# The Briefing: AI for Science

## Metadata
- Channel: Anthropic
- Duration: 107 min
- YouTube: https://www.youtube.com/watch?v=l-7BgzRkq1Y

## Transcript

**[00:00] Speaker A:** Hey hey hey.  
嘿嘿嘿。  
**[01:14] Speaker A:** Please welcome head of go-to-market for healthcare and life sciences at Anthropic, Zubair Jandali.  
请欢迎 Anthropic 医疗健康与生命科学市场拓展负责人 Zubair Jandali。  
**[01:31] Speaker B:** Hello everyone. Good morning and welcome to the briefing. For the leaders in this room and everyone joining us on the live stream, thank you for being here today.  
大家好。早上好,欢迎来到本次发布会。感谢在场的各位领导者以及通过直播加入我们的所有朋友,谢谢你们今天的到来。  
**[01:40] Speaker B:** My name is Zubair Jandali. I lead our commercial work in healthcare and life sciences. And I've been at Anthropic since we shipped the first Claude and I've seen every release since.  
我叫 Zubair Jandali,负责我们在医疗健康和生命科学领域的商业工作。自从我们发布第一代 Claude 以来我就在 Anthropic 工作,见证了之后的每一次版本迭代。  
**[01:50] Speaker B:** There has never been a moment that is more exciting to me than this one right here. And you're about to see why.  
对我来说,从未有哪个时刻比现在更令人兴奋。你们马上就会明白原因。  
**[01:57] Speaker B:** Six months ago on this stage, we made a claim that Claude could help with the work of life sciences R&D.  
六个月前在这个舞台上,我们提出了一个观点:Claude 可以辅助生命科学研发工作。  
**[02:06] Speaker B:** Today, we're going to build on that claim with a new one. The claim is this: Claude can run the work, not help with it, not accelerate it, but run it.  
今天,我们要在此基础上提出一个新观点。这个观点是:Claude 可以运行这些工作,不是辅助,不是加速,而是直接运行。  
**[02:17] Speaker B:** It's a bold thing to say out loud. So, let me tell you why we believe it.  
这是一个大胆的宣言。那么,让我告诉你们我们为什么相信这一点。  
**[02:21] Speaker B:** We've all seen this happen in software development.  
我们都见证了软件开发领域发生的变化。  
**[02:26] Speaker B:** Coding has irreversibly changed.  
编程已经发生了不可逆转的改变。  
**[02:30] Speaker B:** Now software development is a loop: write, run, fix, run again.  
现在软件开发是一个循环:编写、运行、修复、再运行。  
**[02:37] Speaker B:** Once AI could sit inside that loop, the loop could keep turning for longer and longer stretches of time before an engineer had to step back in.  
一旦 AI 能够融入这个循环,这个循环就能持续运转越来越长的时间,然后工程师才需要再次介入。  
**[02:45] Speaker B:** Two years ago, a few minutes. Today, hours. Soon, days.  
两年前是几分钟。今天是几小时。很快就会达到几天。  
**[02:53] Speaker B:** The scientific method is a loop, too. The original loop.  
科学方法也是一个循环。最初的循环。  
**[02:56] Speaker B:** Design the experiment, run it, analyze the data, ask the next question.  
设计实验、运行实验、分析数据、提出下一个问题。  
**[03:04] Speaker B:** The experiment happens at the bench.  
实验在实验台上进行。  
**[03:07] Speaker A:** Analysis happens at a keyboard, and that is where the loop stalls.  
分析工作发生在键盘前,而这正是循环停滞的地方。  
**[03:12] Speaker A:** Every R&D-er in this room knows the shape of it: the experiment that takes three days to run but three or more weeks to analyze.  
在座的每一位研发人员都清楚这个模式:实验运行三天,分析却要三周甚至更久。  
**[03:19] Speaker A:** Those weeks aren't science. They're the toil you endure to get to the science.  
那些周复一周的时间并不是在做科学研究,而是为了抵达科学研究而忍受的苦工。  
**[03:28] Speaker A:** It's our belief that that toil is collapsing, and as it does, your scientists will get back more of what they trained for: time at the question.  
我们相信,这种苦工正在瓦解。随着它的消失,你的科学家们将重新获得他们接受训练时追求的东西:与问题共处的时间。  
**[03:42] Speaker A:** We've designed today's program around this transition.  
我们围绕这一转变设计了今天的议程。  
**[03:46] Speaker A:** First, Dario is going to sit down with the scientist who turned GLP-1 into a medicine about compressing timelines in biology.  
首先,Dario 将与把 GLP-1 转化为药物的科学家坐下来,探讨如何压缩生物学研究的时间线。  
**[03:55] Speaker A:** Then, we're going to introduce you to what we've built and what it looks like in the hands of a scientist.  
然后,我们会向大家介绍我们构建的成果,以及它在科学家手中的实际样子。  
**[04:01] Speaker A:** And finally, we'll wrap with three leading lights from the industry sharing with us how AI has transformed their organizations.  
最后,我们将邀请三位行业领军人物与我们分享 AI 如何变革了他们的组织。  
**[04:11] Speaker A:** Two years ago, our CEO wrote that AI-enabled biology and medicine could compress 50 to 100 years of progress into 5 to 10. At the time, it read as ambition.  
两年前,我们的 CEO 写道,AI 赋能的生物学和医学可以将 50 到 100 年的进展压缩到 5 到 10 年。当时,这读起来像是一种雄心壮志。  
**[04:24] Speaker A:** The morning ahead is our case that it has started.  
今天上午的内容将证明,这一切已经开始了。  
**[04:29] Speaker A:** Now, few people alive have carried the arc of a scientific idea all the way through.  
如今,在世的人中,能够将一个科学想法的整个弧线完整推进的,寥寥无几。  
**[04:38] Speaker A:** Please welcome the scientist behind the GLP-1 medicines, Lasker Award winner and former chief scientific adviser at Novo Nordisk, Lotte Bjerre Knudsen, in conversation with Dario Amodei, our co-founder and CEO.  
请欢迎 GLP-1 药物背后的科学家、Lasker 奖得主、Novo Nordisk 前首席科学顾问 Lotte Bjerre Knudsen,她将与我们的联合创始人兼 CEO Dario Amodei 进行对话。  
**[04:58] Speaker A:** about compressing timelines in biology, moderated by Stat senior writer for medicine, Matt Herper.  
关于压缩生物学时间线,由 Stat 医学资深撰稿人 Matt Herper 主持。  
**[05:26] Speaker B:** So, hello everyone and welcome. I'm Matt Herper. I'm a journalist at Stat. We're an award-winning medical news site. And we're here for a conversation between two people I think we can fairly call revolutionaries about what might be a revolution.  
大家好,欢迎各位。我是 Matt Herper。我是 Stat 的记者。我们是一家屡获殊荣的医学新闻网站。今天我们将进行一场对话,对话双方我认为完全可以称之为革命者,而话题可能关乎一场革命。  
**[05:43] Speaker B:** Lotte Knudsen, and I mean you've heard about her role in creating GLP-1s. She is a visionary who saw the impact these medicines could have not only in diabetes but obesity, a field by the way, and a theme that I think we want to pay attention to, that she joined by accident.  
Lotte Knudsen,你们已经听说过她在开发 GLP-1 药物中的作用。她是一位富有远见的人,看到了这些药物不仅在糖尿病领域、而且在肥胖领域可能产生的影响。顺便说一句,这是一个领域,也是我认为我们要关注的一个主题,就是她是意外加入这个领域的。  
**[06:06] Speaker B:** She set off one of the biggest medical revolutions I've seen in a 25-year career. She has won a lot of important awards including the Lasker Prize which is on her lapel, the Breakthrough Prize, and not least the Stat Biomedical Innovation Award.  
她引发了我在25年职业生涯中见过的最大的医学革命之一。她赢得了许多重要奖项,包括她别在胸前的 Lasker 奖、Breakthrough 奖,以及同样重要的 Stat 生物医学创新奖。  
**[06:25] Speaker B:** Introducing Dario here is a little like introducing Santa at the North Pole, but I want to point out that he did key foundational work in the development of artificial intelligence, but also before that worked in the biophysics of electrophysiology of neural circuits, and that he has said previously that he thinks what we're talking about here is the most important thing that AI can do. So Dario, thank you for  
在这里介绍 Dario 有点像在北极介绍圣诞老人,但我想指出的是,他在人工智能的发展中做了关键的基础性工作,而在那之前他还研究过神经回路电生理学的生物物理学,他之前曾说过,他认为我们今天讨论的内容是 AI 能做的最重要的事情。那么 Dario,感谢你  
**[06:53] Speaker A:** having us and I can't wait to hear what you guys say to each other. I just want to start off Lot, you've been - we've talked a bit about AI. You've seen revolutions in medicine. You also know as well as anyone how the existing system works and what slows it down. Are we on the brink of a big change? Does this change everything?  
感谢你们的到来,我迫不及待想听听你们会聊些什么。我想先从 Lot 开始,你已经——我们之前聊过一些关于 AI 的话题。你见证过医学领域的革命,也比任何人都清楚现有体系是如何运作的,以及是什么在拖慢进展。我们是否正处于重大变革的边缘?这会改变一切吗?  
**[07:13] Speaker B:** Yeah, I - and thank you for having me here. It's exciting times. I really do believe we are at a real inflection point, right, where there's so much data becoming available. Every day there's someone sharing a new idea on how things can be compressed to move forward faster. So I think it's a real inflection point, right? And I've experienced both inflection points as well as hype in my world, right? where the real goal with semaglutide was to actually create a health impact. And we did that first with cardiovascular deaths being reduced, right? But the hype was the weight loss that really got more people to actually use the medicines, but the real goal is the health.  
是的,感谢邀请我来这里。现在是令人振奋的时代。我真的相信我们正处在一个真正的拐点上,有如此多的数据正在变得可用。每天都有人分享关于如何压缩流程以加快推进速度的新想法。所以我认为这是一个真正的拐点,对吧?而且在我的领域里,我既经历过真正的拐点,也经历过炒作,对吧?semaglutide 的真正目标其实是创造健康影响。我们首先做到了减少心血管死亡,对吧?但引发热议的是减重效果,这才真正让更多人开始使用这些药物,但真正的目标是健康。  
**[08:10] Speaker A:** Dario, I mean I'm fascinated by this prediction that - I mean you said the progress of 50 to 100 years over five to 10, but that means you're making the progress of a decade in a year. How is that possible? What kind of evidence can we expect for it over the next six to 12 months? Is this actually - do you still feel this is happening? You laid it out in your essay two years ago.  
Dario,我对这个预测很着迷——我是说你提到过在五到十年内实现五十到一百年的进展,但这意味着你在一年内就要取得十年的进步。这怎么可能?在接下来的六到十二个月里,我们可以期待看到什么样的证据?这真的在发生吗——你现在仍然认为这正在发生吗?你在两年前的文章里就阐述过这一点。  
**[08:39] Speaker C:** Yeah. So, first of all, thank you both  
是的。首先,感谢你们二位  
**[08:40] Speaker A:** For coming.  
感谢你来。  
**[08:42] Speaker B:** We're so thrilled.  
我们非常激动。  
**[08:43] Speaker A:** So, you know, one thing I would distinguish between, and I say this a little bit in the essay, is I do say, like, you know, 10 years from now, I think we may be making progress at a rate of 10 years per year.  
我想区分一下,我在文章里也稍微提到过这一点,就是我确实说过,比如说 10 年后,我认为我们可能会以每年 10 年的速度取得进展。  
**[09:00] Speaker A:** I don't think that today we can make progress at a rate of 10 years per year, for a number of reasons.  
我不认为今天我们就能以每年 10 年的速度取得进展,原因有很多。  
**[09:09] Speaker A:** One, the technology is still getting better. It's on a fast exponential.  
第一,技术还在不断进步。它正处于快速的指数增长曲线上。  
**[09:14] Speaker A:** It's much better than it was before. Every new model released is just better at everything from, you know, computational biology to thinking about proteins and DNA.  
它比以前好多了。每一个发布的新模型在各个方面都更强,从计算生物学到蛋白质和 DNA 的研究。  
**[09:26] Speaker A:** But, you know, we still have some ways to go on the exponential.  
但是,我们在这条指数曲线上还有一段路要走。  
**[09:30] Speaker A:** And the second is, I think, just the inertia in the system—both the inertia of getting used to using these tools and operating in the new way, and figuring out, you know, how do they help with academic biology research, how do they help with new target discovery, how do they help with running clinical trials faster.  
第二个原因,我认为就是系统的惯性——既包括适应使用这些工具、以新方式运作的惯性,也包括搞清楚它们如何帮助学术生物学研究、如何帮助新靶点发现、如何帮助更快地开展临床试验。  
**[09:50] Speaker A:** And I think the longer response to the regulatory system, which is, you know, it's going to take a decade for all this new—  
还有我认为监管系统的响应会更慢,也就是说,所有这些新的——  
**[09:58] Speaker B:** Abilities to be developed with AI, and for the whole system, which is used to operating in a whole bunch of different ways, many parts of which don't yet even believe that AI is going to revolutionize it, to learn to operate.  
用 AI 开发出的能力,以及让整个系统——这个习惯于以各种不同方式运作的系统,其中很多部分甚至还不相信 AI 会给它带来革命性变化——学会如何运作。  
**[10:11] Speaker A:** with AI, but I think once we get it going, particularly in all the parts of the pipeline,  
通过 AI 来实现,但我认为一旦我们让它运转起来,尤其是在整个流程的各个环节,  
**[10:19] Speaker A:** I absolutely believe that we can make 10 years of progress every year.  
我坚信我们每年都能取得相当于十年的进展。  
**[10:24] Speaker A:** And I think the way it works is, you know, I thought about it as I was writing Machines of Loving Grace and thought about my own history, you know, at least trying to do research in biology,  
我认为它的运作方式是这样的,你知道,我在写 Machines of Loving Grace 的时候思考过这个问题,也回想了我自己的经历,至少是我尝试做生物学研究的那段经历,  
**[10:38] Speaker A:** that there are a small number of really essential discoveries. You know, GLP-1 is one, right? But think of CRISPR, think of advances in microscopy for doing connectomics and for understanding systems neuroscience.  
其实真正关键的发现数量是很少的。你知道,GLP-1 就是其中之一,对吧?但想想 CRISPR,想想显微镜技术在连接组学方面的进步,以及在理解系统神经科学方面的作用。  
**[10:54] Speaker A:** Like, there are a small number of these discoveries, and some of them feel like they could have happened decades earlier than they did, right?  
就是说,这类发现的数量很少,而其中有些感觉本可以比实际发生的时间早几十年就出现,对吧?  
**[11:01] Speaker A:** Like, you know, with CRISPR, it's like, oh yeah, there was someone studying genetic engineering, and then they happened to go to a seminar on the bacterial innate immune system, and no one would have thought that there was a connection.  
比如说 CRISPR,当时有人在研究基因工程,然后恰好去听了一个关于细菌先天免疫系统的讲座,没人会想到这两者之间会有联系。  
**[11:12] Speaker A:** But as the AI models get smarter and smarter, I have to think that they're going to be better at discovering these things early, and they're going to discover 10 times or 100 times as many of them.  
但随着 AI 模型变得越来越聪明,我不得不认为它们会更善于提早发现这些联系,而且会发现十倍甚至一百倍数量的这类发现。  
**[11:24] Speaker A:** And then it's all going to be about taking those new discoveries, which have such broad implications across everything, and kind of using them and translating them, both in terms of the science operationally and in terms of the regulatory  
然后关键就在于如何利用这些新发现——它们对各个领域都有广泛的影响——并将它们转化应用,无论是在科学操作层面,还是在监管层面  
**[11:38] Speaker A:** system to make everything truly go faster and it's going to be, you know, it's going to be the work of years but I think it can be done. We know the difficulties in this system. I mean the Tufts Center for the Study of Drug Development estimates it's $2.6 billion to bring a new molecule to market including cost of failure and capital. I think that's actually even low. It's 10 years to bring a drug to market. How fast do you think the system can start to absorb AI and where will that happen?  
系统能让一切真正加速,这将会是,你知道的,这将是多年的工作,但我认为这是可以做到的。我们知道这个系统中的困难。我的意思是,Tufts Center for the Study of Drug Development估计,将一个新分子推向市场需要26亿美元,这还包括失败成本和资本。我认为这个数字实际上还是偏低的。把一款药物推向市场需要10年时间。你认为这个系统能以多快的速度开始吸收AI,会在哪些地方发生?  
**[12:11] Speaker B:** Well, it's already happening, right? There are already areas of the whole drug discovery and development process that have already been revolutionized. And then as the models, as Dario's saying,  
嗯,这已经在发生了,对吧?整个药物发现和开发流程中已经有一些领域被彻底革新了。然后随着模型的发展,就像Dario说的那样,  
**[12:21] Speaker A:** It's just getting so much bigger now. You can break it all down to little areas that you need to work with, but you can compress all of them, right?  
现在规模变得越来越大了。你可以把它拆解成一个个需要处理的小领域,但你也可以把所有这些都压缩,对吧?  
**[12:30] Speaker A:** So I think it's hard to put a number on, right?  
所以我觉得很难给出一个具体数字,对吧?  
**[12:32] Speaker A:** But some of the work that I've experienced, right? So four years to make the right molecule.  
但就我经历过的一些工作来说,制造出正确的分子需要四年时间。  
**[12:39] Speaker A:** We still need to make molecules and prove that they are the right ones, but it could probably go to one year.  
我们仍然需要制造分子并证明它们是正确的,但这个过程可能会缩短到一年。  
**[12:46] Speaker A:** And another area is recruiting patients for clinical trials.  
另一个领域是为临床试验招募患者。  
**[12:49] Speaker A:** You know, some of the stuff we had to do sometimes recruit 20,000 people for a five-year study and it takes two years to recruit the people, yet people have never been more connected than they are now.  
你知道,我们有时需要为一项五年期研究招募两万人,光招募就要花两年时间,但现在人们的联系从未像今天这样紧密。  
**[13:00] Speaker A:** So it must be possible to do that faster.  
所以一定有办法加快这个速度。  
**[13:02] Speaker A:** I also think the regulatory...  
我还认为监管方面……  
**[13:04] Speaker A:** Process could really be revolutionized by AI. So there's just—I could go on for a long time thinking about it. I think so many areas could really become completely—uh, not completely compressed, but really markedly compressed.  
这个过程真的可以被 AI 彻底革新。我可以就这个话题讲很久。我认为很多领域确实可以被大幅压缩——呃,不是完全压缩,而是显著压缩。  
**[13:18] Speaker A:** And then there are areas that will be more difficult, like—uh, I'm very passionate about new target discovery. You know, how do we find the next GLP-1 or the next new biology? And that might seem like the most difficult project right now.  
然后也有一些领域会更困难,比如——呃,我对新靶点发现非常热衷。你知道,我们怎么找到下一个 GLP-1 或下一个新的生物学机制?这可能看起来是目前最困难的项目。  
**[13:33] Speaker A:** But then, uh, every day there's something new being shared—oh, now we can do this faster, this faster—and we can make sense of all of this. And then add on top of all those wonderful agents you can have doing stuff for you while you're sleeping, right?  
但每天都有新东西出现——哦,现在我们可以更快地做这个、更快地做那个——我们能理解所有这些。然后再加上那些很棒的 agent,它们可以在你睡觉的时候帮你干活,对吧?  
**[13:46] Speaker A:** So I think we really will see a large compression, but we still need the clinical trial data. So it's probably hard to imagine that you can go below five years.  
所以我认为我们真的会看到大幅压缩,但我们仍然需要临床试验数据。所以很难想象能压缩到五年以下。  
**[13:59] Speaker B:** I mean, that's a—there's a fundamental—that presents another fundamental problem for Anthropic and for AI compared to other—one of the big use cases obviously for AI changing something has been code. But you write code and put it in to run it, and you find out if it runs.  
我的意思是,这是一个——有一个根本性的——这给 Anthropic 和 AI 带来了另一个根本性问题,相比其他领域。显然,AI 改变某个领域的一个重要用例是代码。但你写代码,运行它,就能立刻知道它能不能跑。  
**[14:17] Speaker B:** How does having an industry with this cycle time, where very often you find out your drug didn't work 10 years later when you were convinced the whole time it would, right—I've watched people go through this process for 25 years, it's an immensely difficult thing—how do you deal with that? How much does the cycle time affect what—  
在这样一个行业里,周期时间这么长,你经常是在十年后才发现药物没有效果,而你一直坚信它会有效,对吧——我看着人们经历这个过程已经 25 年了,这是极其困难的事情——你怎么应对?这个周期时间对什么的影响有多大——  
**[14:39] Speaker A:** What AI can do here?  
AI在这方面能做什么呢?  
**[14:40] Speaker B:** Yeah, I think that's the thing that's going to control the pace. That's going to be the limiting factor.  
是的,我认为这就是控制进度的因素。这将是限制性因素。  
**[14:45] Speaker B:** I think there are a lot of things we can do to speed it up or to work around it, but I think it's always going to be the limiting factor.  
我认为我们可以做很多事情来加快速度或绕过它,但我觉得它始终会是限制性因素。  
**[14:51] Speaker B:** Right? That's why I didn't say, you know, we'll make a thousand years of progress in 10 years, and why I'm very specifically not saying we're going to get all this crazy stuff by 2028 or something.  
对吧?这就是为什么我没有说我们会在10年内取得一千年的进步,也是为什么我特别不会说我们到2028年就能实现所有这些疯狂的东西。  
**[15:03] Speaker B:** We can't get something out the other end of the pipe. There's just absolutely no way to do it.  
我们没法从管道的另一端得到东西。根本就没有办法做到。  
**[15:10] Speaker B:** So there's this inertia, but I think there's lots of things we can do to work around it or to shorten the cycle time.  
所以存在这种惯性,但我认为我们可以做很多事情来绕过它或缩短周期时间。  
**[15:17] Speaker B:** First of all, I think on the fundamental scientific discovery side, I mean, you know, just imagine you're trying to develop the next CRISPR or the next GLP-1 or the next, you know, Ed Boyden's crazy expansion microscopy thing—the cycle time is pretty fast, right?  
首先,我认为在基础科学发现方面,你想象一下你正在开发下一个CRISPR或下一个GLP-1,或者Ed Boyden那个疯狂的扩展显微镜技术之类的东西——周期时间是相当快的,对吧?  
**[15:30] Speaker B:** It may not compile in seconds like code, but you can make your cycle time hours depending on the type of experiment you're doing.  
它可能不会像代码那样在几秒钟内编译完成,但根据你做的实验类型,你可以把周期时间控制在几小时内。  
**[15:40] Speaker B:** So if you can speed that up with a relatively short cycle time, then you have all these additional tools.  
所以如果你能用相对较短的周期时间加快速度,那么你就拥有了所有这些额外的工具。  
**[15:46] Speaker B:** And what those tools allow you to do is lower the cycle time on everything else, right?  
而这些工具能让你做的就是降低其他所有事情的周期时间,对吧?  
**[15:52] Speaker B:** Where, you know, you've done a better job optimizing the drug. You have better measurement ability. You've found a spate of new...  
在这种情况下,你在优化药物方面做得更好了。你有了更好的测量能力。你发现了一大批新的...  
**[16:01] Speaker A:** targets. So it allows you to both shorten the cycle time because you can—you've turned it more from an art into a science, right?  
靶点。所以它能让你既缩短周期时间,因为你可以——你已经把它更多地从一门艺术变成了一门科学,对吧?  
**[16:10] Speaker A:** We've been gradually doing that much more slowly than we'd like over the previous decades, but we can accelerate that process.  
在过去几十年里,我们一直在逐步推进这个转变,但速度远比我们期望的慢得多,不过我们可以加速这个过程。  
**[16:19] Speaker A:** And note that there are some parts of the slow and expensive process that will speed up once we have things that work better.  
还要注意的是,一旦我们有了更有效的方法,这个缓慢而昂贵的过程中的某些环节就会加快。  
**[16:27] Speaker A:** Right? If we have to do less clinical trials because things work more of the time, if when things work they have a stronger effect, then we need to recruit less patients.  
对吧?如果因为药物更常有效而需要做的临床试验更少,如果当药物有效时它们的效果更强,那么我们就需要招募更少的患者。  
**[16:35] Speaker A:** So this five-year thing that you talked about will be shorter. So every part of this long cycle time we can chop.  
所以你提到的这个五年周期会变短。因此这个漫长周期的每个环节我们都可以砍掉一些。  
**[16:41] Speaker A:** Now maybe we can only chop it from, you know, five to ten years to five years or three years or whatever. It's still going to be the controlling thing, and so we're going to have to do a lot of stuff in parallel.  
也许我们只能把它从五到十年缩短到五年或三年之类的。它仍然会是制约因素,所以我们必须并行做很多事情。  
**[16:51] Speaker A:** That's the only way. The biggest issue for drug discovery—the cycle time will help a lot, but the biggest issue is the failure rate, right?  
这是唯一的办法。药物发现的最大问题——周期时间会有很大帮助,但最大的问题是失败率,对吧?  
**[17:00] Speaker A:** And that's your 'Machines of Loving Grace' piece.  
这就是你那篇「Machines of Loving Grace」文章讨论的内容。  
**[17:04] Speaker A:** And I've seen technologists talking with Andy Grove about all the ways drug discovery could change in the times since then.  
我看到过技术专家与 Andy Grove 讨论从那时起药物发现可能改变的所有方式。  
**[17:11] Speaker A:** You know, the technology industry has Moore's Law where things get cheaper and cheaper and faster and faster, and the drug industry has Eroom's Law, which is Moore's Law backward.  
你知道,科技行业有 Moore's Law,事物变得越来越便宜、越来越快,而制药行业有 Eroom's Law,这是 Moore's Law 的反向——越来越贵、越来越慢。  
**[17:19] Speaker A:** So do you think that—why is AI different from all these other technologies we've made?  
那么你认为——为什么 AI 与我们开发的所有其他技术都不同?  
**[17:28] Speaker A:** Genomes go from three billion to $300, and we still spend more on drug development. Why is this different?  
基因组测序的成本从三十亿美元降到了三百美元,但我们在药物研发上的投入反而还在增加。为什么会这样呢?  
**[17:36] Speaker B:** Yeah, so I mean I remember when Andy Grove was talking about these things, there was a writer, I forget who it was, who coined like the Andy Grove fallacy of like if you try and think... Derek Lowe.  
是这样,我记得当 Andy Grove 谈论这些问题的时候,有一位作家,我忘了是谁了,提出了所谓的「Andy Grove 谬论」,就是说如果你试图去想……是 Derek Lowe。  
**[17:45] Speaker A:** Derek Lowe. Okay. If you just think of biology as kind of this engineering system, like we just need to make things more rational, make things make more sense, it just kind of doesn't work, right?  
Derek Lowe,好的。如果你只是把生物学看作某种工程系统,觉得我们只需要让事情变得更理性、更合理,这种想法其实是行不通的,对吧?  
**[17:59] Speaker B:** Because this isn't a designed system. It's this like super messy evolved system. So I basically agree with that. I think that's right.  
因为生物系统不是被设计出来的系统,它是一个极其混乱的演化系统。所以我基本同意这个观点,我觉得说得对。  
**[18:10] Speaker B:** But the way we as humans have only made progress is we've used our human brains, which are capable of comprehending complexity, to kind of wrestle with that, to make sense of it, to make progress against it.  
但人类取得进步的唯一方式,就是利用我们的大脑——它具备理解复杂性的能力——去与复杂性搏斗,去理解它,去战胜它从而取得进展。  
**[18:22] Speaker B:** And I think what I'm saying is AI here is not going to be another engineering technology that organizes our data better or tries to unblock one part of the process when there's a world of complexity.  
我想说的是,AI 在这里不会只是另一种工程技术,只是帮我们更好地组织数据,或者在面对无比复杂的系统时只是疏通流程中的某个环节。  
**[18:35] Speaker B:** It's going to be a general purpose technology that helps us to make sense of that complexity in its full complexity better.  
它将会是一种通用技术,帮助我们更好地理解那些复杂性——是在其完整的复杂性层面上去理解。  
**[18:44] Speaker B:** We don't know for sure if that's going to work out, but I think we're seeing signs that... we're seeing the beginnings of it. It's already starting to. That is the hope.  
我们还不能确定这是否一定会成功,但我觉得我们正在看到一些迹象……我们正在看到它的开端。它已经开始显现了。这就是我们的希望所在。  
**[18:52] Speaker A:** That is what we should shoot for. I am optimistic. I can't be confident because we don't know the future, but I'm optimistic.  
这就是我们应该追求的目标。我持乐观态度。我不能说有十足把握,因为我们无法预知未来,但我是乐观的。  
**[18:58] Speaker B:** I just wonder if you have a thought there, L.  
我想知道你对此有什么想法,L。  
**[19:00] Speaker C:** Yeah. So I think we'll get better at improving the probability of success of new medicines because we'll understand them better. We have a better foundation for why we picked those targets. We can understand the MOA. We can make smarter clinical trials. So I think we will see fewer failures.  
是的。我认为我们会在提高新药成功概率方面做得更好,因为我们会更深入地理解它们。我们对为什么选择这些靶点有了更好的基础。我们能理解作用机制(MOA)。我们能设计更明智的临床试验。所以我认为我们会看到更少的失败案例。  
**[19:19] Speaker C:** But of course we're going to have to listen to Derek Lowe probably for another 10 years. And he's very insightful, right?  
但当然,我们可能还得再听 Derek Lowe 说上10年。他确实很有洞察力,对吧?  
**[19:25] Speaker C:** So, but there will be criticism until some examples have moved forward and people are going to say, oh this is not fully AI designed, but that's again, that's not the point.  
所以,在一些实例取得进展之前,批评声会一直存在,人们会说,哦,这不是完全由 AI 设计的,但这同样不是重点。  
**[19:36] Speaker C:** The point is that there's so many things you need to know in order to choose the right medicines and develop the right medicines, and all the areas of it can be improved upon.  
重点是,为了选择正确的药物并开发正确的药物,你需要了解很多东西,而其中所有领域都可以得到改进。  
**[19:49] Speaker C:** So you will improve your probability of success for whether a new medicine actually comes out successful.  
因此你会提高新药最终成功上市的概率。  
**[19:57] Speaker C:** But again, also many learnings, right? And a very important learning from GLP-1 is that actually a pleiotropic effect is a good one, right? What's really so fantastic about GLP-1 is that you have all these benefits on multiple different organs, yet the whole field of drug discovery is still looking for the genetics with the highest window or the mouse model with the highest window. They should look at also, you know...  
但同样,也有很多经验教训,对吧?从 GLP-1 得到的一个非常重要的教训是,多效性作用实际上是好事,对吧?GLP-1 真正神奇的地方在于它对多个不同器官都有益处,然而整个药物发现领域仍在寻找治疗窗口最大的遗传学证据或小鼠模型。他们也应该关注,你知道的...  
**[20:25] Speaker A:** Finding these broad signals and we can help.  
找到这些广泛的信号，而我们可以提供帮助。  
**[20:29] Speaker B:** So you've also said in addition to this being the area that you have the most hope, that it's one of the ones you worry about risk most.  
所以你也说过，除了这是你最有希望的领域之外，它也是你最担心风险的领域之一。  
**[20:34] Speaker B:** I mean, the obvious worry is could all these cool technologies help people make bioweapons or synthetic organisms that run wild, or every science fiction novel we've read that actually could happen.  
我的意思是，显而易见的担忧是，这些很酷的技术会不会帮助人们制造生物武器，或者制造失控的合成生物体，或者我们读过的每一部科幻小说中的情节真的可能发生。  
**[20:45] Speaker B:** How do you guard against that, Daria?  
Daria，你们如何防范这种情况？  
**[20:47] Speaker A:** Yeah, I mean, you know, I think Anthropic as a company has thought about the risks of AI and different fields quite a lot.  
是的，我的意思是，我认为 Anthropic 作为一家公司已经对 AI 的风险以及不同领域的问题思考了很多。  
**[20:57] Speaker A:** Um, you know, we're currently confronting one of those sets of issues with kind of the cyber risks of AI, right?  
嗯，你知道，我们目前正在面对其中一类问题，就是 AI 的网络安全风险，对吧？  
**[21:03] Speaker A:** We've kind of entered an inflection point or a critical window, you know, as we go along the exponential. Different things in terms of both benefits and economically useful applications and, you know, potentially concerning applications turn on at different times.  
我们已经进入了一个拐点或关键窗口期，你知道，随着我们沿着指数曲线前进，无论是从收益和经济上有用的应用，还是潜在令人担忧的应用，不同的东西会在不同的时间点开启。  
**[21:18] Speaker A:** And so we've had the benefit of seeing the window kind of turn on for cyber.  
所以我们很幸运地看到了网络安全这个窗口期的开启。  
**[21:22] Speaker A:** It is not yet turned on for biology.  
但生物学领域的窗口期还没有开启。  
**[21:25] Speaker A:** Uh, now I think those two examples are very different, right? With cyber, it's like you have the ability to find exploits, and then it's the same model can also patch the exploits, and you can find the exploits in a few seconds, you can patch the exploits in a few seconds.  
呃，现在我认为这两个例子非常不同，对吧？对于网络安全来说，你有能力找到漏洞，然后同一个模型也可以修补这些漏洞，你可以在几秒钟内找到漏洞，也可以在几秒钟内修补漏洞。  
**[21:40] Speaker A:** Biology—  
生物学——  
**[21:42] Speaker A:** As we all know, is rather different from that. As we learned during COVID-19, there's not necessarily the same symmetry, but there are some lessons we can learn.  
众所周知,这与那个相当不同。正如我们在 COVID-19 期间了解到的,不一定存在同样的对称性,但我们可以从中学到一些经验教训。  
**[21:52] Speaker A:** For example, the safeguards that we put on the models to make sure that they can output beneficial content but can't output worrying content, and the difficulty of distinguishing between the two.  
例如,我们在模型上设置的安全防护措施,以确保它们能够输出有益的内容,但不能输出令人担忧的内容,以及区分两者的难度。  
**[22:07] Speaker A:** So we've spent a lot of time putting effort into, you know, what is a helpful query, what is a dangerous query. You see that in cyber where it's like, okay, the model can find bugs. Is that helpful or is it dangerous? It can be sum of both.  
所以我们花了很多时间努力研究什么是有用的查询,什么是危险的查询。你在网络安全领域就能看到这一点,比如模型可以找到漏洞。这是有益的还是危险的?它可能两者兼有。  
**[22:21] Speaker B:** How do you balance putting safeguards on the model versus putting the model in the hands of people who can kind of be the white hat hacker and figure out what you can do with it that's harmful?  
你如何平衡在模型上设置安全防护,与将模型交给那些可以充当白帽黑客、找出可以用它做哪些有害事情的人之间的关系?  
**[22:35] Speaker A:** No, no, exactly. And you know, I think this is another thing, not just in cyber, not just in biology, that's going to be the work of years, where you have a model that depending on what you do and what you say to it and how you interact with it, can do lots of wondrous things, can create enormous economic value, but then there's a tiny slice of things that are actually dangerous, and within that a tinier slice of things that are dangerous and you couldn't do without AI, or where AI actually is the limiting factor, right? Because we should think about that. And so having accurate threat models—  
没错,没错,确实如此。你知道,我认为这是另一个需要多年努力的问题,不仅仅在网络安全领域,也不仅仅在生物学领域。你有一个模型,根据你对它做什么、对它说什么以及如何与它互动,它可以做很多了不起的事情,可以创造巨大的经济价值,但同时也有一小部分事情实际上是危险的,而在这其中又有更小的一部分事情,是没有 AI 你做不到的,或者说 AI 实际上是限制因素,对吧?因为我们应该考虑到这一点。所以要有准确的威胁模型——  
**[23:10] Speaker A:** You both need to have these safeguards and you need to have trusted access programs, right?  
你们都需要有这些防护措施，也需要有可信访问机制，对吧？  
**[23:14] Speaker A:** So, you know, within pharmaceutical companies, people handle dangerous biological material all the time, right?  
你知道，在制药公司内部，人们一直在处理危险的生物材料，对吧？  
**[23:20] Speaker A:** And they have their own protocols for it.  
他们有自己的一套处理规程。  
**[23:21] Speaker A:** So, can we piggyback on those protocols where we say, "Okay, the elements of society that already handle these potentially dangerous things, can we just say, okay, you guys already know how to do this.  
那么我们能不能借鉴这些规程，比如说「好的，社会中那些已经在处理这些潜在危险物品的部门，我们能不能就直接说，好，你们已经知道怎么做这件事了。  
**[23:35] Speaker A:** You're cleared for handling this, but you know, maybe we don't want to give it to just anyone or just anyone without verifying something.  
你们有处理这个的资质，但我们可能不想把它交给随便什么人，或者说不经过某种验证就交给任何人。」  
**[23:46] Speaker A:** So this idea of kind of trusted access, existing elements of society that manage these risks, safeguards on the general models, it's kind of almost a multi-layer cake of how to make sure we get as close as we can to 100% of the benefits while blocking as close as we can to 100% of the real counterfactual actual dangerous harms, which is a very narrow slice, but that we need to make sure that we put a proper buffer around to manage appropriately.  
所以这种可信访问的理念，借助社会中已经在管理这些风险的现有机构，在通用模型上设置防护措施，这就像是一个多层蛋糕，目的是确保我们尽可能接近100%地获得收益，同时尽可能接近100%地阻止那些真实的、反事实的、实际的危险伤害——这是非常窄的一个范围，但我们需要确保在周围设置适当的缓冲区来妥善管理。  
**[24:16] Speaker B:** I mean, you've dealt with the problem of developing a drug in tens of thousands of people and giving it to millions and with the worries of side effects that come, and you've both seen cases outside of your work where there were drugs that had real side effects that required they be withdrawn, and also some of the worries about GLP-1s which have mostly not panned out.  
我是说，你处理过在数万人身上开发药物、然后给数百万人使用的问题，以及随之而来的副作用担忧，而且你们都见过工作之外的案例，有些药物确实出现了真实的副作用，需要被撤回，还有关于GLP-1的一些担忧，这些担忧大多没有成为现实。  
**[24:36] Speaker B:** How do you  
你如何  
**[24:38] Speaker A:** Think about these risk problems?  
如何看待这些风险问题?  
**[24:40] Speaker B:** Yeah. So I think just so in the pharmaceutical industry, many people already know that we're very much used to balancing risk versus benefit, and I think it's just having a strong focus on it like Anthropic absolutely has, and it's even I guess put in your bylaws that how you are thinking about the benefit versus the risk. And I think just being focused on that, having a really strong view on how to evaluate that, and also having maybe an independent—so we have independent data monitoring committees in large clinical trials, maybe that's also something that might already be in place, but really having someone who is not motivated by money to actually look at that this is being dealt with properly.  
是的。我想说的是,在制药行业,很多人都很清楚我们非常习惯于平衡风险与收益,我认为关键在于对此保持强烈的关注,就像 Anthropic 绝对做到的那样,这甚至被写入了你们的章程,即如何权衡收益与风险。我认为就是要专注于此,对如何评估这一点有非常明确的立场,还要有独立的——我们在大型临床试验中有独立的数据监测委员会,也许这也是可能已经存在的机制,但真正重要的是要有不受金钱驱使的人来确保这件事得到妥善处理。  
**[25:33] Speaker A:** Who would that person be?  
那会是什么样的人呢?  
**[25:35] Speaker B:** So you know, I think within Anthropic we have an organization called the Long-Term Benefit Trust. It governs actually the entire company. It appoints a majority of the board seats. And, you know, we have the head of the Clinton Health Access Initiative. We have a former Supreme Court justice on California's Supreme Court. And we're always looking to add others who have different, you know, different experience. I think over time, you know, it's certainly going to include folks who have, you know, experience in biology and medicine. And the one rule for people on the LTBT is they don't have any stock in the...  
你知道,我认为在 Anthropic 内部我们有一个叫做 Long-Term Benefit Trust 的组织。它实际上管理着整个公司,任命董事会的多数席位。而且我们有 Clinton Health Access Initiative 的负责人,我们还有一位加州最高法院的前大法官。我们一直在寻找具有不同经验的其他人加入。我认为随着时间推移,肯定会包括在生物学和医学领域有经验的人士。而 LTBT 成员的一个规则就是他们不持有公司的任何股份...  
**[26:12] Speaker A:** Company, and so I think exactly this kind of independent governance makes sense.  
公司层面,所以我认为这种独立治理模式确实是合理的。  
**[26:18] Speaker A:** Now that's for the company as a unit. That's for the corporate structure overall.  
这是针对作为一个整体的公司而言的,是针对整体企业架构的。  
**[26:21] Speaker A:** It may also be the case that for specific applications like bio and cyber, it may make sense to have independent monitoring committees either associated with specific companies or kind of across the industry.  
也有可能的是,针对特定应用领域,比如生物和网络安全,设立独立的监督委员会是有意义的,可以是与特定公司关联的,也可以是跨行业的。  
**[26:34] Speaker A:** And you know, as we're seeing, the government probably also has a role to play.  
而且正如我们所看到的,政府可能也有其应发挥的作用。  
**[26:37] Speaker B:** So, Dario, I asked Claude what I should ask you.  
那么 Dario,我问了 Claude 我应该问你什么。  
**[26:42] Speaker B:** And this was the first question, and I like it because it's tougher than mine.  
这是第一个问题,我很喜欢,因为它比我自己想的问题更尖锐。  
**[26:47] Speaker B:** Why should pharma trust AI predictions when your models hallucinate? Where's the validation data for claims that AI accelerates timelines?  
既然你们的模型会产生幻觉,制药公司为什么要相信 AI 的预测?有什么验证数据能支持 AI 加速研发时间线的说法?  
**[26:56] Speaker B:** I thought it would actually—  
我觉得这个问题确实——  
**[26:57] Speaker A:** You know, Claude has been among my toughest interviewers over the months and the years.  
你知道,这些月以来这些年以来,Claude 一直是我遇到的最严格的面试官之一。  
**[27:03] Speaker A:** So I would say on hallucinations, actually hallucinations have gotten better and better over time.  
所以关于幻觉问题,我想说的是,实际上幻觉现象随着时间推移已经越来越少了。  
**[27:10] Speaker A:** You don't hear as much about hallucinations as you used to. They still happen, but you know, the situation with hallucinations is a little bit—to make another analogy—like the situation with cars that drive themselves.  
你不会像以前那样频繁听到幻觉的问题了。它们仍然会发生,但是,幻觉的情况有点——打个比方——就像自动驾驶汽车的情况。  
**[27:25] Speaker A:** I don't think we will ever have an AI model that never hallucinates. I think just the probabilistic way in which these models reason, which I suspect is the same as the probabilistic way in which humans—  
我不认为我们会有一个永远不产生幻觉的 AI 模型。我觉得这些模型进行推理的概率性方式,我怀疑与人类进行推理的概率性方式是一样的——  
**[27:36] Speaker A:** The reason is it is prone to a duality between creativity and, you know, basically hallucination, right?  
原因在于它容易陷入创造力和幻觉之间的二元性,对吧?  
**[27:48] Speaker A:** In order to be creative, you're often straddling the boundary between making things up and having good ideas.  
为了具有创造力,你往往游走在编造内容和产生好想法之间的边界上。  
**[27:54] Speaker A:** And so I think it's never going to fully go away. The models will just get better at distinguishing between the two, at having better filters.  
所以我认为这个问题永远不会完全消失。模型只会在区分两者方面变得更好,拥有更好的过滤机制。  
**[28:05] Speaker A:** Again, similar to self-driving cars, we're never going to have a human that can perfectly drive a car. We're never going to have a perfect self-driving machine, perfect self-driving algorithm, right?  
再说一次,就像自动驾驶汽车一样,我们永远不会有能够完美驾驶汽车的人类。我们也永远不会有完美的自动驾驶机器、完美的自动驾驶算法,对吧?  
**[28:20] Speaker A:** It doesn't know how to deal with a duck jumping across the road. It doesn't know how to deal with, you know, there's like 5 inches of ice on the road when previously there's only been 3 inches.  
它不知道如何应对鸭子突然跳过马路的情况。它也不知道如何应对路面上有5英寸厚的冰层,而之前只有3英寸的情况。  
**[28:29] Speaker A:** So I think hallucinations are going to go down and down, and we're just going to have to get used to the idea that, as with humans, oh yeah, this here is a brilliant person.  
所以我认为幻觉会越来越少,而我们只需要习惯这样一个想法:就像对待人类一样,「噢,对,这是一个聪明的人」。  
**[28:41] Speaker A:** They have lots of ideas. Many of them are wrong. Sometimes they'll have misconceptions.  
他们有很多想法。其中许多是错的。有时他们会有误解。  
**[28:47] Speaker A:** Sometimes, you know, like humans, I imagine the AI models may get dogmatically attached.  
有时候,就像人类一样,我想AI模型可能会变得教条式地固执。  
**[28:51] Speaker A:** What if your AI model is a Nobel Prize winner, but it's Kary Mullis and it's talking to a glowing green raccoon?  
如果你的AI模型是诺贝尔奖得主,但它是Kary Mullis,而且它在和一只发光的绿色浣熊说话呢?  
**[28:58] Speaker B:** I mean, you know, there's the list of Nobel Prize winners with that...  
我是说,你知道,有一份具有这种特点的诺贝尔奖得主名单...  
**[29:03] Speaker A:** The line is long, and it just says give everybody vitamin C.  
这个名单还挺长的,而且它就只会说给每个人都补充维生素C。  
**[29:06] Speaker A:** Very long, right? It's a very long list. I think that's maybe another illustration of this balance between creativity and hallucination, right? Like Kary Mullis.  
非常长,对吧?这是一个很长的列表。我觉得这可能是另一个例子,说明了创造力和幻觉之间的平衡,对吧?就像 Kary Mullis 那样。  
**[29:16] Speaker A:** It's like, you know, he came up with brilliant, brilliant ideas, and yet his hallucination rate is pretty high, isn't it? It was really quite high and there were some chemicals involved, I believe.  
就像是,你知道,他提出了非常出色的想法,但他的幻觉率也相当高,不是吗?确实挺高的,而且我相信还涉及到一些化学物质。  
**[29:28] Speaker B:** Yeah, yeah.  
是啊,是啊。  
**[29:30] Speaker B:** What question do you hear from skeptics in the industry? You're getting to talk to everybody.  
你从业内的怀疑者那里听到什么问题?你能接触到所有人。  
**[29:34] Speaker A:** That we haven't asked him yet.  
我们还没问过他的问题。  
**[29:36] Speaker B:** Yeah. You know, the question I most often get when I'm out speaking is like, 'Yeah, but AI doesn't work for my field. I guess lucky for you, but I'm not sure it's true.' Right? So for my field, I would say, you know, I'm a scientist, right? I would say AI is not going to replace scientists. I'm not worried about that at all. But scientists who are not using AI are going to get replaced, and you can expand that to every possible field and every organization. So people should be skeptical, absolutely, but they should lean in because that is how we can create more solutions for the work.  
是的。你知道,我在外面演讲时最常被问到的问题就是:「是啊,但 AI 对我的领域不管用。我猜你挺幸运的,但我不确定这是真的。」对吧?所以对于我的领域,我会说,你知道,我是个科学家,对吧?我会说 AI 不会取代科学家。我完全不担心这个。但是不使用 AI 的科学家会被取代,而且你可以把这个推广到每一个可能的领域和每一个组织。所以人们应该保持怀疑,这完全没问题,但他们应该积极参与进来,因为这才是我们为工作创造更多解决方案的方式。  
**[30:14] Speaker A:** You told me a story of a researcher you saw and you were impressed by how many agents he had arguing about the work he was doing. And my question for a lot of researchers is, look, as a journalist, I can use AI to speed up my work. I can also spend my entire day arguing with certain LLMs, especially...  
你跟我讲过一个故事,关于你见到的一位研究员,你对他让那么多 agent 争论他正在做的工作印象深刻。而我想问很多研究人员的问题是,你看,作为记者,我可以用 AI 来加速我的工作。我也可以花一整天时间跟某些 LLM 争论,尤其是……  
**[30:32] Speaker A:** There are particular LLMs for which that's true. How do you make sure you're using this productively as a researcher?  
确实有一些特定的 LLM 是这样的。作为研究人员,你如何确保有效地使用这些工具呢?  
**[30:42] Speaker B:** Yeah, well, I guess you've got to have a really strong focus on what it actually is that you're trying to do, and then you have to be informed about what the different models are and the ways of using them. And to stay in Anthropic language,  
是的,我想你必须非常清楚自己真正要做的是什么,然后你需要了解不同的模型以及使用它们的方法。用 Anthropic 的术语来说,  
**[30:57] Speaker B:** You know, many people understand what a chat ability is. They've gotten that. Even my mother, who's 92, understands that. But then most people, some people, stop there.  
你知道,很多人都理解什么是聊天能力,他们明白这一点。就连我92岁的母亲都懂这个。但是大多数人,有些人就止步于此了。  
**[31:05] Speaker B:** They don't understand the difference between a chat and a code and Claude design, and why should they have agents, and how should they have their brain bank connected to the agents so the agents keep learning, right? So I think that would be my take. You know, more people just need to understand that in order to get the benefits. But they should be mindful of what the goal was. As I was saying earlier, you know, for GLP, the goal is the health, and not get distracted by the hype of the weight loss, but keep being focused on the multiple health benefits.  
他们不理解聊天、代码和 Claude 设计之间的区别,也不明白为什么应该使用 agents,以及如何将他们的知识库连接到 agents 使其不断学习,对吧?所以我认为这就是我的看法。人们需要理解这些才能获得好处。但他们应该记住最初的目标是什么。就像我之前说的,对于 GLP 来说,目标是健康,而不是被减肥的热度分散注意力,而是要持续关注多重健康益处。  
**[31:42] Speaker A:** So we're running low on time now. We're expecting to see a lot of change very fast, but we're also, as we mentioned, in the pharmaceutical industry where things can happen very slowly.  
我们时间不多了。我们预期会看到很多快速的变化,但正如我们提到的,我们身处制药行业,这个行业的变化可能非常缓慢。  
**[31:54] Speaker A:** I want to ask each of you for two things. What is it you'd most like to see for AI and life sciences over the next year as a positive? And what is it you're hoping you don't see?  
我想问你们每个人两个问题。在未来一年里,你最希望在 AI 和生命科学领域看到什么积极的进展?以及你最不希望看到什么?  
**[32:07] Speaker A:** I wanted to start with Daria.  
我想先从 Daria 开始。  
**[32:09] Speaker B:** Yeah. So, um, I think what I'd like to most see, I think Latte mentioned it, is some success of AI with discovering new targets because I think that is the bottleneck.  
是的。我最想看到的,我想 Latte 也提到过,就是 AI 在发现新靶点方面取得一些成功,因为我认为这才是瓶颈所在。  
**[32:24] Speaker B:** I think the models are just knocking on the door of where they can help a lot with that, and it's on a fast exponential.  
我认为这些模型正处于能够在这方面提供巨大帮助的门槛上,而且正以指数级速度快速发展。  
**[32:31] Speaker B:** So you know exponentials really catch you off guard. So you know the field rising to meet the moment and scientists having the foresight to say well the AI model I had three months ago couldn't help me at all with this but the one I just got today actually is helping me a lot.  
你知道指数级增长真的会让人措手不及。所以这个领域需要迎接这个时刻,科学家们需要有远见地认识到,三个月前的 AI 模型在这方面完全帮不上忙,但今天刚拿到的模型实际上帮了我很多。  
**[32:52] Speaker B:** Just the attention and the foresight to keep pace with the technology and just keep revisiting and understanding how fast it's improving.  
关键是要保持关注和远见,跟上技术的步伐,不断重新审视并理解它改进的速度有多快。  
**[33:04] Speaker B:** So that's what I hope I see. What I hope I don't see is also something we've alluded to, which is the kind of reflexive skepticism.  
这就是我希望看到的。我不希望看到的,也是我们提到过的,就是那种条件反射式的怀疑态度。  
**[33:15] Speaker B:** So I have the benefit I know less about biology than some people on this stage, but I have seen how AI gets applied to many different fields.  
我有个优势是,虽然我对生物学的了解不如台上的一些人,但我见证了 AI 在许多不同领域的应用。  
**[33:28] Speaker B:** And there's a story that's the same everywhere. I've seen it run through when AI first beat the world Go champion to how well it's performing on code to what we've seen with Mythos and Cyber over the last few months to how AI is doing on mathematics to the quality of AI's writing to cite something that like biology is not as...  
而且到处都是同样的故事。我见证了这一切,从 AI 首次击败世界围棋冠军,到它在代码方面的出色表现,到我们在过去几个月看到的 Mythos 和 Cyber,到 AI 在数学方面的表现,到 AI 写作的质量,举个例子,生物学并不像……  
**[33:52] Speaker A:** Not as easily verifiable, and the pattern is always the same within any area, or it's true also for sub-areas of biology. The models are useless, useless, useless. They're dismissed, and then they get to a point where they can help the ordinary practitioner, and then those who are the most skilled, the most advanced, still dismiss them.  
不像那么容易验证,但模式总是一样的,在任何领域都是如此,生物学的子领域也是这样。这些模型没用、没用、没用。人们不屑一顾,然后它们发展到能够帮助普通从业者的程度,而那些最有技能、最先进的专家仍然对它们不屑一顾。  
**[34:13] Speaker A:** They're like, "Ah, this may help the median person, but it won't actually advance the field."  
他们会说:「啊,这可能对水平中等的人有帮助,但它不会真正推动这个领域前进。」  
**[34:16] Speaker A:** And then the exponential does its thing, and they actually get to the point where they can help you a lot.  
然后指数增长发挥了作用,它们真的发展到了能够帮你很多的程度。  
**[34:22] Speaker A:** And it happens so fast, and people get set in their ways because they've seen 12 generations of AI models that don't help them at all.  
而且这发生得太快了,人们已经固步自封,因为他们见过12代对他们毫无帮助的AI模型。  
**[34:29] Speaker A:** And then suddenly overnight the thing—you know, again, the last six months, the thing that's fresh in my mind is cyber.  
然后突然一夜之间——你知道,就是过去六个月,我印象最深刻的是网络安全领域。  
**[34:39] Speaker A:** It's more verifiable than biology. It won't be an exact analogy, but I think it's going to happen.  
它比生物学更容易验证。虽然不会是完全相同的类比,但我认为这会发生。  
**[34:47] Speaker A:** And the thing I don't want is the scientists, the pharmaceutical companies, the regulatory system to be incredibly slow to recognize it, because then we'll get the benefits years later than we would otherwise.  
而我不希望看到的是科学家、制药公司、监管系统对此反应极其迟缓,因为那样的话,我们获得这些好处的时间就会比本可以的时间晚好几年。  
**[35:00] Speaker B:** So I'll start with the negatives. You know, I have witnessed an absolute revolution in science. It's become a global sport, and it creates wonderful variability of opinions, which means we can solve greater questions together.  
那我先说说负面的方面。你知道,我见证了科学领域的一场彻底革命。它已经变成了一项全球性的运动,这创造了观点的精彩多样性,这意味着我们可以共同解决更重大的问题。  
**[35:14] Speaker B:** So I hope—I fear the world will stop collaborating because US—  
所以我希望——我担心世界会停止合作,因为美国——  
**[35:19] Speaker A:** will have their models, China will have their models, Europe not so much yet, right?  
都会有自己的模型,中国会有自己的模型,欧洲目前还不太行,对吧?  
**[35:24] Speaker A:** But that's my fear, right? And my hope is that really within the next year that institutions and companies really see the potential and actually help their people to understand the full implications of how they can be helped with AI.  
但这正是我的担忧所在。我的希望是,在明年之内,各类机构和企业能够真正看到 AI 的潜力,并切实帮助他们的员工充分理解 AI 能够如何帮助到他们。  
**[35:42] Speaker A:** So I want more bilingual people in all teams everywhere, and of course I don't mean people who speak two languages. I mean people who are completely fluent in some scientific topic as well as in digital and AI. And then one person in each team can do wonders. You cannot just say to people 'use AI,' right? You need to actually—they need to be there and embedded in teams. So if everyone does that, and please do, everyone who's listening, because then we can truly accomplish great things together for the benefit of the world.  
所以我希望每个团队里都能有更多「双语人才」,当然我指的不是会说两种语言的人,而是那些既精通某个科学领域、又精通数字技术和 AI 的人。这样每个团队里有一个这样的人就能创造奇迹。你不能只是对大家说「用 AI 吧」,对不对?你需要让这些人真正在那里,融入到团队当中。所以如果每个人都这样做——请所有正在收听的人都这样做——那么我们就能真正携手为世界的福祉成就伟大的事情。  
**[36:19] Speaker B:** Well, I hope maybe this helps some people become a little more fluent. Thank you very much. We're at time.  
好的,我希望这能帮助一些人变得更加精通。非常感谢你。我们的时间到了。  
**[36:42] Speaker B:** I took two things away from that conversation. One, we can't neatly reverse engineer our way to the vision that Dario set forth in Machines of Loving Grace. And two, that shouldn't inhibit us from simply starting.  
我从那次对话中得到了两点启示。第一,我们无法通过简单的逆向工程来实现 Dario 在 Machines of Loving Grace 中提出的愿景。第二,这不应该阻止我们现在就开始行动。  
**[36:58] Speaker B:** And as we've seen AI continue to grow exponentially, organizations that start to track the exponential are beginning to understand that it matters a lot less where you start than the fact that you simply do start. And at Anthropic, the person who's been tasked with starting our journey in life sciences is a gentleman  
随着我们看到 AI 持续呈指数级增长,那些开始追踪这种指数增长的组织逐渐意识到,你从哪里开始远不如你真正开始行动来得重要。而在 Anthropic,负责带领我们开启生命科学之旅的是一位先生  
**[37:19] Speaker A:** named Eric Carter Abrams. I'd like to welcome onto the stage and please join me in doing so, our life sciences leader, Eric Carter Abrams.  
有请 Eric Carter Abrams。我想邀请他上台,也请大家和我一起欢迎我们的生命科学负责人 Eric Carter Abrams。  
**[37:43] Speaker B:** It's great to be here with you all today.  
很高兴今天能和大家在这里相聚。  
**[37:49] Speaker B:** So returning to the image of the exponential and how it affects different disciplines,  
回到指数增长的话题,以及它如何影响不同的学科领域,  
**[37:55] Speaker B:** we've seen it happen in coding where we went very quickly from AI participating in autocompleting work to working perhaps at the level of a junior engineer and then rapidly progressing from there to be increasingly autonomous and capable thanks to the underlying progress in our models and products.  
我们已经在编程领域见证了这一点,AI 从参与自动补全工作,迅速发展到初级工程师的水平,然后又快速进步,变得越来越自主、越来越强大,这都得益于我们模型和产品底层技术的进步。  
**[38:16] Speaker B:** Now, it's going to take longer for this same change to happen in the life sciences because to give us some credit for a moment, it's a much harder problem.  
现在,同样的变革在生命科学领域需要更长的时间才能发生,因为坦白说,这是一个困难得多的问题。  
**[38:26] Speaker B:** In the life sciences, our feedback loops take longer and they involve running real experiments in the physical world and there's so much uncertainty and noise in all biological data.  
在生命科学中,我们的反馈循环需要更长时间,还涉及在物理世界中进行真实实验,而且所有生物数据中都存在大量的不确定性和噪声。  
**[38:40] Speaker B:** But our message is that though it's harder in the life sciences and it's going to take a little longer, the same change is absolutely coming.  
但我们想传达的信息是,尽管生命科学领域更难,需要更长时间,但同样的变革绝对会到来。  
**[38:51] Speaker B:** For our life science efforts, we have two primary objectives that we're pursuing.  
对于我们在生命科学方面的工作,我们追求两个主要目标。  
**[38:56] Speaker B:** The first is accelerating scientific discovery as an end in itself, pure pursuit of basic research.  
第一个是加速科学发现本身,即纯粹的基础研究追求。  
**[39:01] Speaker B:** And the second is alleviating the burden of disease and aging.  
第二个是减轻疾病和衰老的负担。  
**[39:07] Speaker B:** Everything that we're doing is constructed to build  
我们所做的一切都是为了构建  
**[39:12] Speaker A:** A full stack approach to pursue these objectives as fast as we can.  
我们采用全栈方法,以最快速度去实现这些目标。  
**[39:18] Speaker A:** And the approach that we're taking includes many parts. It starts at the foundational layer with our foundation models Claude.  
我们采取的方法包含很多部分。首先是基础层,也就是我们的基础模型 Claude。  
**[39:27] Speaker A:** It includes the product layer of optimizing the product for scientists, because even with the most capable models in the world, we still need to have the right product features to make the model intelligence integrated into workflows and accessible to scientists.  
还包括产品层面,为科学家优化产品,因为即使拥有世界上最强大的模型,我们仍然需要合适的产品功能,才能将模型智能融入工作流程,让科学家能够使用。  
**[39:41] Speaker A:** And then on top of that, as we'll talk more about today, there's all the work that we're doing with all of our partners and customers and internally to directly pursue these objectives that we have.  
然后在此之上,我们今天会详细讨论的,是我们与合作伙伴、客户以及内部团队一起开展的所有工作,直接推进我们的这些目标。  
**[39:54] Speaker A:** So we'll start at the model layer. Can AI actually tackle scientific problems?  
那么我们先从模型层开始。AI 真的能解决科学问题吗?  
**[40:01] Speaker A:** What we've demonstrated over the course of the last six months or so is rapid progress in the underlying foundation model capabilities. Here we're showing our Claude Opus series going from Opus 4.5, which we released about 6 months ago, to Opus 4.8, our most recent Opus model.  
在过去大约六个月里,我们展示了底层基础模型能力的快速进步。这里展示的是我们的 Claude Opus 系列,从大约六个月前发布的 Opus 4.5,到我们最新的 Opus 4.8 模型。  
**[40:19] Speaker A:** And we're seeing rapid progress in several benchmarks here. We've chosen a few from organic chemistry, bioinformatics, and structural biology.  
我们在这些基准测试中看到了快速进步。我们选择了有机化学、生物信息学和结构生物学的几个基准。  
**[40:28] Speaker A:** And the models over this time frame have gone from being not all that useful to performing at a level that is on par or greater than the average PhD level scientist in these fields.  
在这段时间里,模型已经从不太有用,发展到能够达到甚至超过这些领域普通博士水平科学家的表现。  
**[40:41] Speaker A:** So what else do we need to do to solve the problems that scientists face? As I said before, it's not enough to build  
那么,要解决科学家面临的问题,我们还需要做什么呢?正如我之前所说,仅仅构建模型是不够的  
**[40:47] Speaker A:** Great models. We also need to build great products.  
优秀的模型。我们还需要打造优秀的产品。  
**[40:52] Speaker A:** So the way that we think about it is on top of our foundation models we have a series of products that are designed for different users.  
所以我们的思路是这样的:在我们的基础模型之上,我们有一系列针对不同用户设计的产品。  
**[41:01] Speaker A:** For developers we have Claude Code, for knowledge work we have...  
针对开发者,我们有 Claude Code;针对知识工作,我们有……  
**[41:11] Speaker A:** But for scientists we believe that we need something else.  
但对于科学家,我们认为需要提供不同的东西。  
**[41:15] Speaker A:** There's so many things that are unique about scientific use cases that aren't well captured by the use cases and the workflows of these other fields.  
科学用例有太多独特之处,这些是其他领域的用例和工作流程所无法很好涵盖的。  
**[41:25] Speaker A:** In science, there are all these problems that we have that really have little to do with the model capabilities and everything to do with things like connecting to tens of different databases and perfecting every last iteration of the figures that go into manuscripts.  
在科学研究中,我们遇到的许多问题其实与模型能力关系不大,而是与连接几十个不同的数据库、完善论文中每一版图表这类事情有关。  
**[41:40] Speaker A:** And performing literature reviews and converting between file formats. There's so much work to do that should be captured in the product layer.  
还有进行文献综述、转换文件格式等等。有大量工作应该在产品层面来解决。  
**[41:47] Speaker A:** And so for scientists, I'm very excited to announce that today we're launching our newest product, Claude Science.  
所以针对科学家群体,我非常兴奋地宣布,今天我们推出最新产品 Claude Science。  
**[42:05] Speaker A:** Claude Science is intended to be your AI workbench that drives all of the work that you do across the whole scientific workflow.  
Claude Science 旨在成为你的 AI 工作台,驱动你在整个科研工作流程中的所有工作。  
**[42:13] Speaker A:** I'll go through some of the key features now that make it well suited to scientific use cases. The first is that Claude Science has a rich set of scientific artifacts that it supports that are fully reproducible.  
现在我来介绍一些使它非常适合科学用例的关键特性。首先,Claude Science 支持丰富的科学工件集,并且是完全可复现的。  
**[42:28] Speaker A:** So every time that you create a figure...  
所以每次你创建一个图表时……  
**[42:30] Speaker A:** You have the code history associated with it, so you know the exact set of analyses that got you there.  
你拥有与之关联的代码历史记录,因此你清楚地知道是哪些具体的分析步骤让你得出了这个结果。  
**[42:36] Speaker A:** In addition, it supports a wide array of different types of artifacts.  
此外,它还支持各种各样不同类型的工件。  
**[42:39] Speaker A:** Science is a very visual affair. You need to deal with protein structures and small molecules and multiple sequence alignments, figures in your papers, whole manuscript drafts, and you need to be able to iterate with Claude in real time and explore these different types of artifacts together.  
科学研究是一项高度视觉化的工作。你需要处理蛋白质结构、小分子、多序列比对、论文中的图表、完整的手稿草稿,而且你需要能够与 Claude 实时迭代,共同探索这些不同类型的工件。  
**[42:59] Speaker A:** Next, Claude Science manages your compute and scales on demand.  
其次,Claude Science 会管理你的计算资源并按需扩展。  
**[43:04] Speaker A:** It's increasingly becoming the case that many scientific workflows are leaning more and more on these high-performance scientific computing jobs.  
越来越多的科学工作流程正日益依赖这些高性能科学计算任务。  
**[43:10] Speaker A:** For example, in biology, running folding models and molecular design models.  
例如,在生物学领域运行折叠模型和分子设计模型。  
**[43:12] Speaker A:** So we need Claude Science to, first of all, run wherever your data lives—if it's on your local laptop or your cluster—and it needs to be able to set up, execute, and manage all of the computing jobs that you have.  
因此我们需要 Claude Science 首先能够在你的数据所在的任何地方运行——无论是在你的本地笔记本电脑还是集群上——并且它需要能够设置、执行和管理你所有的计算任务。  
**[43:27] Speaker A:** If you want it to run on your cluster, it will run the jobs on your cluster. If you don't have the compute, you prefer that Claude handles it, it will spin up its own compute and GPUs and get the jobs done.  
如果你想让它在你的集群上运行,它就会在你的集群上运行任务。如果你没有计算资源,你希望 Claude 来处理,它就会启动自己的计算资源和 GPU 并完成这些任务。  
**[43:43] Speaker A:** And the next feature that I'll talk about is it comes ready for each domain on day one.  
我要讲的下一个功能是,它从第一天起就为每个领域做好了准备。  
**[43:49] Speaker A:** And in biology, this means connecting to tons of different databases and specialized tools so that you don't have to go in and manually configure that.  
在生物学领域,这意味着连接到大量不同的数据库和专业工具,这样你就不必手动进行配置。  
**[43:59] Speaker A:** But it is rapidly reconfigurable so that if there are...  
但它可以快速重新配置,因此如果有……  
**[44:04] Speaker A:** Additional tools that you want to connect to, you can set those up very quickly as well.  
如果你想连接其他工具,也可以很快地完成设置。  
**[44:11] Speaker A:** So we have our underlying frontier models and on top of that we have our products: Claude Code for developers, Claude Co-work for knowledge work, and now Claude Science for scientists.  
我们有底层的前沿模型,在此基础上我们推出了系列产品:面向开发者的 Claude Code,面向知识工作的 Claude Co-work,以及现在面向科学家的 Claude Science。  
**[44:26] Speaker A:** And next I will welcome up Alec Terashansky, the engineer who has led the development of Claude Science to walk through a demo.  
接下来,让我们欢迎 Alec Terashansky,他是领导 Claude Science 开发的工程师,将为我们演示产品。  
**[44:50] Speaker B:** I lead product development for Claude Science, but earlier in my career, I spent years as a computational biologist, so I've experienced the toil in the scientific process firsthand.  
我负责 Claude Science 的产品开发,但在我职业生涯早期,我做了多年计算生物学家,所以我亲身体验过科研过程中的种种艰辛。  
**[45:01] Speaker B:** Old databases with poorly documented schemas, pipelines that break when a dependency updates, Jupyter notebooks scattered throughout your messy file system with cells that were executed out of order, all those horrible hours spent making figures in Matplotlib and Illustrator.  
文档糟糕的旧数据库、依赖项更新就会崩溃的流程、散落在混乱文件系统中单元格执行顺序错乱的 Jupyter notebooks,还有那些在 Matplotlib 和 Illustrator 里制作图表的无数痛苦时光。  
**[45:17] Speaker B:** Entire fields are underexplored because the research cycle is too slow and too tedious.  
许多研究领域都未被充分探索,就是因为科研周期太慢、太繁琐。  
**[45:24] Speaker B:** So how can we accelerate science progress overall and for biology in particular? That question led us to build Claude Science, an AI workbench for every stage of scientific research.  
那么我们如何才能加速整体科学进展,特别是生物学领域的进展呢?这个问题促使我们开发了 Claude Science,一个覆盖科研各个阶段的 AI 工作台。  
**[45:36] Speaker B:** We've been running it internally on real problems for months and the results have completely transformed what we think is possible.  
我们已经在内部用它解决实际问题好几个月了,结果彻底改变了我们对可能性的认知。  
**[45:43] Speaker B:** To demonstrate its capabilities, I'm going to walk you through one workflow, the example of a real drug program end to end.  
为了展示它的能力,我将带大家完整走一遍工作流程,以一个真实的药物研发项目为例。  
**[45:51] Speaker A:** The disease is phenylketonuria, PKU.  
这种疾病是苯丙酮尿症,简称 PKU。  
**[45:56] Speaker A:** One broken enzyme and an amino acid builds up until it damages the brain.  
一个失效的酶导致氨基酸堆积,最终损伤大脑。  
**[46:01] Speaker A:** There's an approved small molecule drug for it, but it doesn't work in the most common severe mutation.  
目前有一款已批准的小分子药物,但对最常见的严重突变类型无效。  
**[46:07] Speaker A:** And in fact, that's why other groups are pursuing new therapies for it.  
事实上,这也是为什么其他团队正在研发新疗法的原因。  
**[46:11] Speaker A:** Month one of a program like this is a lot of manual work: literature review, structure prep, doing the genetics, scoping the screen, building the business case.  
这类项目的第一个月需要大量人工工作:文献综述、结构准备、遗传学分析、筛选范围界定、构建商业案例。  
**[46:20] Speaker A:** Three to six weeks before a single experiment can be run.  
需要三到六周才能开始做第一个实验。  
**[46:26] Speaker A:** So to start, I just gave Claude one sentence.  
所以一开始,我只给 Claude 一句话。  
**[46:35] Speaker A:** Find me a stabilizer for PAH variants. That's the enzyme. And get me up to speed on PKU as the indication so we can put a program together.  
帮我找一个 PAH 变体的稳定剂——这是那个酶。然后让我快速了解 PKU 作为适应症的情况,这样我们就能组建项目了。  
**[46:43] Speaker A:** Claude went ahead and built a plan.  
Claude 随即制定了一个计划。  
**[46:52] Speaker A:** So this plan it built has three phases.  
它制定的这个计划分为三个阶段。  
**[46:55] Speaker A:** First, do the landscaping analysis. You can see here it's going to execute this phase with three parallel sub-agents.  
首先是做全局分析。你可以看到这里它会用三个并行的子智能体来执行这个阶段。  
**[47:02] Speaker A:** Claude Science is natively multi-agent and uses sub-agents to execute the work.  
Claude Science 本身就是多智能体架构,会使用子智能体来执行工作。  
**[47:07] Speaker A:** First, it'll study the variant biology.  
首先,它会研究变体的生物学特性。  
**[47:10] Speaker A:** Then it'll look at the structure and assess where the pockets are in the enzyme.  
然后它会查看结构并评估酶上的结合口袋位置。  
**[47:13] Speaker A:** And then it'll build the investment case.  
接着它会构建投资论证。  
**[47:18] Speaker A:** Phase two, Claude will build a library. It'll assemble a set of compounds.  
第二阶段,Claude 会构建化合物库。它会组装一组化合物。  
**[47:24] Speaker A:** It'll fan them out across a bunch of GPUs to assess their binding affinity to the enzyme that we're interested in.  
它会把这些化合物分散到多个 GPU 上,评估它们与目标酶的结合亲和力。  
**[47:32] Speaker A:** Score the results, and then finally, it'll produce the deliverables, ultimately leading to a go/no-go verdict.  
对结果进行评分,最后生成可交付成果,最终给出是否继续推进的决策。  
**[47:41] Speaker A:** Claude, even in Claude Science, will tell you its confidence in the scope and feasibility of the plan.  
Claude 甚至在 Claude Science 中会告诉你它对计划范围和可行性的置信度。  
**[47:46] Speaker A:** Now normally with plans like this, where it can take an hour or two to even conduct the computational screen or do the landscaping analysis, you would iterate with Claude to refine the plan, and in fact that is, you know, that is encouraged to make sure that it's going to do what you want it to do.  
通常对于这类计划,光是进行计算筛选或做景观分析就可能需要一两个小时,你会与 Claude 迭代来优化计划,事实上这是被鼓励的,以确保它会按你的预期执行。  
**[48:00] Speaker A:** Here, for the sake of the demo, I just approved it, and then I'll show you what it came back with.  
这里为了演示,我直接批准了,然后我会展示它返回的结果。  
**[48:10] Speaker A:** So it spun up three sub-agents. You can see them here: Variant Biology, Structure and Pockets, Investment Case.  
它启动了三个子智能体。你可以在这里看到它们:变异生物学、结构与口袋、投资案例。  
**[48:18] Speaker A:** I'll drill into them in a second. The first thing it did though is it confirmed that R48W is the variant that matters. It is in fact the most common severe mutation.  
我待会再深入讲解。它做的第一件事是确认了 R48W 是关键的变异体。事实上,它是最常见的严重突变。  
**[48:30] Speaker A:** And then it did three things that I didn't need to ask it to do.  
然后它还做了三件我没有要求它做的事情。  
**[48:34] Speaker A:** So drilling into the structure and pocket sub-agent, you can see here this is the brief that the parent agent gave.  
深入看 structure and pocket 子代理,你可以看到这是父代理给出的简报。  
**[48:40] Speaker A:** You are the structure and product sub.  
你是结构和产品分支。  
**[48:42] Speaker A:** agent for a PAH stabilizer discovery  
用于 PAH 稳定剂发现项目的子智能体。  
**[48:44] Speaker A:** Campaign. Here's what you have access to. Here's the steps you should, you know, you should complete and here's the...  
这是一个活动。这是你可以访问的内容。这是你应该完成的步骤，然后...  
**[48:50] Speaker A:** Output schema that you should return. Now, because the work is being done through sub-agents here in this product in particular, it's extremely important to make sure that you guys  
你应该返回的输出模式。现在,因为在这个产品中工作是通过 sub-agents 来完成的,所以确保你们能够正确处理这一点就变得极其重要  
**[49:02] Speaker A:** Have as much transparency and visibility into what's going on at any given time as possible.  
在任何时候都尽可能地保持透明度和可见性,了解正在发生的事情。  
**[49:09] Speaker A:** So, at the very bottom of this transcript, it produced two figures.  
所以,在这个记录的最底部,它生成了两张图。  
**[49:14] Speaker A:** The first thing it did is it mapped the mutation onto the crystal structure. So, this is the enzyme. The active site is here and then the mutation site is over here.  
它做的第一件事是将突变映射到晶体结构上。这是酶,活性位点在这里,而突变位点在那边。  
**[49:24] Speaker A:** And you can see it's over 20 angstroms apart. So immediately we can tell that this is likely not a problem with a broken active site.  
你可以看到它们相距超过20埃。所以我们立刻就能判断出这很可能不是活性位点损坏的问题。  
**[49:29] Speaker A:** This is more likely a folding problem. So a stabilizer is the right call.  
这更可能是一个折叠问题。所以使用稳定剂是正确的选择。  
**[49:37] Speaker A:** Then it checked to see if we could do the obvious thing. Is there a pocket where the mutation is?  
然后它检查了我们能否采用最直接的方法。突变位置那里有口袋结构吗?  
**[49:40] Speaker A:** So it checked, found it was essentially a smooth surface. Zero druggability on a scale of 0 to one.  
所以它检查后发现那基本上是一个光滑的表面。在0到1的评分标准下,可成药性为零。  
**[49:48] Speaker A:** So that dead end is closed before we even had it, before we even tried it.  
所以这条死路在我们尝试之前就已经被排除了。  
**[49:54] Speaker A:** And then it built us the business case. It told us who we have to beat, the separatory precedent.  
然后它为我们构建了商业案例。它告诉我们必须超越谁,也就是现有的基准药物。  
**[50:07] Speaker A:** So I came in asking for a stabilizer and it told me why that's right, where we put it or where we don't put it, and who we have to beat.  
所以我一开始是想要一个稳定剂,它告诉我为什么这是对的,我们应该把它放在哪里或不应该放在哪里,以及我们必须超越谁。  
**[50:15] Speaker A:** All this was done before I ran a single computational screen.  
这一切都是在我运行任何一次计算筛选之前就完成了。  
**[50:22] Speaker A:** I want to take a step back a bit and talk about the product primitives that make this possible.  
我想退一步,谈谈使这一切成为可能的产品基础能力。  
**[50:27] Speaker A:** First, Claude for science ships with capabilities in many different domains. Whether it's  
首先,Claude for science 具备许多不同领域的能力。无论是  
**[50:34] Speaker A:** Proteomics, structural biology, chemistry, genomics, literature review—altogether more than 60 plus databases and scientific resources that it has access to.  
蛋白质组学、结构生物学、化学、基因组学、文献综述——总共超过 60 多个数据库和科学资源,它都可以访问。  
**[50:46] Speaker A:** These are through the skills that come built into the product and are available for you to use, as well as the connectors.  
这些能力是通过产品内置的 skills 来实现的,你可以直接使用,还有各种 connectors。  
**[50:54] Speaker A:** Cloud Science comes ready for your domain out of the box. And in the event that it doesn't, it's missing capabilities, the product is extraordinarily customizable.  
Cloud Science 开箱即用,已经为你的领域做好了准备。如果它还缺少某些能力,这个产品的可定制性非常强。  
**[51:03] Speaker A:** You can add a skill simply by chatting with Claude. You can write one from scratch. You can upload it from a zip file, or you can import it from your favorite GitHub repository.  
你可以直接通过和 Claude 对话来添加一个 skill,也可以从零开始写一个,或者从 zip 文件上传,还可以从你喜欢的 GitHub 仓库导入。  
**[51:14] Speaker A:** For MCPs, any local or remote MCP that you have access to, you can add them.  
对于 MCP,任何你能访问的本地或远程 MCP,都可以添加进来。  
**[51:22] Speaker A:** The second pillar of the product that I want to talk about are the artifacts.  
我想讲的产品的第二个核心支柱是 artifacts。  
**[51:29] Speaker A:** Every output that Claude produces comes with its full history attached.  
Claude 生成的每一个输出都附带完整的历史记录。  
**[51:34] Speaker A:** So going back to the figure I showed you initially, if you click here and look into the provenance of this artifact, you'll see that the artifact has the code that produced it, including the input artifacts it depends on, the full execution log of all the cells executed in that session leading up to the production of the artifact, the conversation around the artifact, the exact environmental snapshot that produced this artifact.  
回到我一开始展示的那张图,如果你点击这里查看这个 artifact 的来源,你会看到这个 artifact 包含了生成它的代码,包括它依赖的输入 artifacts,在这个会话中执行的所有 cells 的完整执行日志,一直到生成这个 artifact 的整个过程,围绕这个 artifact 的对话,以及生成这个 artifact 时的精确环境快照。  
**[52:06] Speaker A:** What this means is that you can come  
这意味着你可以  
**[52:08] Speaker A:** Back to the product six months, a year, two years later, every artifact is reproducible by construction.  
回到产品本身,无论是六个月、一年还是两年后,每个生成物都可以通过构建过程完全重现。  
**[52:15] Speaker A:** And because Claude knows how every artifact was made, this supports a lot of incredibly powerful interaction patterns.  
正因为 Claude 知道每个生成物是如何制作的,这就支持了许多极其强大的交互模式。  
**[52:22] Speaker A:** Here, I think the label is hard to see, so I'm going to just click and say this label is hard to see.  
这里我觉得标签不太清楚,所以我直接点击并说这个标签看不清楚。  
**[52:32] Speaker A:** And I want to point out one thing here. I could say this without referring to what I'm labeling specifically, because Claude sees every annotation that you make with its vision capabilities as well as its ability to read text, of course.  
我想在这里指出一点,我可以不用具体说明我在标注什么就能这样表达,因为 Claude 能通过它的视觉能力看到你做的每一个标注,当然还有它的文本阅读能力。  
**[52:44] Speaker A:** So you can send this off. I pre-ran this, so I'll show you over here.  
所以你可以发送这个请求。我提前运行过了,我在这边展示给你看。  
**[52:53] Speaker A:** First message that I sent: these labels are hard to see.  
我发送的第一条消息是:这些标签看不清楚。  
**[52:57] Speaker A:** You can see the first thing it does is, I'll look at how this figure was generated so I can fix the label legibility. It looks at the code provenance.  
你可以看到它做的第一件事是,我会看看这个图表是如何生成的,这样我就能修复标签的可读性。它会查看代码来源。  
**[53:05] Speaker A:** It makes the precise edit and then it saves a new version of the figure, and you can see at the top over here v2.  
它做出精确的修改,然后保存图表的新版本,你可以在这上面看到 v2。  
**[53:13] Speaker A:** And that's the other thing. Every artifact is versioned. Versions are immutable, and each version has its own provenance attached to it.  
还有一点,每个生成物都有版本控制。版本是不可变的,每个版本都附带着自己的来源信息。  
**[53:20] Speaker A:** Later on I thought that this dotted line needed to be more visible, so I asked for it. It was a little bit off center, so I asked it again. All four versions are permanently on the record.  
后来我觉得这条虚线需要更明显一些,所以我提出了要求。它稍微有点偏离中心,所以我又要求了一次。所有四个版本都永久记录在案。  
**[53:36] Speaker A:** And artifacts are checked.  
而且生成物都经过了检查。  
**[53:41] Speaker A:** Going back to the structure and pocket sub-agent,  
回到结构和 pocket 子代理,  
**[53:45] Speaker A:** Underneath every agent is a built-in reviewer that is assessing the accuracy of every claim that the agent is making and every artifact it produces.  
每个 agent 底层都内置了一个审查器,用来评估 agent 所做的每一个声明以及它生成的每一个产物的准确性。  
**[53:51] Speaker A:** You can see here the agent wrote a brief. This is a markdown document.  
你可以看到这里 agent 写了一份简报,这是一个 markdown 文档。  
**[54:00] Speaker A:** The reviewer caught a mistake, injected a notice into the agent thread, the agent corrected it, both versions on the record.  
审查器发现了一个错误,向 agent 线程中注入了一条通知,agent 随即进行了修正,两个版本都有记录在案。  
**[54:10] Speaker A:** You can see the diff here.  
你可以在这里看到差异对比。  
**[54:14] Speaker A:** To give you a sense of the power here, in my own personal research project where I continued my own PhD, at this point, Claude has produced thousands and thousands of artifacts all leading up to one output, a manuscript.  
为了让你体会这个功能的强大,在我个人的研究项目中——也就是我继续攻读博士学位的项目里,到目前为止,Claude 已经生成了成千上万个产物,它们全都汇聚到一个最终输出:一份学术手稿。  
**[54:27] Speaker A:** A manuscript that is by construction fully reproducible end to end. I honestly think that that is the future of science.  
这份手稿从构建方式上就做到了端到端完全可重现。我真诚地认为这就是科学的未来。  
**[54:36] Speaker A:** Okay, going back to the campaign.  
好,回到这个研究项目上。  
**[54:40] Speaker A:** So, what's next?  
那么,接下来是什么?  
**[54:43] Speaker A:** We did the landscaping analysis, the next thing Claude did is it compiled a focused library of 2,200 compounds and then distributed them across 80 GPUs.  
我们完成了全景分析,Claude 接下来做的事情是编译了一个包含 2200 个化合物的聚焦库,然后把它们分配到 80 个 GPU 上进行计算。  
**[54:58] Speaker A:** And that leads us to the third pillar of the product: Compute.  
这就引出了产品的第三大支柱:计算资源。  
**[55:05] Speaker A:** Cloud science is portable. It runs. We wanted to be as much a drop-in replacement for Jupyter notebooks as possible to make sure that computational biologists can use this without worrying about it not running where their data is.  
Cloud science 是可移植的,它能运行在各种环境中。我们希望它尽可能成为 Jupyter notebooks 的无缝替代品,确保计算生物学家能够使用它,而不用担心在他们数据所在的地方无法运行。  
**[55:17] Speaker A:** So it can run on your laptop, it can run on a cloud VM, it can run on a Linux workstation, and it can connect to any SSH host that you have.  
所以它可以运行在你的笔记本电脑上,可以运行在云虚拟机上,可以运行在 Linux 工作站上,而且它可以连接到你拥有的任何 SSH 主机。  
**[55:25] Speaker A:** access to, whether it's your lab cluster, an EC2 machine you provisioned on AWS in the cloud. We also have built-in integrations with Modal as a cloud provider, and also we can access model endpoints through Nvidia.  
可以访问的资源,无论是你的实验室集群,还是在 AWS 云上预配的 EC2 机器。我们还内置了与 Modal 作为云服务提供商的集成,也可以通过 Nvidia 访问模型端点。  
**[55:42] Speaker A:** So Claude collected all 80 jobs, they completed, two of them failed, no matter.  
所以 Claude 收集了全部 80 个任务,它们都完成了,其中两个失败了,不过没关系。  
**[55:51] Speaker A:** And then it did the scoring.  
然后它进行了打分。  
**[55:55] Speaker A:** So 2200 went in, 2100 or so were folded and assessed for binding affinity with the compounds of interest with the enzyme. 723 passed the thresholds, and then four survived a check with a second independent model.  
所以输入了 2200 个,大约 2100 个进行了折叠,并评估了与目标化合物和酶的结合亲和力。723 个通过了阈值,然后有 4 个通过了第二个独立模型的检验。  
**[56:14] Speaker A:** And that leads us to the deliverables.  
这就引出了我们的交付成果。  
**[56:20] Speaker A:** So as you can see, Cloud Science is a very visual product. We have built-in artifact renderers for images, but not just images. As you can see elsewhere here, markdown documents, HTML dashboards, chemistry structures, structure viewers, and more. And we're only going to keep building more because we know that the tail of data types in this space is heavy.  
所以你可以看到,Cloud Science 是一个非常可视化的产品。我们内置了图像的 artifact 渲染器,但不仅仅是图像。正如你在这里看到的其他地方,还有 markdown 文档、HTML 仪表板、化学结构、结构查看器等等。而且我们会继续构建更多,因为我们知道这个领域的数据类型长尾非常庞大。  
**[56:46] Speaker A:** Ultimately Claude produced a dashboard.  
最终 Claude 生成了一个仪表板。  
**[56:51] Speaker A:** Let me zoom out a little bit here.  
让我在这里缩小一点。  
**[56:55] Speaker A:** So on the left here are the compounds we've screened.  
所以左边这里是我们筛选的化合物。  
**[56:59] Speaker A:** Here's the protein with some of the compounds overlaid.  
这里是蛋白质,上面叠加了一些化合物。  
**[57:03] Speaker A:** I'll turn them all off here. At the very bottom of the list, you can't see it because the legend is covering it, but we can always just annotate and say the legend is covering it.  
我把它们都关掉。在列表的最底部,你看不到它,因为图例挡住了,但我们总是可以标注一下,说图例挡住了它。  
**[57:14] Speaker A:** And you can iterate with Claude on HTML  
而且你可以与 Claude 在 HTML 上进行迭代  
**[57:16] Speaker A:** documents as well. But at the bottom of the list is serpterin. This is the approved drug, and is ranked dead last for binding affinity to the pocket.  
还有文档。但在列表底部是 serpterin。这是已获批的药物,在与口袋的结合亲和力排名中垫底。  
**[57:24] Speaker A:** At the top of the list, our top four candidates, you can even see the stabilizing arms of the compound.  
在列表顶部,我们的前四个候选化合物,你甚至可以看到化合物的稳定臂。  
**[57:32] Speaker A:** Claude also produced a ranked list of genes with all the metadata attached of compounds. I mean, the top four are the ones that survived the check with the second model and then the 700 trailing behind them.  
Claude 还生成了一个排序的基因列表,附带了化合物的所有元数据。我是说,前四个是通过第二个模型检查的,然后是排在它们后面的 700 个。  
**[57:46] Speaker A:** And most importantly, it led to the go no-go memo.  
最重要的是,它生成了一份 go/no-go 决策备忘录。  
**[57:52] Speaker A:** Claude gave us a conditional go. It told us the first decisive experiment we would need to do to fund a larger experimental campaign and then it gave us the kill criteria for that experiment. At what point do we decide to look for another approach?  
Claude 给了我们一个有条件的通过。它告诉我们需要做的第一个关键实验,以便为更大规模的实验活动提供资金支持,然后它给出了该实验的终止标准。在什么时候我们应该决定寻找另一种方法?  
**[58:09] Speaker A:** And that's the workflow. One sentence in and a full campaign and a go no-go verdict out. We went all the way from the bench to the boardroom in a single session.  
这就是整个工作流程。输入一句话,输出完整的研究方案和 go/no-go 决策。我们在一次会话中就从实验台走到了董事会。  
**[58:22] Speaker A:** Now, why stop there?  
那么,为什么要止步于此呢?  
**[58:25] Speaker A:** Why do it just for one disease?  
为什么只针对一种疾病呢?  
**[58:28] Speaker A:** In parallel, I had asked Claude to run a month one stabilizer assessment for 100 rare monogenic diseases. For each one, tell me which variant matters, what the structure says, and whether a screen is worth running. It did this with 100 parallel sub-agents. Each one doing the landscaping analysis and assessing whether or not this compound is worth computationally screening.  
与此同时,我让 Claude 对 100 种罕见单基因疾病进行第一个月的稳定剂评估。对于每一种疾病,告诉我哪个变异重要,结构说明了什么,以及是否值得进行筛选。它用 100 个并行的子代理完成了这项工作。每个子代理都在做全景分析,评估该化合物是否值得进行计算筛选。  
**[58:52] Speaker A:** At the end of it, we found out of 100  
最终,我们从 100 种疾病中发现  
**[58:56] Speaker A:** Rare diseases, 32 were worth a full computational screen. Altogether, the total time for this was under an hour.  
罕见病方面,有32种值得进行完整的计算筛选。总共花费的时间不到一小时。  
**[59:04] Speaker A:** Total time for the last session including the entire computational screen, under two hours.  
上一次会话的总时间,包括整个计算筛选在内,不到两小时。  
**[59:12] Speaker A:** So why stop there even? Why a hundred? Why not a thousand? Why not 5,000? Why not 10,000?  
那为什么要止步于此呢?为什么是一百?为什么不是一千?为什么不是五千?为什么不是一万?  
**[59:17] Speaker A:** At this point, scale is no longer an issue because of how good Claude is at actually executing on the campaign and on any computational biology problem in general.  
现在,规模已经不再是问题了,因为Claude在实际执行这类研究任务以及任何计算生物学问题上都表现得非常出色。  
**[59:31] Speaker A:** So, I said at the start, entire fields sit underexplored because the cycle is too slow and expensive. This is what it looks like when it isn't.  
所以,我在开头说过,很多整个领域都未被充分探索,因为研究周期太慢、成本太高。而这就是当这些限制不复存在时的样子。  
**[59:37] Speaker A:** To be honest, it's a dream to be able to build this product because if I had this when I was in the lab, the scale at which I could have worked and the ideas I could have afforded to try would have been completely different.  
老实说,能打造这样的产品是我的梦想,因为如果我在实验室时就有这个工具,我能够开展的工作规模以及能够尝试的想法都会完全不同。  
**[59:50] Speaker A:** And I think that's about to be true for everyone in this room.  
而且我认为在座的各位很快都会拥有这样的能力。  
**[59:56] Speaker B:** To share more about what we've been hearing from our early access partners, please help me in welcoming Eric back to the stage.  
为了分享更多我们从早期访问合作伙伴那里听到的反馈,请大家欢迎Eric重新上台。  
**[01:00:19] Speaker C:** Everything that Alec just showed you is powered by an open ecosystem of connectors to so many different partners throughout the scientific world, including key partners like Benchling, Bolts, Latch Bio.  
Alec刚才展示的所有内容都是由一个开放的连接器生态系统驱动的,这些连接器对接了整个科学界的众多不同合作伙伴,包括Benchling、Bolts、Latch Bio等关键合作伙伴。  
**[01:00:31] Speaker C:** We've designed this to be easy to add connectors to all the tools that scientists need to use every day.  
我们的设计让添加连接器变得很容易,可以对接科学家每天需要使用的所有工具。  
**[01:00:43] Speaker C:** So now I want to go through a few examples of what some of our early access partners have been doing.  
现在我想介绍几个例子,看看我们的一些早期访问合作伙伴都在做什么。  
**[01:00:46] Speaker A:** Access customers have been doing with Claude Science. The first example I'll start with begins with a very familiar experience for those of us who have done genomics research.  
Access 的客户一直在使用 Claude Science 做各种工作。我要讲的第一个例子,对于我们这些做过基因组学研究的人来说,是一个非常熟悉的场景。  
**[01:00:54] Speaker A:** There is a scientist at UCSF who was working on an RNA sequencing data analysis, and before using Claude Science, over the course of a year, something wasn't quite right with the data, and eventually a year later his team figured out that there was a virus contaminating the sample that explained the unusual result.  
UCSF 有一位科学家在做 RNA 测序数据分析,在使用 Claude Science 之前,有一年时间数据一直有些不对劲,最终一年后他的团队才发现是病毒污染了样本,这解释了那些异常结果。  
**[01:01:19] Speaker A:** When they provided the data into Claude Science, on the first go, over the course of a few minutes, Claude Science noticed the virus contaminant, and it's these sort of experiences that are critical for building confidence in using AI so thoroughly in scientific workflows.  
当他们把数据输入 Claude Science 后,第一次尝试,仅仅几分钟内,Claude Science 就发现了病毒污染物。正是这类经验对于建立信心、在科学工作流程中如此深入地使用 AI 至关重要。  
**[01:01:36] Speaker A:** The next example is from a scientist at Manifold Bio doing drug development, and here they call out that with Claude Science they're able to go all the way from raw data to publication quality figures in a single session, going through all of the analyses, all of the visual iterations on the figures, and the whole time having the full history of the changes available.  
下一个例子来自 Manifold Bio 的一位做药物开发的科学家,他们特别指出,使用 Claude Science 能够在一个会话中从原始数据一路做到发表级别的图表,完成所有分析、所有图表的视觉迭代,并且全程保留完整的更改历史记录。  
**[01:01:59] Speaker A:** Our last example hits on what I consider to be one of the most important themes of Claude Science. And this is a professor at the Whitehead Institute who shares how Claude Science is able to take scientists who have primarily experimental biology backgrounds and make them able to perform entire [analyses].  
我们的最后一个例子触及了我认为 Claude Science 最重要的主题之一。这是 Whitehead Institute 的一位教授分享的,他说 Claude Science 能够让那些主要是实验生物学背景的科学家完成整套[分析工作]。  
**[01:02:20] Speaker A:** Workflows that include the computational part as well.  
包含计算部分在内的工作流程。  
**[01:02:24] Speaker A:** And so in all of these examples, the important themes here are accelerating workflows by an order of magnitude in some cases and enabling small teams of scientists to do what previously would have taken, you know, much larger teams including a larger array of different skill sets.  
所以在所有这些例子中,重要的主题是将工作流程加速一个数量级,同时让小型科学家团队能够完成以前需要更大规模团队才能做到的工作,那些团队需要更广泛的不同技能组合。  
**[01:02:46] Speaker A:** So in the life science world and taking drug development as an example, there's a lot else going on besides R&D. So in this presentation we focused mostly on R&D, but of course there is critical work happening in clinical and regulatory, in commercial, in manufacturing and operations, and our cloud offerings from the models and products that we have are designed to address all of it.  
在生命科学领域,以药物开发为例,除了研发之外还有很多其他工作在进行。在本次演示中我们主要关注研发,但当然在临床和法规、商业、制造和运营方面也有关键工作在进行,我们的云服务以及我们拥有的模型和产品都旨在解决所有这些问题。  
**[01:03:13] Speaker A:** So next I'll go through a demo to help illustrate the different use of our products for different parts of this value chain of a scientist that's doing biostats work using Cloud Code, our product for software developers. Let's see it in action.  
接下来我会演示一个demo,来说明我们的产品在这个价值链的不同环节中的不同用途,演示一位使用 Cloud Code 做生物统计工作的科学家的场景,Cloud Code 是我们面向软件开发者的产品。让我们看看它的实际效果。  
**[01:03:33] Speaker A:** So here we have a scientist that's responsible for biostats at a hypothetical therapeutics company, and they're tasked with the very difficult job of taking a legacy code base that's written in SAS and porting it into Python.  
这里我们有一位负责生物统计的科学家,在一家假设的治疗公司工作,他们面临一项非常困难的任务,就是要把用 SAS 编写的遗留代码库迁移到 Python。  
**[01:03:48] Speaker A:** So to do this they fire up Cloud Code and they give the prompt in that they need to perform this migration, and you see Cloud Code gets to work.  
为了完成这项工作,他们启动 Cloud Code,输入提示说明需要执行这次迁移,然后你会看到 Cloud Code 开始工作。  
**[01:03:58] Speaker A:** The first thing that Cloud Code does is it makes a detailed plan for all of the...  
Cloud Code 做的第一件事是为所有的……制定详细计划  
**[01:04:03] Speaker A:** steps that it's going to follow to perform this job.  
它将遵循的步骤来执行这项任务。  
**[01:04:07] Speaker A:** The first step of the plan is to build a dashboard so that you can follow along and see where the agent is over the whole process.  
计划的第一步是构建一个仪表板,这样你就可以跟踪并查看 agent 在整个过程中的位置。  
**[01:04:13] Speaker A:** After that, it goes through to actually performing the migrations and then critically producing the validation data and all of the documentation that's required by regulations.  
之后,它会继续实际执行迁移,然后关键的是生成验证数据以及法规要求的所有文档。  
**[01:04:24] Speaker A:** So we see it start and you see first Claude has quickly whipped up this dashboard so that you can follow along and very quickly it gets into the work of going module by module and actually converting this codebase from SAS to Python.  
所以我们看到它启动了,你会看到 Claude 首先快速搭建了这个仪表板,这样你就可以跟进,然后它很快就开始工作,逐个模块地将这个代码库从 SAS 转换到 Python。  
**[01:04:40] Speaker A:** So as Claude is working, it's going through the codebase and a key feature of how Claude Code works is that it's set up to flag you and stop if it needs your input on something.  
所以当 Claude 工作时,它会遍历代码库,Claude Code 工作方式的一个关键特性是,如果它需要你对某些事情提供意见,它会设置标记并停下来。  
**[01:04:50] Speaker A:** And we put a lot of work into making sure that the models have the right judgment and knowing when to grab your attention.  
我们投入了大量工作来确保模型具有正确的判断力,知道何时需要引起你的注意。  
**[01:04:57] Speaker A:** So in this case, you can see through the dashboard side by side the legacy SAS along with the migrated Python.  
所以在这种情况下,你可以通过仪表板并排查看旧版 SAS 代码和迁移后的 Python 代码。  
**[01:05:05] Speaker A:** So you can follow along in real time as the agent is performing this job.  
这样你就可以实时跟踪 agent 执行这项任务的过程。  
**[01:05:12] Speaker A:** But as it's going through the course of this analysis, it's going to come upon an issue and flag that it needs our input on something.  
但是在进行分析的过程中,它会遇到一个问题,并标记出它需要我们对某件事提供意见。  
**[01:05:21] Speaker A:** So we can see up there that it's flagging that there's something that it requires our attention for.  
所以我们可以在上面看到它标记出有一些需要我们注意的事情。  
**[01:05:28] Speaker A:** And in this case, when we click into it to go see what's going on, we can see  
在这种情况下,当我们点击进去查看发生了什么时,我们可以看到  
**[01:05:32] Speaker A:** Here that it's detected that the migration is trying to use a package that isn't yet approved, right? And so here we go in and we resolve this particular flag.  
可以看到这里检测到迁移过程试图使用一个尚未批准的包，对吧？所以我们进入并解决这个特定的标记。  
**[01:05:41] Speaker A:** And this is a good representation of the way that we find works best with using these long-running agents.  
这很好地展示了我们发现使用这些长时运行 agent 的最佳方式。  
**[01:05:50] Speaker A:** You need to set them up and give them ambitious tasks, but make sure that they are programmed to get your attention when they need it and have the right judgment.  
你需要设置它们并给它们分配有挑战性的任务，但要确保它们被编程为在需要时能引起你的注意，并具备正确的判断力。  
**[01:05:56] Speaker A:** So here we go through, we resolve that conflict. Over the course of the migration, there are one or two others that will come up.  
所以我们继续进行，解决了那个冲突。在迁移过程中，还会出现一两个其他问题。  
**[01:06:04] Speaker A:** And when we resolve those from there, it quickly goes through and it completes the migration.  
当我们解决这些问题后，它就会快速完成整个迁移。  
**[01:06:09] Speaker A:** And it not only migrates the code, but it produces all of the documentation that's required to perform validation. And at that point, it's complete.  
它不仅迁移代码，还生成执行验证所需的所有文档。到那时，就完成了。  
**[01:06:21] Speaker A:** So what we've seen here in this a little bit sped up demo is a team of agents through Cloud Code performing a migration from SAS to Python of a production biostats codebase that otherwise would have taken a team of software engineers and clinical scientists perhaps a few months and can be completed here within a single session over the course of a few hours.  
所以我们在这个稍微加速的演示中看到的是一组 agent 通过 Cloud Code 执行从 SAS 到 Python 的生产生物统计代码库迁移，这项工作原本需要一个由软件工程师和临床科学家组成的团队花费数月时间，而在这里可以在几个小时的单次会话中完成。  
**[01:06:57] Speaker A:** So let's keep moving now, looking across the broader life science industry.  
那么我们继续，来看看更广泛的生命科学行业。  
**[01:07:06] Speaker A:** People are building with Claude in all of these different areas today from discovery and pre-clinical, manufacturing and regulatory, commercial, clinical, interfacing with lab equipment itself.  
如今人们正在所有这些不同领域使用 Claude 进行构建，从发现和临床前阶段、制造和监管、商业化、临床，到与实验室设备本身的交互。  
**[01:07:18] Speaker A:** And there's already customers doing enterprise-wide deployments that hit on all of these areas today.  
现在已经有客户在进行企业级部署,覆盖了所有这些领域。  
**[01:07:28] Speaker A:** We're very proud to be working with many of the leading organizations throughout the life sciences that have been at the forefront of doing the most aggressive and forward-looking deployments of Claude and AI generally into all facets of their operations.  
我们非常自豪能够与生命科学领域的众多领先机构合作,这些机构一直走在最前沿,以最积极和前瞻性的方式将 Claude 以及 AI 技术全面部署到其运营的各个方面。  
**[01:07:41] Speaker A:** And you can see on the screen here some of their testimonials about this experience.  
你可以在屏幕上看到他们对这一体验的一些评价。  
**[01:07:52] Speaker A:** So coming back to the bigger picture now, our objectives, the reason that we're doing all of this, is to participate in accelerating basic science and reducing the burden of disease and aging.  
那么回到更宏观的层面,我们的目标,也就是我们做这一切的原因,是为了参与加速基础科学研究,并减轻疾病和衰老带来的负担。  
**[01:08:08] Speaker A:** So in doing that, our approach is to start with the model training. Everything rests on the models having the right underlying capabilities.  
为了实现这一目标,我们的方法是从模型训练开始。一切都基于模型具备正确的底层能力。  
**[01:08:20] Speaker A:** And to add on top of that the product layer, as we have announced today with our Claude Science product. We believe that we need the right products to make the model capabilities integrated into workflows and really solve the problems that scientists are facing.  
在此基础上再增加产品层,就像我们今天发布的 Claude Science 产品。我们认为需要合适的产品来将模型能力整合到工作流程中,真正解决科学家们面临的问题。  
**[01:08:35] Speaker A:** But we've also been asking ourselves, what else should we be doing besides training models and building products?  
但我们也一直在问自己,除了训练模型和构建产品之外,我们还应该做些什么?  
**[01:08:43] Speaker A:** And I'm excited to announce that one answer to that question is that we've decided to start running some drug programs ourselves.  
我很激动地宣布,这个问题的一个答案是,我们已经决定自己开始运行一些药物研发项目。  
**[01:08:54] Speaker A:** And we've chosen to do this by running drug programs in the pre-clinical stage, so the earlier stages of discovery, and choosing indications for neglected diseases.  
我们选择在临床前阶段运行药物项目,也就是发现的早期阶段,并选择被忽视疾病的适应症。  
**[01:09:05] Speaker A:** disease. So these are areas that are outside the scope of what the traditional pharma and biotech landscape might consider attractive targets, but nonetheless, you know, have real burden associated with them.  
这些疾病领域超出了传统制药和生物技术行业可能认为有吸引力的靶点范围,但它们确实给患者带来了真实的负担。  
**[01:09:18] Speaker A:** And we're doing this because we believe first and foremost that to build the right models and products and tools that accelerate the whole industry, we need to live it along with all of you. We believe in the power of tight feedback loops. And there's no substitute for having our own experiences alongside you all in the trenches trying to develop drugs. And the second reason is that we're fortunate due to our public benefit mission to be able to go after these neglected diseases that otherwise wouldn't be addressed.  
我们这样做,首先是因为我们相信,要构建能加速整个行业的正确模型、产品和工具,我们需要和大家一起亲身经历这一切。我们相信紧密反馈循环的力量。没有什么能替代我们与大家一起在药物开发的一线战壕中获得的亲身经验。第二个原因是,得益于我们的公益使命,我们有幸能够去攻克这些原本不会被关注的被忽视疾病。  
**[01:09:49] Speaker A:** So we're very excited about this new direction and you'll be hearing more from us about this soon.  
我们对这个新方向感到非常兴奋,很快你们会听到我们更多关于这方面的消息。  
**[01:09:59] Speaker A:** With that, I will welcome to the stage our head of life science partnerships, Jonah Cool, to moderate a discussion with some of our largest customers.  
接下来,让我欢迎我们生命科学合作伙伴关系负责人 Jonah Cool 上台,主持与我们一些最大客户的讨论。  
**[01:10:19] Speaker B:** It's a pleasure to be here. Thank you all for joining us. So in the partnerships and deployment group, we are the beneficiaries like all of you of the models and capabilities that you've been hearing about and we get to ask two key questions. One, where are the frontiers of science that we believe we can have a positive impact on? And two, who are the partners that we believe that we can invest and spend our time with to bring that possibility to reality. Now, as a scientist, I want to  
很高兴来到这里。感谢大家的参与。在合作伙伴关系和部署团队中,我们和大家一样,都是你们刚才听到的这些模型和能力的受益者,我们需要思考两个关键问题。第一,我们相信自己能够产生积极影响的科学前沿领域在哪里?第二,我们相信可以投入时间合作、将这种可能性变为现实的合作伙伴是谁?作为一名科学家,我想要  
**[01:10:46] Speaker A:** See these capabilities and models in as many scientists' hands as possible across academia, biotech, pharma, and the wider ecosystem.  
让尽可能多的科学家能够使用这些能力和模型,覆盖学术界、生物技术公司、制药企业以及更广泛的生态系统。  
**[01:10:54] Speaker A:** But I also know that as scientists get new technologies, that is actually where many experiments begin.  
但我也知道,当科学家获得新技术时,这实际上正是许多实验的起点。  
**[01:11:02] Speaker A:** We've all probably experienced a new technology, whether it be CRISPR, next generation sequencing, or perhaps even a computational collaborator that unearths and opens up our mind to what is possible and how we can pursue that science.  
我们可能都体验过新技术,无论是 CRISPR、下一代测序技术,还是某个计算协作工具,它们能够发掘并打开我们的思维,让我们看到什么是可能的,以及如何去推进这些科学研究。  
**[01:11:15] Speaker A:** But it's also this big question, and even an experimental question, of how we integrate that into our work, how we integrate that into our organizations, and how we start to rethink the questions that are possible.  
但这也是一个重大问题,甚至是一个实验性的问题:我们如何将这些技术整合到工作中,如何整合到我们的组织里,以及如何开始重新思考那些可能性。  
**[01:11:27] Speaker A:** And so it's this topic of how that we're going to jump off and focus on the panel.  
所以我们接下来就从这个「如何」的话题切入,这也是我们小组讨论的重点。  
**[01:11:34] Speaker A:** And it's really a privilege to have an incredible group that come at the pharma industry from very different perspectives but have a similarly expansive view.  
能够邀请到这样一个杰出的团队,真的是一种荣幸,他们从非常不同的角度来看待制药行业,但又都有着同样开阔的视野。  
**[01:11:42] Speaker A:** With that, I would like to welcome our panelists to the stage.  
那么,让我欢迎我们的嘉宾上台。  
**[01:11:49] Speaker A:** First, Chris Boerner, board chair and CEO of Bristol Myers Squibb. Aviv Regev, the executive vice president and head of research and early development at Genentech, and finally Vas Narasimhan, who is both a board member of Anthropic and also the CEO of Novartis.  
首先是 Chris Boerner,Bristol Myers Squibb 的董事会主席兼首席执行官。Aviv Regev,Genentech 的执行副总裁兼研发与早期开发负责人,最后是 Vas Narasimhan,他既是 Anthropic 的董事会成员,也是 Novartis 的首席执行官。  
**[01:12:09] Speaker A:** Please welcome Chris, Aviv, and Vas.  
欢迎 Chris、Aviv 和 Vas。  
**[01:12:27] Speaker B:** Great. Thank you all for joining us.  
很好。感谢大家的参与。  
**[01:12:30] Speaker B:** We're going to jump right in. Aviv, I'd like to start with you.  
我们直接开始吧。Aviv,我想先从你开始。  
**[01:12:32] Speaker B:** Yes, and the earliest stages of discovery and thinking a lot about this notion of compressing biology that—  
好的,在最早的发现阶段,你一直在思考这个「压缩生物学」的概念——  
**[01:12:39] Speaker A:** We've heard a lot and Dario talk about, and you have thought a lot about as well, biology as we know it really is complex in its nature, and so sometimes it's less of, you know, what question you're answering and more the space that you're searching for it in.  
我们听到了很多讨论,Dario 也谈到过,而你也对此思考了很多。我们所知的生物学本质上确实非常复杂,所以有时候问题不在于你要回答什么问题,而更多在于你在什么空间里去寻找答案。  
**[01:12:54] Speaker B:** Help us maybe understand where this notion of lab-in-the-loop and how you're thinking about early stage discovery and integrating AI at Genentech and where that's going.  
能否帮我们理解一下「实验室在环」这个概念的由来,以及你们在 Genentech 如何思考早期阶段的药物发现和 AI 集成,还有这个方向将走向何方。  
**[01:13:01] Speaker B:** So I think it's worth understanding why biology in general is difficult. Drug R&D is worse than the general difficulties of biology, and it comes from certain kind of inherent properties of biology, chemistry, and that world. It's not just complex, it's actually just huge, like huge, huge. All the numbers are big no matter what they are, and it's huge enough that it's bigger than our experimental capacity ever. It gets bigger than, you know, number of cells on the planet, definitely people on the planet, stars in the universe, atoms in the universe. Choose your number, it's bigger than that.  
所以我认为值得先理解为什么生物学本身就很困难。药物研发比生物学的一般性困难还要更糟糕,这源于生物学、化学以及那个世界的某些内在属性。它不仅仅是复杂,实际上是巨大,非常非常巨大。所有的数字都很大,无论是什么。而且它大到超出了我们任何时候的实验能力。它比地球上细胞的数量更大,肯定比地球上人口的数量、宇宙中恒星的数量、宇宙中原子的数量都要大。你随便选个数字,它都比那个更大。  
**[01:13:37] Speaker B:** So that's problem number one. It's not the only one. Second problem is it's multiscale. So you have to think at many, many, many scales. We actually saw that in the beautiful presentations from Eric and others. You have to think at the level of atoms, and then molecules, and then cells, and then tissues, and then patients, and then populations of patients. So that's a lot of scales.  
所以这是第一个问题,但不是唯一的问题。第二个问题是它具有多尺度特性。你必须在许多许多不同的尺度上思考。我们实际上在 Eric 和其他人精彩的演讲中看到了这一点。你必须在原子层面思考,然后是分子,然后是细胞,然后是组织,然后是患者,然后是患者群体。这涉及很多尺度。  
**[01:13:56] Speaker B:** And then on top of that, we have a lot of measurement limitations. So we don't measure one thing for all of its...  
除此之外,我们还有很多测量上的局限性。所以我们无法对一个事物的所有方面进行测量……  
**[01:14:01] Speaker A:** Properties. We kind of get a lot of separate views of any one entity. Say if you look at a cell, you can look at it with many different measures in many different ways.  
属性。我们对任何一个实体都会得到很多不同的视角。比如说,如果你观察一个细胞,你可以用许多不同的测量方法、从许多不同的角度去观察它。  
**[01:14:09] Speaker A:** And then on top of all of those things, that piece that people often call complexity is within this vastness having to wade through it.  
然后在所有这些之上,人们常说的那种复杂性,就是在这种广阔的空间中你必须去摸索前行。  
**[01:14:17] Speaker A:** You can't automate that because it's not exactly the same way. Even when you ask a really similar question to the one you asked before, it's not like perfectly the same.  
你无法将其自动化,因为它不是完全相同的方式。即使你问一个跟之前非常相似的问题,它也不会是完全一样的。  
**[01:14:25] Speaker A:** So you can't just automate it in the standard way. You have to somehow operate through it, which is what humans have done.  
所以你不能用标准的方式将其自动化。你必须以某种方式在其中运作,而这正是人类一直在做的事情。  
**[01:14:30] Speaker A:** Now you look at AI. So when you have really huge spaces kind of nominally, but actually the real dimensionality is lower, well AI is great at that.  
现在你看AI。当你有表面上非常巨大的空间,但实际的维度其实更低时,AI在这方面非常擅长。  
**[01:14:42] Speaker A:** When you look at things that are multiscale and as you move from one scale to another you have some nonlinear transformation, AI has proven itself being really great at that.  
当你观察多尺度的事物,并且当你从一个尺度移动到另一个尺度时会有某种非线性变换,AI已经证明了它在这方面真的非常出色。  
**[01:14:50] Speaker A:** When you have two views of the same thing, language and video, video and audio, whatever, take your pick, AI is great at that.  
当你对同一事物有两种视角时,语言和视频、视频和音频,随便什么,任你挑选,AI在这方面很擅长。  
**[01:14:58] Speaker A:** And that last piece, which is kind of winging through a world that is not exactly specified, although there's kind of a playbook, but it's not precise. Well, AI agents are actually great at that.  
还有最后一点,就是在一个没有被精确定义的世界中随机应变,虽然有某种行动指南,但它并不精确。而AI智能体在这方面其实也很擅长。  
**[01:15:07] Speaker A:** So, that's kind of a great, great, great, and great. And so, now really the promise is that you put all of these things together somehow and it changes.  
所以,这就是擅长、擅长、擅长、再擅长。因此,现在真正的希望在于,你以某种方式把所有这些东西组合在一起,然后它就会带来改变。  
**[01:15:16] Speaker A:** And then it's important to remember that AI is not magic. It really needs certain conditions to be true. It needs, um, surely it needs algorithms.  
然后很重要的是要记住,AI不是魔法。它确实需要某些条件成立。它需要,嗯,它肯定需要算法。  
**[01:15:26] Speaker A:** And models and it needs GPUs, but that's not enough. It needs a lot of data and it needs really the ability to understand whether what it did means anything, and that means that it needs iterations.  
还有模型，它需要 GPU，但这还不够。它需要大量数据，还需要真正理解自己所做的事情是否有意义的能力，这意味着它需要不断迭代。  
**[01:15:37] Speaker A:** So that really leads to this idea of a lab in the loop or a clinic in the loop.  
所以这就引出了「实验室在环」或者「临床在环」这个概念。  
**[01:15:41] Speaker A:** By the way, the faster you are in the clinic, the same idea applies that your data are the basis really for learning your models.  
顺便说一句，你在临床中的速度越快，同样的道理也适用——你的数据才是学习模型的真正基础。  
**[01:15:48] Speaker A:** Your models hold the answer, not the data, but the models hold the answer and you have to shift to that worldview.  
答案掌握在你的模型中，而不是数据本身，是模型掌握着答案，你必须转变到这种世界观。  
**[01:15:56] Speaker A:** But the model will really guide you to the next step because the space is so big. We're not going to measure all of it, but we are going to get to a general model in this way.  
但模型会真正引导你走向下一步，因为这个空间太大了。我们不可能测量所有东西，但我们会通过这种方式得到一个通用模型。  
**[01:16:05] Speaker A:** So it will allow you to iterate, repeat, etc., etc. In some way, it's exactly what biologists and scientists have done all along, except it operates at the scale of biology that before was beyond us, and it actually does actually move the needle.  
所以它会让你能够迭代、重复，等等等等。从某种意义上说，这正是生物学家和科学家一直在做的事情，只不过它运作的生物学尺度是以前我们无法企及的，而且它确实能产生实质性的进展。  
**[01:16:20] Speaker B:** It's not the same as it used to be.  
这已经不像过去那样了。  
**[01:16:22] Speaker A:** And I think Dario touched on this earlier, but like there is a very unique shape of generative biology that I think matches this space in biology so well.  
我想 Dario 之前提到过这一点，但生成式生物学有一种非常独特的形态，我认为它与生物学这个领域非常契合。  
**[01:16:31] Speaker B:** We heard a lot of great, great, great promise. Where is, you know, where is...  
我们听到了很多很棒很棒的承诺。那么在哪里呢，你知道，在哪里……  
**[01:16:37] Speaker A:** Difficult. Yeah, there's a lot where we're falling over here.  
困难的地方。是的，我们在这里遇到了很多问题。  
**[01:16:40] Speaker A:** The first is, first of all, let's start with the fact that it's not enough on its own today.  
首先，第一点是，让我们从一个事实开始，那就是在今天，光靠它本身是不够的。  
**[01:16:45] Speaker A:** Biology and chemistry and so on, they have a huge long tail. So people love...  
生物学、化学等等，它们都有一条巨大的长尾。所以人们喜欢……  
**[01:16:49] Speaker A:** Focusing on examples that are in the areas where data abound and models perform well. But actually for those things to work, yes, the AI might show you a great starting point.  
我们关注的是那些数据丰富、模型表现良好的领域的案例。但实际上,要让这些东西真正发挥作用,AI 确实可能会给你一个很好的起点。  
**[01:16:59] Speaker A:** We have a nice example like that in oncology. The outcome was what I would call an alien. No human would have thought of that.  
我们在肿瘤学领域就有这样一个很好的例子。那个结果我称之为「异类」——没有任何人类会想到那样的方案。  
**[01:17:05] Speaker A:** Humans say the same, but that was just the starting point. When you needed to figure out the mechanism of action, you actually needed the kind of experimental science and thinking that AI cannot help you with right now.  
人类也会这么说,但那只是起点。当你需要弄清楚作用机制时,你实际上需要的是那种实验科学和思维方式,而这是 AI 目前还帮不上忙的。  
**[01:17:16] Speaker A:** There's no data like that. The experiments are all small scale, they're very bespoke, very specific, very imaginative, and that's great. You put these two things together, you get your answer.  
这类数据根本不存在。这些实验都是小规模的,非常定制化、非常具体、非常富有想象力,而这很好。你把这两样东西结合在一起,就能得到答案。  
**[01:17:25] Speaker A:** So that's one layer of real difficulty that needs to be solved. Many of these problems, AI performs in an interesting way, but it doesn't take you all the way.  
所以这是需要解决的一层真正的难题。对于这些问题中的很多,AI 的表现方式很有意思,但它并不能带你走完全程。  
**[01:17:35] Speaker A:** I think it can one day help you get much closer, but what it does let you do is it lets you work in a comprehensive way.  
我认为有一天它能帮你接近得多,但它现在能让你做的是以一种全面的方式工作。  
**[01:17:43] Speaker A:** That's why I say the answer is in the model, not in the data. Before, it was only in the data, and people have to change how they operate with it, and that shift is also quite difficult.  
这就是为什么我说答案在模型里,而不在数据里。以前答案只在数据中,而人们必须改变他们使用它的方式,这种转变也是相当困难的。  
**[01:17:54] Speaker B:** So on the tone of comprehensiveness and also, you know, the breadth of tasks, Chris, I'd like to transition to you.  
那么关于全面性这个话题,还有任务的广度,Chris,我想把话题转向你。  
**[01:18:02] Speaker B:** You've been outspoken and talked a lot about how at BMS you don't want AI simply to accelerate the current processes, but actually transform how you operate and do work. You've also now rolled...  
你一直直言不讳,谈了很多关于在 BMS,你们不希望 AI 只是加速现有流程,而是真正转变你们的运作和工作方式。你们现在也已经推出了...  
**[01:18:15] Speaker A:** So tell us a little bit about both your vision for that and then how it's going.  
那么跟我们讲讲你对此的愿景,以及目前进展如何。  
**[01:18:23] Speaker B:** Well look, I mean at a macro level, our view is that this technology is ultimately going to transform every piece of the value chain in the slides that you've seen in the previous discussion. And while we're still in the early innings of that journey, we think we're already starting to see promise across each of the areas that we're utilizing this technology.  
从宏观层面来看,我们认为这项技术最终会变革价值链的每一个环节,就像你在之前讨论中看到的那些幻灯片展示的那样。虽然我们仍处于这段旅程的早期阶段,但我们认为在应用这项技术的各个领域已经开始看到希望。  
**[01:18:45] Speaker B:** We're placing at the company really three big bets. The first bet is that AI and machine learning can really help us identify those hidden patterns in biology that ultimately will enable us to drug previously undruggable targets and hopefully be able to bring the next new medicine to patients.  
我们在公司层面真正押注了三个大方向。第一个赌注是,AI和机器学习能够真正帮助我们识别生物学中那些隐藏的模式,最终使我们能够针对以前无法成药的靶点开发药物,并有望将下一代新药带给患者。  
**[01:19:06] Speaker B:** We're making good progress. We already today, all of our small molecules and a large percentage of our large molecules go through an AI screening and validation process before they ever get into the wet lab. So we feel good about where we are there. There's a lot more work to do.  
我们正在取得良好进展。如今我们所有的小分子药物,以及很大比例的大分子药物,在进入湿实验室之前都会经过AI筛选和验证流程。所以我们对目前的进展感到满意,当然还有很多工作要做。  
**[01:19:21] Speaker B:** The second bet is that AI will fundamentally change drug development. It will reduce the times, it will reduce the cost, and hopefully improve the probability of success.  
第二个赌注是,AI将从根本上改变药物开发。它会缩短时间,降低成本,并有望提高成功概率。  
**[01:19:31] Speaker B:** We've set the target internally that we can reduce cycle times by 30%. We're well on our way to that, will likely beat that target. Good progress there.  
我们在内部设定的目标是将周期时间缩短30%。我们正在朝着这个目标稳步前进,很可能会超额完成这个目标。这方面进展不错。  
**[01:19:40] Speaker B:** And third, we do believe this technology will create productivity.  
第三,我们确实相信这项技术将创造生产力。  
**[01:19:46] Speaker A:** Tailwinds really across the organization, and we've got thousands of use cases that we can point to.  
这在整个组织内都形成了顺风,我们可以列举出成千上万个应用案例。  
**[01:19:51] Speaker A:** If you double click on that, we rolled out the first ChatGPT models in early January, February of 2023, very early on.  
如果深入了解的话,我们在2023年1月、2月初就推出了第一批 ChatGPT 模型,非常早期。  
**[01:20:00] Speaker A:** Today, over 30,000 employees have access to a whole suite of AI tools.  
如今,超过3万名员工可以使用一整套 AI 工具。  
**[01:20:06] Speaker A:** And so we've generated thousands and thousands of use cases that we believe ultimately will show a 5 to 10% at a minimum increase in productivity across various parts of the organization.  
因此我们创造了成千上万个应用案例,我们相信最终将在组织的各个部门实现至少5%到10%的生产力提升。  
**[01:20:19] Speaker A:** So great promise. We're in the early part of the journey, but it's a journey this industry ultimately needs to take, because Vos, you can speak to this as well.  
前景非常好。我们还处在旅程的早期阶段,但这是这个行业最终必须踏上的旅程,因为 Vos 你也可以谈谈这一点。  
**[01:20:26] Speaker A:** The business model of this industry will change over time, and we think AI will enable that.  
这个行业的商业模式会随着时间而改变,我们认为 AI 将推动这种改变。  
**[01:20:32] Speaker A:** And if we're successful, it should make us better at delivering on our mission, which is bring more medicines to patients faster and ultimately change patient outcomes.  
如果我们成功了,它应该能让我们更好地实现我们的使命,也就是更快地将更多药物带给患者,并最终改善患者的治疗结果。  
**[01:20:41] Speaker B:** On that note, Vos, you are a physician, have developed drugs, now the CEO of Novartis.  
说到这里,Vos,你是一名医生,开发过药物,现在是 Novartis 的首席执行官。  
**[01:20:50] Speaker B:** We've spoken about the lab, we've talked about the operations. I'd love to hear your perspective on the vision and where you see this work impacting patients.  
我们谈了实验室,也谈了运营。我很想听听你对愿景的看法,以及你认为这项工作会在哪些方面影响患者。  
**[01:20:58] Speaker C:** Yes, so thanks. It's great to be here also with two great colleagues.  
是的,谢谢。很高兴能和两位优秀的同事一起来到这里。  
**[01:21:03] Speaker C:** You know, I think it's worth putting it in perspective. When you look at the 120 years of this sector, we've really only discovered around 800 to 1,000 medicines.  
你知道,我认为有必要从全局来看这个问题。当你回顾这个行业120年的历史,我们实际上只发现了大约800到1000种药物。  
**[01:21:13] Speaker C:** And actually, when you look at actual mechanisms of action that are  
而实际上,当你看真正的作用机制时  
**[01:21:18] Speaker A:** Notable, it's actually quite small. I mean this is an industry where we spend 150 to 200 billion a year in drug R&D amongst the bigger companies, and yet we've only found a handful of medicines, which I think actually shows you how hard this is—unpacking billions of years of evolution.  
值得注意的是,这个数字其实相当小。要知道这是一个每年在药物研发上投入1500到2000亿美元的行业,而且是那些大型公司的投入,但我们却只找到了屈指可数的几种药物,我认为这恰恰说明了这件事有多难——要去解析数十亿年的进化过程。  
**[01:21:35] Speaker A:** And so the tools now that you guys are talking about here today can hopefully get us to another level, but I think it's important to break down how that's going to work.  
所以你们今天在这里讨论的这些工具,有望把我们带到另一个层次,但我认为重要的是要详细分析这将如何实现。  
**[01:21:45] Speaker A:** So when you think about it from a development time standpoint, I think of three categories of latency that drive drug development timelines.  
所以从开发时间的角度来看,我认为有三类延迟因素在驱动着药物开发的时间线。  
**[01:21:54] Speaker A:** Information latency, operational latency, and biological latency.  
信息延迟、运营延迟和生物学延迟。  
**[01:21:56] Speaker A:** And information and operational latency are actually about 40% of this time.  
而信息延迟和运营延迟实际上占了整个时间的大约40%。  
**[01:22:01] Speaker A:** And I think with the tools you saw today, we can bring information latency down almost to zero. I mean, the information will be at scientists' fingertips.  
我认为借助你们今天看到的这些工具,我们可以把信息延迟降低到几乎为零。我的意思是,信息将触手可及,就在科学家们的指尖。  
**[01:22:12] Speaker A:** Operational latency we can probably reduce significantly. This is organizing trials, organizing experiments, getting all of the work to happen in a large organization.  
运营延迟我们可能可以大幅减少。这包括组织试验、组织实验,以及让大型组织中的所有工作得以进行。  
**[01:22:22] Speaker A:** But I think the biological latency we're stuck with. I mean that is actually having to run the experiment in an animal model and a cellular model or in humans, and that's about 60% of this timeline.  
但我认为生物学延迟是我们无法避免的。我的意思是,我们必须实际在动物模型、细胞模型或人体中进行实验,而这占了整个时间线的大约60%。  
**[01:22:36] Speaker A:** So what does that translate to? That means that I think, as Chris rightfully said, you can get this down from 12 years—from when we actually have a candidate to the end of this journey—down to seven to eight years, which if you compound...  
那么这意味着什么呢?我认为正如Chris正确指出的那样,你可以把这个时间从12年——也就是从我们实际拥有候选药物到整个旅程结束——缩短到7到8年,如果你把这个复合计算的话......  
**[01:22:49] Speaker A:** over this entire industry is massive. So I think that's where we'll get the speed, and then when you think about probability of success, again, I think of four components in general.  
对整个行业的影响是巨大的。所以我认为这就是我们获得速度提升的地方,然后当你思考成功概率时,我会从四个方面来考虑。  
**[01:23:00] Speaker A:** Drugs fail in phase one because of safety. Can we get a lot better at predicting safety with these models? I hope so. We're certainly trying.  
药物在一期临床失败是因为安全性问题。我们能否用这些模型更好地预测安全性?我希望可以。我们肯定在努力尝试。  
**[01:23:08] Speaker A:** I think all of us are trying to do better on that front. Second is the biophysical properties of the molecule.  
我认为我们所有人都在努力做得更好。第二是分子的生物物理特性。  
**[01:23:14] Speaker A:** There I think these models will have a huge benefit. We can leverage all of the historical knowledge to design a better drug that hopefully is better behaved, more manufacturable.  
在这方面我认为这些模型会带来巨大的好处。我们可以利用所有历史知识来设计更好的药物,希望它表现更好、更易于生产。  
**[01:23:25] Speaker A:** Third is the patient selection and actually getting the right indication—there remains to be seen. Hopefully we can really get there.  
第三是患者选择和真正找到正确的适应症,这还有待观察。希望我们能真正做到。  
**[01:23:32] Speaker A:** The hard one again is the underlying biology of: is this a good drug target for this disease? And that comes back to the first topic, the first thing I said—that's really hard.  
再次强调,困难的是底层生物学问题:这个靶点对这种疾病来说是好的药物靶点吗?这又回到了我说的第一个话题,那真的很难。  
**[01:23:43] Speaker A:** And in the end, after 10 years, we might learn that some of these targets aren't the right targets. So take that all together, I think we could probably move from 8% to 16%.  
最终,10年后我们可能会发现其中一些靶点并不是正确的靶点。所以综合考虑这一切,我认为我们大概可以从8%提升到16%。  
**[01:23:52] Speaker A:** And if we were actually able to, across all of our pipelines, go from 12 years to seven years and 8% POS to 16%, the impact on public health is massive.  
如果我们真的能够在所有管线中把时间从12年缩短到7年,把成功概率从8%提高到16%,对公共卫生的影响将是巨大的。  
**[01:24:03] Speaker A:** And it's important to understand that those sound like small moves, but compounded over the size of our industry pipelines, would be big.  
重要的是要理解,这些听起来像是小的改进,但在整个行业管线的规模下复合起来,影响会很大。  
**[01:24:11] Speaker A:** And then hopefully some of that means that more diseases get treated, undruggable targets get drugged.  
然后希望这意味着更多疾病能得到治疗,原本无法成药的靶点能够成药。  
**[01:24:17] Speaker A:** And we have, you know, a much bigger public health impact.  
这样我们就能产生更大的公共卫生影响。  
**[01:24:20] Speaker B:** Yeah, those kinds of compounding effects even if it's not perfection, right? And yesterday chatting with Lotus, she gave this great example of the four years and several thousand compounds that it took to test and get the stability right of GLP-1 and semaglutide. And again, 25%, you know, if you go to a thousand compounds in a year, let alone maybe even more aggressive, you know, that's an incredible acceleration and impact.  
是的,即使不完美,那种复合效应也很显著,对吧?昨天和 Lotus 聊天时,她举了个很好的例子,说为了测试并确保 GLP-1 和 semaglutide 的稳定性,花了四年时间测试了几千种化合物。如果一年能测试一千种化合物,甚至更激进一些,那就是 25% 的提升,这是非常惊人的加速和影响。  
**[01:24:44] Speaker A:** All the steps of this process are very difficult.  
这个过程的每一步都非常困难。  
**[01:24:46] Speaker B:** Very difficult.  
非常困难。  
**[01:24:47] Speaker A:** And every step that you move the needle on is great. And there is not one that if you just solve that one, everything else would be solved. So even if you had the perfect oracle for targets, you would still have to drug them. You would still have to find the right patients. You would still have to execute a clinical trial. That's why it's kind of nice that you can take an all of the above approach and any benefit that you get is a real benefit as it compounds. And also as models can let you work more end to end rather than look at each step separately, then they also can be more foreshadowing challenges. Think about safety way earlier than you would normally think because you have so much that's already happening on the virtual side and on the targets. We look actually as an industry at a very small sliver of them because of this difficulty to manage the...  
每一步取得进展都很好。并不存在解决了某一步就能解决所有问题的情况。即使你有完美的靶点预测工具,你还是得开发药物,还是得找到合适的患者,还是得执行临床试验。所以「全方位推进」的策略很不错,你获得的任何收益都是真实的,而且会不断累积。而且随着模型让你能够更端到端地工作,而不是分别看待每个步骤,它们也能更好地预见挑战。你可以比通常更早地考虑安全性问题,因为在虚拟层面和靶点层面已经有很多工作在进行。实际上,作为一个行业,我们只关注了其中很小的一部分,就是因为管理这些的难度...  
**[01:25:33] Speaker A:** Comprehensiveness and because of the inability to really track experimentally what you call the biological latency.  
全面性,以及因为无法通过实验真正追踪你所说的生物学延迟。  
**[01:25:39] Speaker A:** But if some of this biological latency, as we have seen, can start becoming virtualized, you can run, for example, not just a small molecule virtual screen for in vitro. You can also run small molecule virtual screens phenotypically for cells.  
但如果我们已经看到的这些生物学延迟中的一部分可以开始被虚拟化,你就可以运行,比如说,不仅仅是体外的小分子虚拟筛选,你还可以针对细胞运行表型层面的小分子虚拟筛选。  
**[01:25:53] Speaker B:** Which is possible now. And when you do that, then biology becomes a lot more accessible in a much broader sense than it has been historically.  
这在现在是可能的。当你这样做的时候,生物学就会变得比历史上任何时候都更加易于获取,而且是在更广泛的意义上。  
**[01:26:00] Speaker B:** But I think what you're hearing really across the collective is that there's a lot of possibility and opportunity here with this technology. But we also need to make sure we don't set expectations for what we're going to be...  
但我认为你从整个群体那里真正听到的是,这项技术带来了很多可能性和机会。但我们也需要确保不要对我们将要...  
**[01:26:13] Speaker A:** able to accomplish that we simply can't deliver on. So when you hear the sort of "we're going to cure cancer in our lifetime,"  
能够完成的目标,而我们根本无法兑现。所以当你听到那种「我们将在有生之年治愈癌症」的说法时,  
**[01:26:19] Speaker B:** we're going to make a lot of progress on cancer in our lifetime, but let's—we don't want to get over our skis.  
我们确实会在有生之年在癌症治疗上取得很大进展,但是——我们不想过度承诺。  
**[01:26:24] Speaker B:** So maybe on this topic of both what didn't work, what is challenging—we just heard that we are starting to pursue some of our preclinical work internally. The big goal there is both to focus on neglected diseases, but also for ourselves to learn these lessons and learn alongside you and really have skin in the game.  
所以也许在这个话题上,谈谈什么没成功、什么有挑战性——我们刚听说我们开始在内部推进一些临床前研究工作。那里的主要目标既是关注被忽视的疾病,同时也是让我们自己学习这些经验教训,与你们一起学习,真正参与其中、承担风险。  
**[01:26:44] Speaker B:** You all have lived this experience. So, you know, phoning some friends here as we get into this. You know, what has not worked? What are the lessons that you  
你们都经历过这一切。所以,你知道,在我们开始做这件事时向朋友们请教一下。那么,什么东西没有奏效?你们学到的经验教训是什么  
**[01:26:54] Speaker A:** Would kind of impart and, you know, set those expectations at the right time?  
会在合适的时机传达并设定这些预期吗?  
**[01:26:57] Speaker B:** Yeah. Well, maybe I'll start from our standpoint. I mentioned that we have thousands of use cases.  
是的。那么,我先从我们的角度来说吧。我之前提到过,我们有数千个使用场景。  
**[01:27:02] Speaker B:** When we started rolling out this technology, our intention was to let a thousand flowers bloom. Let people take this technology and do what they would like and try to improve their ability to be productive and efficient.  
当我们开始推广这项技术时,我们的初衷是让百花齐放。让大家使用这项技术,做他们想做的事情,尝试提高他们的生产力和效率。  
**[01:27:13] Speaker B:** And that worked. But what we found was that the use cases tended to be quite narrow and were very difficult to scale.  
这个策略确实有效。但我们发现,这些使用场景往往非常狭窄,而且很难扩展。  
**[01:27:24] Speaker B:** And that sort of makes sense, right? Because you're developing these tools for a very individualized case. You're putting it on top of processes, but you're not changing the process itself.  
这也说得通,对吧?因为你是在为非常个性化的场景开发这些工具。你只是把它叠加在现有流程之上,但并没有改变流程本身。  
**[01:27:34] Speaker B:** And so we really were struggling with scaling. I'll give you one sort of somewhat mundane example, but we knew two years ago that AI could predict and forecast our business and many parts of our businesses anyway better than the battalions of people that we had actually doing the forecasting, but we couldn't scale it.  
所以我们在扩展方面确实遇到了困难。我举一个相对平常的例子,两年前我们就知道,AI在预测和预报我们的业务以及业务的许多部分时,反正都比我们大批从事预测工作的员工做得更好,但我们无法将其扩展开来。  
**[01:27:53] Speaker B:** And the reason we couldn't scale is when you talk about changing the forecast and planning process of a company of our size, that's not a business analytics exercise. That's a finance, commercial, manufacturing—the list goes on—and you simply couldn't get the scale we needed.  
我们无法扩展的原因是,当你谈到改变我们这种规模公司的预测和规划流程时,这不是一个业务分析练习。这涉及财务、商业、制造等等,名单很长,你根本无法达到我们所需的规模。  
**[01:28:09] Speaker B:** So what we realized very quickly is that we needed to supplement this sort of bottom-up innovation with a very robust and...  
所以我们很快意识到,我们需要用一个非常强大的……来补充这种自下而上的创新。  
**[01:28:17] Speaker A:** rigorous top-down approach. So we created this AI accelerator in the company. We had my team say where within the key verticals in the organization can you get real productivity improvements?  
严格的自上而下的方法。所以我们在公司内部创建了这个 AI 加速器。我让我的团队去找,在组织的关键垂直领域中,哪里能获得真正的生产力提升?  
**[01:28:26] Speaker A:** Where can you change the processes leveraging this technology?  
哪里可以利用这项技术来改变流程?  
**[01:28:29] Speaker A:** We put small teams that had six to eight weeks to prove out the concept and then we pushed to scale that and so we have I think 40 or 50 projects in the incubator now.  
我们组建了小团队,给他们六到八周的时间来验证概念,然后我们就推动规模化,所以现在孵化器里大概有 40 或 50 个项目。  
**[01:28:43] Speaker A:** We've gotten 30 ongoing efforts that are pretty far down the path.  
我们有 30 个正在进行的项目已经推进得相当深入了。  
**[01:28:48] Speaker A:** So this ability to take bottoms up and top down approaches to these problems so that you can overcome the organizational and people issues associated with fully leveraging this technology is super important.  
所以这种能够同时采取自下而上和自上而下方法来解决这些问题的能力非常重要,这样你才能克服充分利用这项技术时面临的组织和人员问题。  
**[01:28:59] Speaker B:** So I'll give you a perspective more from the research side.  
那我从研究这一侧给你一个视角。  
**[01:29:04] Speaker B:** So before development just to complement what Chris described is actually what we would call reshaping of a process.  
所以在开发之前,作为对 Chris 所描述内容的补充,实际上我们称之为流程重塑。  
**[01:29:10] Speaker B:** Like the AI problem is actually not an AI problem at all. You kind of know how to solve it. It demonstrates success.  
就像这个 AI 问题其实根本不是 AI 问题。你大概知道怎么解决它。它已经证明了成功。  
**[01:29:15] Speaker B:** How do you take the whole company's process to change?  
你怎么让整个公司的流程发生改变?  
**[01:29:19] Speaker B:** But on the research side, we didn't come from a million flowers bloom. We came from a very particular hypothesis that is now years old.  
但在研究这一侧,我们不是百花齐放的路子。我们源于一个非常特定的假设,这个假设现在已经有好几年了。  
**[01:29:25] Speaker B:** So we started in 2020 on this that is like a big idea this lab in the loop that you can really change how you do the drug discovery part the target and the drug discovery part and that required several shifts in like scientific work.  
所以我们在 2020 年就开始做这个,这是一个大想法,就是实验室在环(lab in the loop),你可以真正改变药物发现的方式,包括靶点和药物发现部分,这需要在科学工作方式上做出几次转变。  
**[01:29:40] Speaker B:** So for example we needed certain styles of data to exist. So we invested a lot in data generation.  
比如说我们需要某些特定类型的数据存在。所以我们在数据生成方面投入了大量资源。  
**[01:29:47] Speaker A:** capacity that would be the right fit for what AI needs. And then something unexpected happened. It wasn't a problem to set up the data generation capacity, but once the data were generated, you know, these are kind of cool data. People forgot that the goal was to end up with an AI model. They started looking at the data.  
能够满足 AI 需求的合适容量。然后意想不到的事情发生了。建立数据生成能力本身并不是问题,但是一旦数据生成出来之后,你知道,这些数据真的很酷。人们忘记了最终目标是要得到一个 AI 模型。他们开始专注于研究数据本身。  
**[01:30:06] Speaker B:** And it wasn't just the experimental people, the computational people too, because there's like real results there. We of course built them for our systems and so on.  
而且不仅仅是实验人员,计算人员也是如此,因为那里有真实的结果。我们当然会把这些用于构建我们的系统等等。  
**[01:30:13] Speaker A:** So we had to pivot for that and we called something foundational datasets for foundation models as an initiative so that people remember that the goal is to train the model. And at first the reaction was like, but why are you making me do this? The data are there. I want this for my problem, for my system. And within not a long period of time, a few months, people were like, why is the foundation model not trained yet? They were realizing how much more is in a model than in data.  
所以我们不得不为此做出调整,我们推出了一个叫做「foundation models 的基础数据集」的计划,让人们记住目标是训练模型。一开始大家的反应是:为什么要让我这么做?数据都在这里了,我想用这些数据解决我的问题、我的系统。但在不长的时间内,几个月后,人们开始问:foundation model 怎么还没训练好?他们逐渐意识到模型中蕴含的东西远比数据本身多得多。  
**[01:30:42] Speaker A:** And similarly, and this might actually be useful for you, once you start on projects, people want that molecule to succeed.  
同样地,这一点对你可能会有用,一旦你开始做项目,人们会希望那个分子能够成功。  
**[01:30:48] Speaker B:** And now our goal is not to learn how to do the process in general, and rightfully so, because we want to change the lives of patients in the end. It's not to make the model general. It's not to learn lessons. It's to actually get a project into patients.  
现在我们的目标不是学习如何泛化这个流程,这也是理所当然的,因为我们最终想要改变患者的生活。目标不是让模型更通用,不是总结经验教训,而是真正把项目推进到患者那里。  
**[01:31:04] Speaker A:** And when that happens, they're like, why should I spend synthesis money just to make the model better? Even though that would be good for all projects, I need to move my project forward.  
当这种情况发生时,他们会说:我为什么要花合成实验的钱只是为了让模型变得更好?虽然这对所有项目都有好处,但我需要推进我自己的项目。  
**[01:31:11] Speaker A:** Project. And so shifting that also required some thinking and work. How do you still move all your projects but also generalize from it? And that's very natural when your goal is to impact patients but you still want to think big, so you have to build something in there. And like, you know, it's kind of in all molecules now there's something with AI, sometimes a big thing, sometimes a smaller thing depending on the project, the challenges and so on, but there's something. And that's a big shift from a few years ago.  
项目。所以这种转变也需要一些思考和努力。你如何既推进所有项目,又能从中进行泛化呢?当你的目标是影响患者,但同时又想要有大格局的思考时,这是很自然的,所以你必须在其中构建一些东西。就像,你知道的,现在几乎所有分子中都有 AI 的成分,有时候是很大的部分,有时候是较小的部分,取决于项目和挑战等等,但总有一些 AI 的存在。这与几年前相比是一个巨大的转变。  
**[01:31:43] Speaker B:** Maybe the only thing I'll add in terms of advice, I think all the math shows when you look at pipelines that to Viv's point, before phase 2A, you need to set high bars and be willing to let things fail. And I think one of the...  
也许我在建议方面唯一要补充的是,我认为当你看药物管线时,所有的数据都表明,正如 Viv 所说的,在 2A 期之前,你需要设定高标准,并且愿意让一些项目失败。我认为其中一个...  
**[01:31:57] Speaker A:** Challenges we're going to have with AI is the ability to generate molecules is going to increase dramatically. So decision-making criteria is going to be really important.  
我们在 AI 领域即将面临的挑战是,分子生成能力将会大幅提升。因此,决策标准将变得非常重要。  
**[01:32:04] Speaker A:** So setting really good experiments whether it's in pre-clinical or in early clinical to really kill early based on hard data.  
所以无论是在临床前还是早期临床阶段,都要设计非常好的实验,以便基于确凿数据及早淘汰不合格的候选药物。  
**[01:32:12] Speaker A:** And then the other part of the math is once you find an active drug and it actually is hitting the target and has a pharmacodynamic effect, you need to find its use case in late stage and that's where you don't want to give up too soon, and the examples are many in our industry where we eventually after many years actually find the right use case in late stage for a drug.  
另一方面,一旦你找到了一个有活性的药物,它确实击中了靶点并产生了药效学效应,你就需要在后期阶段找到它的适用场景,这时候你不能太早放弃。在我们这个行业有很多例子,最终经过多年努力,我们确实在后期为某个药物找到了正确的适用场景。  
**[01:32:34] Speaker B:** I have to comment on that and the decision-making because I agree so much for  
我必须对此以及决策问题发表一下看法,因为我非常赞同  
**[01:32:38] Speaker A:** The entries into the research portfolio more than doubled in two years. Doubled is a lot more than like 10%. It was, you know, 130% or so.  
研究项目组合的入选数量在两年内增长了一倍多。翻倍可比 10% 多得多，实际上是 130% 左右。  
**[01:32:48] Speaker A:** That's huge. And we didn't add any more biologists during that process.  
这是个巨大的增长。而且在这个过程中我们没有增加任何生物学家。  
**[01:32:53] Speaker A:** It's not just the AI, it's also the data and data generation capacity that we had.  
这不仅仅是因为 AI，还因为我们拥有的数据和数据生成能力。  
**[01:32:56] Speaker A:** And so actually how you manage the process of vetting and entering became also a thing we had to change, because the old process was just not made for this type of approach.  
所以实际上我们如何管理审核和录入流程也成了必须改变的事情，因为旧流程根本不是为这种方法设计的。  
**[01:33:10] Speaker A:** And now if you multiply it across the whole industry, and tools like the ones that we saw today will surely get there, then the volume just in the whole system all of a sudden rises. And there is also the risk that people will do the same thing again and again and again, which is not actually a very desirable outcome.  
现在如果你把这个规模扩展到整个行业，而且像我们今天看到的这些工具肯定会达到那个水平，那么整个系统中的工作量会突然激增。而且还存在一个风险，就是人们会一遍又一遍地做同样的事情，这其实不是一个理想的结果。  
**[01:33:28] Speaker A:** So I think there's a lot of room for thinking about how we make our decisions. Yeah. Yeah, in science.  
所以我认为我们在如何做决策方面还有很大的思考空间。是的，在科学研究中。  
**[01:33:34] Speaker B:** The vision and perspective that you all have of leading some of the largest organizations in the world, and the challenges of scale here, are very real ones.  
你们作为世界上一些最大组织的领导者所拥有的视野和视角，以及这里涉及的规模化挑战，都是非常真实的。  
**[01:33:43] Speaker B:** And with that scale, of course, also comes difficulties and a change of shape of how work is getting implemented, the power, the potential.  
而伴随着这种规模，当然也会带来困难，以及工作实施方式的形态变化、力量和潜力。  
**[01:33:54] Speaker B:** And there's risks associated with that. And perhaps one of the risks that we've chatted a little bit about is this creep towards mediocrity.  
这其中也存在风险。我们之前稍微讨论过的一个风险，就是这种向平庸化蔓延的趋势。  
**[01:34:02] Speaker B:** And especially if we're talking about, you know, again, all phases of science from discovery to drug development to...  
特别是如果我们讨论的是科学的所有阶段，从发现到药物开发再到……  
**[01:34:08] Speaker A:** Patients, what we do not want more of is more mediocre discoveries or mediocre drugs. We want to maintain that level of excellence.  
对患者而言,我们不希望出现更多平庸的发现或平庸的药物。我们要保持卓越的水准。  
**[01:34:16] Speaker B:** This in my mind is probably the biggest thing I worry about. There are all sorts of things that you worry about, malfeasance and the like, and those are very real and I don't want to downplay them. But in my mind, my worry is that employees and society in general goes on autopilot, because when you have tools that can create content, they can edit content, summarize, take action on your behalf, it's very easy to phone it in and effectively say, well, you know what, I'm going to stop checking my homework or interrogating the output. And in our sector, that's existential.  
在我看来,这可能是我最担心的事情。当然还有各种其他值得担心的问题,比如恶意行为之类的,这些都是非常现实的威胁,我不想低估它们。但在我看来,我真正担心的是员工和整个社会进入自动驾驶模式,因为当你拥有能够创建内容、编辑内容、总结内容、代你采取行动的工具时,敷衍了事就变得非常容易,实际上就是在说:「好吧,我不再检查我的作业,也不再质疑输出结果了。」而在我们这个行业,这是生死攸关的问题。  
**[01:34:52] Speaker B:** I was very fortunate when I was at Genentech and started my career. The person who hired me said, we hire the best and the brightest because the problems that we have to deal with in our industry are the most complex problems that we have to deal with. You've heard that in spades on this discussion. And the moment we stop bringing our intelligence and our creativity and our discipline and the approach that we take towards science to the table, that's a big problem.  
我很幸运,在 Genentech 开始我的职业生涯时,雇用我的那个人说:「我们雇用最优秀、最聪明的人,因为我们行业需要处理的问题是最复杂的问题。」你在这次讨论中已经深刻体会到这一点了。而一旦我们不再把我们的智慧、创造力、严谨态度以及对待科学的方法带到工作中,那就是一个大问题。  
**[01:35:15] Speaker B:** And so, you know, I think you don't solve that with governance or guardrails. You solve that with a culture of how you engage this technology. That's much harder to do, and you got to be super thoughtful about how you approach it.  
所以,我认为你不能用治理机制或护栏来解决这个问题。你要用一种与这项技术互动的文化来解决它。这要困难得多,而且你必须非常慎重地思考如何应对。  
**[01:35:32] Speaker A:** I will take a slight — I worry about mediocrity a lot, but in a slightly  
我想稍微补充一下——我也非常担心平庸化的问题,但角度稍有不同  
**[01:35:36] Speaker A:** Different style than Chris's, although I agree with everything he said.  
我的看法跟 Chris 的风格不太一样,尽管我同意他说的所有观点。  
**[01:35:41] Speaker A:** I worry about the narcissistic version of the model. So like our narcissism, meaning the model reflects back to us ourselves.  
我担心的是模型的自恋版本。也就是说,一种我们自己的自恋——模型把我们自己反射回给我们。  
**[01:35:50] Speaker B:** Uh-huh.  
嗯哼。  
**[01:35:51] Speaker A:** And it makes us more and more embroiled in our previous collective scientific self and makes it harder and harder in a sense to break free.  
它让我们越来越深地陷入我们过去的集体科学认知中,从某种意义上说,让我们越来越难以挣脱出来。  
**[01:36:01] Speaker A:** So people do all the right things in my nightmare. They still check their homework, they still read everything that's written there, but that almost infects their brain and they become even more and more ingrained in that.  
所以在我的噩梦场景里,人们做了所有正确的事情。他们仍然会检查自己的作业,仍然会阅读那里写的所有内容,但这几乎感染了他们的大脑,让他们变得越来越根深蒂固地沉浸其中。  
**[01:36:17] Speaker A:** And the reason I worry about that is that this is something that is actually not new, is something that happens in scientific communities all the time.  
我之所以担心这一点,是因为这其实并不是什么新鲜事,这种情况在科学界一直都在发生。  
**[01:36:23] Speaker A:** And you see that happen to fields that, as we call them, are well-established. Everyone talks in this way and gets kind of embroiled in their own field.  
你会看到这种情况发生在那些我们所说的成熟领域里。每个人都用这种方式交流,然后就陷入了自己领域的思维定式中。  
**[01:36:31] Speaker A:** It happens to large and illustrious companies over their many, many decades and more than decades, centuries of existence. They have like their way of doing things.  
这也发生在大型知名公司身上,在它们存在的几十年、甚至数百年的历史中,它们形成了自己的做事方式。  
**[01:36:41] Speaker A:** And then some other upstart comes and disrupts the whole thing because they're able to break free from that.  
然后就会有其他新兴势力出现,颠覆整个格局,因为他们能够摆脱那种束缚。  
**[01:36:45] Speaker A:** So there is a risk that the model kind of takes over our thinking, emulates it beyond perfection, has the broader, bigger brain because of this, and everyone gets sucked into that world and is unable to break free.  
所以有一个风险是,模型在某种程度上接管了我们的思维,把它模拟得比完美还完美,因此拥有了更广阔、更强大的大脑,然后每个人都被吸进那个世界,无法挣脱。  
**[01:37:00] Speaker A:** And I think there's ways to mitigate for that. You can mitigate for that with the modeling itself, but we have to think about it, we have to be...  
我认为有办法可以缓解这个问题。你可以在建模本身上做些改进来缓解,但我们必须思考这个问题,我们必须...  
**[01:37:09] Speaker A:** We have to be extraordinarily careful for the evals not to make that happen to our models, and it is not a trivial problem to solve.  
我们必须格外小心,不能让评估系统对我们的模型造成这种影响,这不是一个简单的问题。  
**[01:37:16] Speaker A:** There's real examples now of people getting excited about, you know, you can get a quad to beat any other approach, but a lot of it is actually how it's evaluated.  
现在有真实的例子,人们很兴奋,觉得可以让某个模型击败任何其他方法,但实际上很大程度上取决于如何评估。  
**[01:37:24] Speaker A:** And those evals are actually where it gets an edge, but they're not the right evals for novelty.  
那些评估指标确实让它获得了优势,但它们并不是衡量创新性的正确指标。  
**[01:37:30] Speaker A:** And then you can also mitigate for that, especially in our field, by using the real world, which is the lab, to let you kind of escape free from where you're already at.  
你也可以缓解这个问题,特别是在我们这个领域,通过使用真实世界——也就是实验室——让你能够跳出现有的框架。  
**[01:37:39] Speaker A:** And unless we do this, I actually would be very worried that we'll keep kind of regurgitating the same thing again and again and again and again.  
如果我们不这样做,我真的会很担心我们会不断地重复同样的东西,一遍又一遍。  
**[01:37:49] Speaker B:** Instilling this discernment.  
培养这种辨别力。  
**[01:37:51] Speaker A:** And that's narcissism. It's like watching yourself like Narcissus in the pond and liking it so much and just getting fixated on that.  
那就是自恋。就像 Narcissus 在池塘里看着自己的倒影,太喜欢了,然后就执迷于此。  
**[01:37:58] Speaker A:** So, you know, the ancient Greeks already told us everything.  
所以你看,古希腊人早就告诉我们一切了。  
**[01:38:00] Speaker A:** Maybe the only thing I'd add is from a public policy standpoint.  
也许我唯一要补充的是从公共政策的角度来看。  
**[01:38:05] Speaker A:** I think clearly we need to both have regulation, and I think Anthropic is a leader on this, and Dario's written, I think, eloquently recently about the need for regulation on AI systems to mitigate a lot of the threats.  
我认为显然我们需要监管,而且我觉得 Anthropic 在这方面是领导者,Dario 最近写得很有说服力,阐述了对 AI 系统进行监管以缓解许多威胁的必要性。  
**[01:38:22] Speaker A:** And at the same time, I think we're going to have to modernize our drug regulatory system to fully leverage these technologies.  
与此同时,我认为我们必须现代化我们的药物监管系统,才能充分利用这些技术。  
**[01:38:28] Speaker A:** If we actually can better predict preclinical safety, maybe we can do more streamlined animal models. If we can better model...  
如果我们真的能更好地预测临床前安全性,也许我们可以使用更精简的动物模型。如果我们能更好地建模……  
**[01:38:37] Speaker A:** Control arms and control groups, maybe we can truly do synthetic control arms, which has always been a challenge with acceptance in specific instances with the FDA. Maybe we can better model dosing and not have to do as much dose range finding. I mean, all of these things are possible, but we also need to bring the drug regulator along with us to get the full benefits of speed and probability of success we've been talking about.  
对照组和对照臂,也许我们可以真正实现合成对照臂,这在 FDA 的特定案例审批中一直是个挑战。也许我们可以更好地建模给药方案,不必做那么多剂量范围探索。我的意思是,所有这些都是可能的,但我们也需要让药品监管机构与我们同行,才能获得我们一直在讨论的速度和成功概率方面的全部收益。  
**[01:39:00] Speaker B:** Okay. So, we're going to end with a final question for everyone. It's the same question that Elon and Dario were asked. So what, when we come back next year and hopefully we're all sitting here or catching up, what is the one thing that you want to be held to and also one thing that you hope to be proven wrong about? Moss, why don't you start this?  
好的。那么,我们将以一个最后的问题来结束今天的讨论。这是同样问过 Elon 和 Dario 的问题。当我们明年再见面,希望我们都能坐在这里或者碰面叙旧时,你希望被问责兑现的一件事是什么?以及你希望被证明是错误的一件事是什么? Moss,你先来回答吧?  
**[01:39:21] Speaker A:** Yeah, I think to be held to that we can actually demonstrate that these technologies are starting to have an impact on the speed of our drug R&D process, because I think we've made a lot of bold proclamations and now I think we need to actually show for patients that we are delivering real results. And I hope I'm proven wrong that it's not going to take a crisis to actually get good regulation to happen, because I think it would be a shame that a crisis is what pushes us to get appropriate AI regulation in place.  
是的,我希望被问责的是,我们能够实际证明这些技术正在开始对我们药物研发流程的速度产生影响,因为我认为我们已经做了很多大胆的宣言,现在我们需要真正向患者展示我们正在交付实实在在的成果。而我希望被证明是错的是,不需要一场危机才能实现良好的监管,因为我认为如果要靠危机来推动我们建立适当的 AI 监管,那将是一件遗憾的事。  
**[01:39:54] Speaker B:** Um,  
嗯,  
**[01:39:56] Speaker C:** I think for many of the pieces, we already have had them in place now for a while, including, you know, real  
我认为对于很多方面,我们其实已经部署实施了一段时间了,包括,你知道,真实的  
**[01:40:02] Speaker A:** deployment and use. I want to see more of the end-to-end, like the model looks further down the road and the pieces are more connected together, rather than here's my solution for target discovery and here's my solution for a small molecule and for this step in the process and for that step in the process, for the other step.  
部署和使用。我希望看到更多端到端的方案,比如模型能够看得更长远,各个环节之间能够更紧密地连接在一起,而不是这里有一个靶点发现的解决方案,那里有一个小分子的解决方案,然后流程中这一步有个方案,那一步又有个方案,另一步还有个方案。  
**[01:40:18] Speaker A:** I think putting them together would be a good aspiration, and I hope I'll be proven wrong.  
我认为把它们整合到一起会是一个很好的目标,而且我希望自己的担心是错的。  
**[01:40:25] Speaker A:** I think it is reasonable to assume that people's stress levels are rising, and increasingly so, in companies, in other scientific communities like in academia, in society as a whole.  
我认为可以合理地假设,人们的压力水平正在上升,而且越来越严重,无论是在企业里,还是在其他科研群体比如学术界,还是在整个社会中。  
**[01:40:43] Speaker A:** I really hope, because I really want to be proven wrong on that rising stress level, that we'll start seeing people's joy in the process,  
我真心希望,因为我真的希望自己对压力上升的判断是错的,我希望我们能开始看到人们在这个过程中的喜悦,  
**[01:40:52] Speaker A:** their real enjoyment from the fact that something new and exciting is in their hands, that they can use a new thing in new ways, not just in old ways, and to see the arc of that positivity, because I think that's extremely important for getting a good outcome in the end.  
看到他们真正享受手中拥有新鲜而令人兴奋的东西这一事实,他们可以用新的方式使用新工具,而不只是用老办法,并且能看到这种积极性的发展轨迹,因为我认为这对最终获得好的结果极其重要。  
**[01:41:10] Speaker B:** From my standpoint, I just build on what Vos said. I think that there's a lot of hard work for us to put around how do we actually start to demonstrate the return on the significant investments that we're making.  
从我的角度来看,我只是在 Vos 所说的基础上补充一下。我认为我们有很多艰苦的工作要做,那就是我们如何真正开始展示我们所做的这些重大投资的回报。  
**[01:41:20] Speaker B:** Increasingly, investors are asking us not what you're investing in, or begging us to make more investments. Oh, there's some of that, but now they're saying, 「Okay, what's the return?」 And that's a tricky—that's  
投资者们越来越多地不是在问我们你们在投资什么,或者恳求我们增加投资。哦,这种情况也有一些,但现在他们在说:「好吧,回报是什么?」这是个棘手的——这是个  
**[01:41:32] Speaker A:** A tricky thing to get at. We had a month ago or so, one of our phase one programs went into phase two.  
这是个很难搞清楚的事情。大约一个月前,我们有一个一期项目进入了二期。  
**[01:41:37] Speaker A:** We started to see this kind of gnarly side effect that fortunately we had the data curated.  
我们开始看到一些很棘手的副作用,幸运的是我们有经过整理的数据。  
**[01:41:44] Speaker A:** We had AI tools sitting on top of it. And within a week, we were able to very clearly articulate a very small change that had been made in manufacturing that probably saved the program and certainly saved the program 6 to 12 months of time.  
我们在数据之上部署了AI工具。在一周之内,我们就能非常清楚地指出生产过程中做出的一个很小的改动,这个改动很可能挽救了整个项目,肯定为项目节省了6到12个月的时间。  
**[01:41:59] Speaker A:** How do you define the return on that investment? That's tricky, but we got to figure it out.  
你如何定义这项投资的回报呢?这很难说,但我们必须弄清楚。  
**[01:42:04] Speaker A:** And if we don't figure it out, the thing I worry about and I hope I'm proven wrong on is at some point organizations, certainly large organizations, are going to become put under pressure to ration the use of this technology.  
如果我们搞不清楚这一点,我担心的是——我希望我是错的——在某个时候,组织,尤其是大型组织,将会面临压力,不得不限量使用这项技术。  
**[01:42:16] Speaker A:** And in fact, one of the reasons we did the Anthropic partnership was we've now embedded Anthropic across all of our work processes and we hopefully will be able to now apply the right tools to the right problems so that we're not using the most advanced tools to summarize people's emails.  
事实上,我们与Anthropic合作的原因之一就是,我们现在已经将Anthropic嵌入到所有工作流程中,我们希望能够针对不同的问题应用合适的工具,这样就不会用最先进的工具去做总结邮件这种事情了。  
**[01:42:34] Speaker A:** Because that's not the way to get a good return on this investment.  
因为那不是获得良好投资回报的方式。  
**[01:42:38] Speaker A:** And I think we've got to get more sophisticated around that concept generally.  
我认为总体上我们需要在这个概念上变得更加精细化。  
**[01:42:43] Speaker B:** So as a scientist, I always feel compelled and it's most critical to be both rigorous and really honest with these new technologies and to think about how we're deploying them.  
作为一名科学家,我总是觉得有必要,而且最关键的是要对这些新技术既严谨又真正诚实,并思考我们如何部署它们。  
**[01:42:52] Speaker B:** When I ran a research group, I used to always remind the group that the only bad experiment was the one that we didn't...  
当我管理一个研究小组时,我过去常常提醒团队,唯一糟糕的实验是我们没有...  
**[01:42:58] Speaker A:** Carefully analyze, and I'm really grateful to the panel for the time because I think we heard how AI and Claude is touching all parts of science and drug development as well as the business.  
仔细分析一下,我非常感谢各位嘉宾抽出时间参与讨论,因为我们了解到 AI 和 Claude 是如何触及科学和药物开发的各个领域以及商业层面的。  
**[01:43:09] Speaker A:** And also that there's still plenty of room to improve, right? We're at early stages and the impacts are real, but the upside is still ahead of us, and so very grateful for the time.  
同时我们也看到仍有很大的改进空间,对吧?我们还处于早期阶段,虽然影响是真实存在的,但更大的收益还在前方,所以非常感谢大家的时间。  
**[01:43:18] Speaker A:** A little bit of a performance review for Claude. So, you know, we'll make sure we keep working, and finally thank all the audience again for that discussion and partnership, and also the work going forward both on model capabilities but also doing these experiments and how we get them to patients.  
这也算是对 Claude 的一次绩效评估吧。所以,我们会确保继续努力,最后再次感谢所有观众的讨论和合作,以及未来在模型能力方面的工作,还有进行这些实验以及如何将成果惠及患者。  
**[01:43:32] Speaker A:** Thank you all.  
谢谢大家。  
**[01:43:34] Speaker B:** Great. Thank you.  
很好。谢谢。  
**[01:43:42] Speaker B:** Please welcome back to the stage Zoubin Ghahramani.  
请欢迎 Zoubin Ghahramani 重返舞台。  
**[01:43:55] Speaker C:** All right, bringing us home. Six months ago, we told you that Claude could help with the digital work of life sciences R&D. This morning, I said something different: that Claude could run that work.  
好的,进入收尾环节。六个月前,我们告诉大家 Claude 可以辅助生命科学研发的数字化工作。今天早上,我说的是不同的事情:Claude 可以运行这些工作。  
**[01:44:08] Speaker C:** We told you that the morning ahead would include us laying out the case for it. And here's how we spent it.  
我们说过今天上午会为这一论断提供论证。以下是我们的安排。  
**[01:44:17] Speaker C:** Eric and D— sorry, Lattice and Dario walked us through their unvarnished views on how compression may or may not happen, the challenges and opportunity ahead.  
Eric 和 D——抱歉,是 Lattice 和 Dario 向我们坦率地阐述了他们对压缩可能发生或不发生的看法,以及前方的挑战和机遇。  
**[01:44:25] Speaker C:** Eric and Alec walked you through the basis of our claim: models that meaningfully improve in biology with every release, and a workbench that runs the analysis your scientists actually need.  
Eric 和 Alec 向大家展示了我们论断的基础:在生物学领域随着每次发布都有显著改进的模型,以及一个能运行科学家实际所需分析的工作平台。  
**[01:44:38] Speaker A:** Run pipelines, figures, every step reproducible and traceable. And Voss, Aviv, and Chris just walked us through how they're charting the path of AI transformation at BMS, Genentech, and Artis.  
运行流程、图表,每一步都可重现、可追溯。Voss、Aviv 和 Chris 刚刚向我们展示了他们如何在 BMS、Genentech 和 Artis 规划 AI 转型的路径。  
**[01:44:54] Speaker A:** At the end of the day, this is just the beginning. To build an intelligence platform where scientific discovery truly happens is going to be a collaboration with the industry and our partners represented here.  
归根结底,这只是一个开始。要构建一个真正能够实现科学发现的智能平台,需要与在座的行业伙伴们共同协作。  
**[01:45:07] Speaker A:** So we'd love to invite you into this journey, and there are a couple of ways to start.  
因此我们诚邀各位加入这一旅程,这里有几种参与方式。  
**[01:45:16] Speaker A:** Number one, as far as Claude Science goes, try it. The QR code on this screen will give you access to it and put it right in your hands.  
第一,关于 Claude Science,请亲自试用。屏幕上的二维码可以让您直接访问并上手体验。  
**[01:45:23] Speaker A:** And secondly, join us for a hackathon with the Gladstone Institutes from July 6th to the 12th, our first ever that'll feature Claude Science.  
第二,请加入我们与 Gladstone Institutes 合作举办的黑客马拉松,时间是 7 月 6 日至 12 日,这是我们首次以 Claude Science 为主题的活动。  
**[01:45:36] Speaker A:** In the end, this is about time and the question, giving your scientists back what they trained for and you deciding which questions they'll spend it on.  
最终,这关乎时间与问题本身——把科学家们接受训练时追求的那些东西还给他们,由您来决定让他们把时间花在哪些问题上。  
**[01:45:46] Speaker A:** We'd like to invite you to ask the questions as well. The demos are out that door. So pick the question, give it the work, and see how far it runs. To everyone who joined us here today, both in person and virtually, thank you.  
我们也希望邀请各位提出问题。演示区就在那扇门外。所以请选择您的问题,交给它去执行,看看它能走多远。感谢今天到场以及线上参与的所有人,谢谢大家。  
**[01:46:24] Speaker B:** Hey hey hey hey hey hey hey.  
嘿嘿嘿嘿嘿嘿嘿。  

---

## Deep Dive Summary

### Topic 1: Opening and Introduction to Claude's New Capability Claim
开场及 Claude 新能力宣言
_[00:00]_

**Q:** What is Anthropic announcing about Claude's capabilities in life sciences R&D?
**问：** Anthropic 宣布 Claude 在生命科学研发领域有什么新能力?

**A:** Zubair Jandali, head of go-to-market for healthcare and life sciences at Anthropic, presents a bold evolution in Claude's role in scientific research. Six months prior, Anthropic claimed Claude could "help with" life sciences R&D work. Today's announcement escalates that claim dramatically: Claude can now "run the work" - not merely assist or accelerate it, but autonomously execute it. Jandali frames this as the most exciting moment since Claude's initial release, suggesting a fundamental shift in how AI interfaces with scientific workflows. The claim is positioned as deliberately provocative ("a bold thing to say out loud"), signaling Anthropic's confidence in a qualitative leap in capability.
**答：** Anthropic 医疗与生命科学商业负责人 Zubair Jandali 宣布了 Claude 在科研角色上的重大跃升。六个月前,Anthropic 声称 Claude 能够"帮助"生命科学研发工作。今天的发布会将这一主张大幅升级:Claude 现在可以"运行工作"——不仅仅是辅助或加速,而是自主执行。Jandali 称这是 Claude 发布以来最激动人心的时刻,暗示 AI 与科学工作流的交互方式发生了根本性转变。这一宣言被刻意定位为挑战性的("大胆地说出来"),显示出 Anthropic 对能力质变的信心。

### Topic 2: The Software Development Loop Analogy
软件开发循环类比
_[02:21]_

**Q:** How does the evolution of AI in software development provide a model for scientific research?
**问：** AI 在软件开发中的演进如何为科学研究提供模型?

**A:** The speaker establishes software development as the precedent for AI's transformative potential in research. Coding has "irreversibly changed" into a continuous loop: "write, run, fix, run again." Once AI could operate within this feedback cycle, the autonomous operation time expanded dramatically - from minutes two years ago, to hours today, with days anticipated soon. This progression demonstrates AI's capacity to maintain momentum within iterative processes without constant human intervention. The analogy sets up the parallel to scientific research, suggesting that if AI can sustain software development loops for extended periods, a similar capability should emerge in the experimental cycle.
**答：** 演讲者将软件开发确立为 AI 变革潜力的先例。编程已经"不可逆转地改变"成了一个持续循环:"编写、运行、修复、再运行"。一旦 AI 能够在这个反馈循环中运作,自主操作时长就急剧扩展——两年前是几分钟,今天是几小时,很快将达到几天。这一进展展示了 AI 在迭代过程中保持动力而无需持续人工干预的能力。这个类比为科学研究奠定了平行基础,暗示如果 AI 能够长时间维持软件开发循环,类似的能力应该会出现在实验周期中。

### Topic 3: The Scientific Method as a Loop and the Analysis Bottleneck
作为循环的科学方法与分析瓶颈
_[02:53]_

**Q:** What is the fundamental bottleneck in the scientific research cycle that AI can address?
**问：** AI 可以解决科学研究周期中的哪个根本瓶颈?

**A:** The scientific method is reframed as "the original loop": design experiment, run it, analyze data, ask the next question. The critical bottleneck occurs not at the bench during experimentation, but "at a keyboard" during analysis. The speaker identifies a familiar pain point for R&D professionals: "the experiment that takes three days to run but three or more weeks to analyze." These analytical weeks represent "toil" rather than actual science - work scientists must "endure to get to the science." The claim is that this toil is "collapsing," which will return scientists to "more of what they trained for: time at the question." This positions AI not as replacing scientific thinking, but as eliminating the mechanical overhead that distances scientists from inquiry itself.
**答：** 科学方法被重新定义为"最初的循环":设计实验、运行实验、分析数据、提出下一个问题。关键瓶颈不在实验台上的实验阶段,而在分析阶段的"键盘前"。演讲者指出了研发专业人士熟悉的痛点:"实验运行三天,但分析需要三周甚至更久"。这些分析周构成的是"苦工"而非真正的科学——科学家为了"到达科学"而必须"忍受"的工作。主张是这种苦工正在"崩塌",这将让科学家回归"他们接受训练时的初衷:思考问题的时间"。这将 AI 定位为不是取代科学思维,而是消除让科学家远离探究本身的机械性开销。

### Topic 4: Event Structure and the 50-100 Year Compression Thesis
活动结构与50-100年压缩论题
_[03:42]_

**Q:** How is the briefing structured and what is the overarching thesis about AI's impact on biomedical progress?
**问：** 发布会如何组织,关于 AI 对生物医学进步影响的总体论题是什么?

**A:** The program is structured around demonstrating AI's transformative impact through three segments: a conversation between Dario Amodei and a GLP-1 medicine scientist about "compressing timelines in biology," a demonstration of the new technology "in the hands of a scientist," and insights from three industry leaders on organizational transformation. The event anchors itself to Dario's essay from two years prior, which claimed that "AI-enabled biology and medicine could compress 50 to 100 years of progress into 5 to 10." What was initially received as "ambition" is now being presented as empirical reality: "it has started." The framing emphasizes moving from theoretical potential to observable transformation.
**答：** 发布会围绕展示 AI 变革性影响设计了三个环节:Dario Amodei 与 GLP-1 药物科学家关于"压缩生物学时间线"的对话、新技术"在科学家手中"的演示,以及三位行业领袖关于组织转型的见解。活动锚定在 Dario 两年前的文章上,该文章声称"AI 赋能的生物学和医学可以将50到100年的进展压缩到5到10年"。最初被视为"雄心"的论述现在被呈现为经验现实:"已经开始了"。框架强调从理论潜力转向可观察的转型。

### Topic 5: Introduction to Lotte Bjerre Knudsen and the GLP-1 Revolution
Lotte Bjerre Knudsen 介绍与 GLP-1 革命
_[04:29]_

**Q:** Who is Lotte Bjerre Knudsen and what makes her perspective on medical revolution relevant to AI's potential?
**问：** Lotte Bjerre Knudsen 是谁,她对医学革命的视角为何与 AI 潜力相关?

**A:** Matt Herper, senior writer at Stat, introduces the conversation's participants as "revolutionaries" discussing a potential revolution. Lotte Knudsen is positioned as a "visionary" behind GLP-1 medicines, who saw their potential impact not only in diabetes but also in obesity - a field she notably "joined by accident." Her work sparked "one of the biggest medical revolutions" Herper has witnessed in 25 years of medical journalism. Knudsen's credentials include the Lasker Prize, Breakthrough Prize, and Stat Biomedical Innovation Award. Dario Amodei is introduced with dual expertise: foundational AI research and earlier work in "biophysics of electrophysiology of neural circuits." Critically, Dario has stated that AI's impact on biomedicine represents "the most important thing that AI can do," establishing the high-stakes framing for the conversation.
**答：** Stat 资深作者 Matt Herper 将对话参与者介绍为讨论潜在革命的"革命者"。Lotte Knudsen 被定位为 GLP-1 药物背后的"远见者",她看到了这些药物不仅在糖尿病而且在肥胖症方面的潜在影响——值得注意的是她"意外加入"了肥胖症领域。她的工作引发了 Herper 在25年医学新闻报道生涯中见证的"最大医学革命之一"。Knudsen 的荣誉包括 Lasker 奖、Breakthrough 奖和 Stat 生物医学创新奖。Dario Amodei 以双重专长被介绍:基础 AI 研究和早期在"神经回路的电生理生物物理学"方面的工作。关键是,Dario 曾表示 AI 对生物医学的影响代表"AI 能做的最重要的事情",为对话建立了高风险框架。

### Topic 6: Lotte's Perspective on the AI Inflection Point in Medicine
Lotte 对医学中 AI 拐点的看法
_[06:53]_

**Q:** Does Lotte Knudsen believe we are at a transformative moment in biomedical AI, and how does she distinguish real breakthroughs from hype?
**问：** Lotte Knudsen 是否认为我们正处于生物医学 AI 的变革时刻,她如何区分真正的突破与炒作?

**A:** Knudsen affirms that "we are at a real inflection point" in biomedical AI, driven by the explosion of available data and daily sharing of new approaches to "compress" progress timelines. However, she brings critical perspective from her experience with GLP-1 development, having witnessed both "inflection points as well as hype." She draws a crucial distinction: the "real goal" with semaglutide was creating "health impact" through reduced cardiovascular deaths, while the "hype was the weight loss that really got more people to actually use the medicines." This reveals her framework for evaluating transformative claims - true inflection points deliver measurable health outcomes, not just attention or adoption. The implication is that AI's value in medicine will ultimately be judged by health impact, not technical capability alone.
**答：** Knudsen 确认"我们正处于一个真正的拐点",这是由可用数据的爆炸式增长和每天分享的"压缩"进展时间线的新方法驱动的。然而,她从 GLP-1 开发经验中带来了批判性视角,她既见证过"拐点也见证过炒作"。她划出了关键区别:semaglutide 的"真正目标"是通过减少心血管死亡来创造"健康影响",而"炒作是减重,这真正让更多人实际使用药物"。这揭示了她评估变革性主张的框架——真正的拐点提供可测量的健康结果,而不仅仅是关注或采用。潜在含义是,AI 在医学中的价值最终将由健康影响而非仅技术能力来评判。

### Topic 7: The 50-100 Year Compression Timeline and Evidence Timeline
50-100年压缩时间线与证据时间线
_[08:10]_

**Q:** How does Dario defend the ambitious claim of compressing decades of progress into years, and what evidence timeline does he propose?
**问：** Dario 如何为将数十年进展压缩到数年的雄心主张辩护,他提出了什么证据时间线?

**A:** When asked about the "50 to 100 years of progress in 5 to 10" prediction and what evidence can emerge in the next 6-12 months, Dario begins to articulate his framework. He makes an important temporal distinction that he references from his original essay: the claim is not that this acceleration is happening uniformly now, but that "10 years from now, I think we may be making progress at a rate of 10 years per year." This suggests an exponential or compound acceleration model rather than immediate 10x progress. The response implies that near-term evidence (6-12 months) will show early indicators of this trajectory rather than the full realization of decade-per-year progress rates. The setup suggests Dario is building toward explaining how current capabilities connect to future acceleration.
**答：** 当被问及"5到10年内取得50到100年的进展"预测以及未来6-12个月能出现什么证据时,Dario 开始阐述他的框架。他做出了一个重要的时间区分,引用了他原始文章中的内容:这一主张并不是说加速现在正在均匀发生,而是"10年后,我认为我们可能会以每年10年的速度取得进展"。这暗示了指数或复合加速模型,而非立即的10倍进展。回应暗示近期证据(6-12个月)将显示这一轨迹的早期指标,而非每年十年进展率的完全实现。铺垫暗示 Dario 正在构建如何将当前能力与未来加速联系起来的解释。

### Topic 8: Current Limitations on AI's Impact in Biology
AI 在生物学领域影响的当前局限性
_[09:00]_

**Q:** Why can't AI accelerate biological progress at a rate of 10 years per year today?
**问：** 为什么 AI 现在还不能以每年十年的速度加速生物学进步?

**A:** Speaker A identifies two major bottlenecks preventing AI from achieving "10 years of progress every year" in biology currently. The first is technological: AI models are rapidly improving on an exponential curve and getting better at tasks from "computational biology to thinking about proteins and DNA," but they still have "some ways to go on the exponential." The second bottleneck is systemic inertia, encompassing both the learning curve for scientists adopting these tools in academic research, target discovery, and clinical trials, and the much slower adaptation of regulatory systems. Speaker A estimates it will take "a decade" for the full system, much of which doesn't yet believe in AI's revolutionary potential, to learn to operate with these new capabilities.
**答：** Speaker A 指出了两个主要瓶颈阻碍 AI 在生物学领域实现"每年十年进步"的目标。第一个是技术瓶颈:AI 模型正在快速指数级改进,在"计算生物学到蛋白质和 DNA 思考"等任务上越来越好,但在指数曲线上"还有一段路要走"。第二个瓶颈是系统惯性,包括科学家在学术研究、靶点发现和临床试验中采用这些工具的学习曲线,以及监管系统更缓慢的适应过程。Speaker A 估计整个系统需要"十年时间"才能学会使用这些新能力,而且系统中很多部分甚至还不相信 AI 会带来革命性变化。

### Topic 9: AI Tools Reducing Drug Discovery Cycle Time
AI工具缩短药物发现周期时间
_[15:46]_

**Q:** How do AI tools help reduce the cycle time in drug discovery?
**问：** AI工具如何帮助缩短药物发现的周期时间？

**A:** Speaker B and A explain that AI tools lower cycle time by transforming drug discovery "from an art into a science," enabling better optimization, measurement, and target identification. They note this acceleration builds on gradual progress over previous decades. Multiple parts of the lengthy process can be compressed: clinical trials become shorter when drugs work more reliably and have stronger effects, requiring fewer patient recruits. While the timeline might only compress from "five to ten years to five years or three years," cycle time remains the controlling constraint, necessitating extensive parallel work.
**答：** Speaker B和A解释说，AI工具通过将药物发现"从艺术转变为科学"来降低周期时间，实现更好的优化、测量能力和靶点识别。他们指出这种加速是建立在过去几十年的渐进进展之上的。漫长流程的多个环节都可以压缩：当药物更可靠地发挥作用并产生更强效果时，临床试验会缩短，需要招募的患者更少。虽然时间线可能只能从"5到10年压缩到5年或3年"，但周期时间仍然是控制性约束，需要大量并行工作。

### Topic 10: Failure Rate as the Biggest Drug Discovery Challenge
失败率是药物发现的最大挑战
_[16:51]_

**Q:** What is the biggest issue in drug discovery that AI needs to address?
**问：** AI需要解决的药物发现中的最大问题是什么？

**A:** Speaker A emphasizes that while cycle time reduction helps significantly, "the biggest issue is the failure rate." This connects to the broader challenge referenced in the 'Machines of Loving Grace' piece. The discussion contrasts the technology industry's Moore's Law (exponential improvement) with the drug industry's Eroom's Law (Moore's Law backward), where costs increase despite technological advances. The paradox is highlighted: genome sequencing costs dropped "from three billion to $300," yet drug development spending continues to rise, raising the question of why AI might succeed where previous technologies failed.
**答：** Speaker A强调，虽然周期时间缩短有很大帮助，但"最大的问题是失败率"。这与'Machines of Loving Grace'文章中提到的更广泛挑战相关。讨论对比了科技行业的Moore定律（指数级改进）与制药行业的Eroom定律（Moore定律的倒退），即尽管技术进步，成本仍在增加。这个悖论很明显：基因组测序成本从"30亿美元降到300美元"，但药物开发支出持续上升，引发了为什么AI可能在以往技术失败的地方成功的问题。

### Topic 11: Why AI Differs from Previous Technologies: The Andy Grove Fallacy
AI为何不同于以往技术：Andy Grove谬误
_[17:36]_

**Q:** Why is AI different from previous technologies that failed to transform drug discovery, and what was the Andy Grove fallacy?
**问：** 为什么AI与以往未能转变药物发现的技术不同，Andy Grove谬误是什么？

**A:** Speaker B references Derek Lowe's concept of the "Andy Grove fallacy"—the mistaken belief that biology can be treated as a rational engineering system when it's actually "a super messy evolved system." Previous engineering technologies failed because they tried to organize data or unblock single process steps while confronting systemic complexity. Speaker B argues AI is fundamentally different as "a general purpose technology that helps us to make sense of that complexity in its full complexity better." Humans have historically made progress using their brains to comprehend complexity, and AI extends this capability. While success isn't guaranteed, Speaker B sees "the beginnings of it" already emerging and expresses optimism tempered by uncertainty.
**答：** Speaker B引用了Derek Lowe的"Andy Grove谬误"概念——错误地认为生物学可以被当作理性的工程系统，而实际上它是"一个超级混乱的进化系统"。以往的工程技术失败是因为它们试图组织数据或解除单个流程步骤的阻塞，同时面对系统性复杂性。Speaker B认为AI根本不同，是"一种通用技术，帮助我们更好地理解其全部复杂性"。人类历来通过大脑理解复杂性来取得进展，AI扩展了这种能力。虽然成功不能保证，Speaker B看到"它的开端"已经出现，并表达了谨慎的乐观。

### Topic 12: Improving Probability of Success Through Better Understanding
通过更好理解提高成功概率
_[19:00]_

**Q:** How will AI improve the success rate of drug development beyond just cycle time?
**问：** AI如何在周期时间之外提高药物开发的成功率？

**A:** Speaker C explains that AI will "improve the probability of success of new medicines because we'll understand them better," establishing stronger foundations for target selection, understanding mechanisms of action (MOA), and designing smarter clinical trials. This leads to fewer failures, though Speaker C acknowledges Derek Lowe's ongoing skepticism will persist "for another 10 years" until concrete examples emerge. Critics will argue drugs aren't "fully AI designed," but Speaker C emphasizes the real value is incremental improvement across "so many things you need to know" for choosing and developing medicines. A key insight from GLP-1 drugs is that "pleiotropic effects"—benefits across multiple organs—are valuable, yet the field still overemphasizes narrow targeting based on genetics or mouse models rather than seeking "broad signals."
**答：** Speaker C解释说，AI将"提高新药的成功概率，因为我们会更好地理解它们"，为靶点选择、理解作用机制（MOA）和设计更智能的临床试验建立更强的基础。这会导致更少的失败，尽管Speaker C承认Derek Lowe的持续怀疑将持续"再10年"，直到具体案例出现。批评者会争辩药物不是"完全由AI设计"，但Speaker C强调真正的价值是在"选择和开发药物需要知道的这么多事情"中的渐进改进。GLP-1药物的关键洞察是"多效性"——对多个器官的益处——是有价值的，但该领域仍过分强调基于遗传学或小鼠模型的狭窄靶向，而不是寻求"广泛信号"。

### Topic 13: Biosecurity Risks and Multi-Layered Safety Approach
生物安全风险与多层安全方法
_[20:29]_

**Q:** What are the biosecurity risks of AI in drug discovery, and how should they be managed?
**问：** AI在药物发现中的生物安全风险是什么，应该如何管理？

**A:** Speaker B raises concerns about AI enabling bioweapons or synthetic organisms, prompting Speaker A to discuss Anthropic's approach to AI risks across domains. They note the company is currently addressing cyber risks as that capability window has opened, while "biology has not yet turned on." Drawing lessons from cyber security—where models can both find and patch exploits—Speaker A acknowledges biology lacks the same symmetry, as COVID-19 demonstrated. The challenge involves distinguishing helpful from dangerous queries through model safeguards. Speaker A proposes a "multi-layer cake" approach: safeguards on general models, trusted access programs for entities like pharmaceutical companies that already handle dangerous biological materials, and piggybacking on existing protocols. The goal is capturing "100% of the benefits while blocking as close as we can to 100% of the real counterfactual actual dangerous harms, which is a very narrow slice."
**答：** Speaker B提出了AI使生物武器或合成生物体成为可能的担忧，促使Speaker A讨论Anthropic对各领域AI风险的方法。他们指出公司目前正在解决网络安全风险，因为该能力窗口已经打开，而"生物学尚未开启"。从网络安全中吸取教训——模型既能发现也能修补漏洞——Speaker A承认生物学缺乏同样的对称性，正如COVID-19所展示的。挑战在于通过模型安全措施区分有用和危险的查询。Speaker A提出"多层蛋糕"方法：通用模型的安全措施、为已经处理危险生物材料的制药公司等实体提供的可信访问程序，以及利用现有协议。目标是获取"100%的益处，同时尽可能阻止100%真正的反事实实际危险危害，这是一个非常狭窄的部分"。

### Topic 14: Risk-Benefit Balancing and Independent Governance
风险收益平衡与独立治理
_[24:16]_

**Q:** How should AI companies balance risk versus benefit, and what governance structures support this?
**问：** AI公司应该如何平衡风险与收益，什么治理结构支持这一点？

**A:** Drawing from pharmaceutical industry experience, Speaker B emphasizes the importance of "balancing risk versus benefit" with strong focus and evaluation frameworks, suggesting independent oversight similar to data monitoring committees in clinical trials—specifically, oversight not "motivated by money." Speaker A describes Anthropic's Long-Term Benefit Trust (LTBT), which governs the entire company and appoints a majority of board seats. The trust includes leaders like the head of the Clinton Health Access Initiative and a former California Supreme Court justice, with plans to add experts in biology and medicine. The crucial rule: LTBT members "don't have any stock in the company." Beyond corporate governance, Speaker A suggests application-specific monitoring committees for domains like biology and cyber, either within companies or industry-wide, noting "the government probably also has a role to play."
**答：** 借鉴制药行业经验，Speaker B强调"平衡风险与收益"的重要性，需要强有力的关注和评估框架，建议类似临床试验中数据监测委员会的独立监督——特别是不"受金钱驱动"的监督。Speaker A描述了Anthropic的Long-Term Benefit Trust（LTBT），它管理整个公司并任命董事会多数席位。该信托包括Clinton Health Access Initiative负责人和前加州最高法院法官等领导者，计划增加生物学和医学专家。关键规则：LTBT成员"在公司中没有任何股份"。除了公司治理，Speaker A建议针对生物学和网络安全等领域的应用特定监测委员会，可以在公司内部或行业范围内，并指出"政府可能也有作用"。

### Topic 15: Addressing AI Hallucinations and Validation Concerns
解决AI幻觉和验证担忧
_[26:42]_

**Q:** How should pharma trust AI predictions when models hallucinate, and where is the validation data?
**问：** 当模型产生幻觉时制药公司应该如何信任AI预测，验证数据在哪里？

**A:** Speaker A addresses Claude's own skeptical question about hallucinations and validation by noting that hallucinations have improved significantly over time and occur less frequently now. Using self-driving cars as an analogy, Speaker A argues "we will never have an AI model that never hallucinates" due to the probabilistic nature of AI reasoning, which mirrors human cognition. This probabilistic approach creates an inherent "duality between creativity and hallucination"—to be creative requires "straddling the boundary between making things up and having good ideas." Like humans who are brilliant but sometimes wrong or hold misconceptions, AI models will improve their filters for distinguishing creativity from fabrication but never achieve perfection. Speaker A humorously references Nobel Prize winners with eccentric beliefs, noting Kary Mullis talked to "a glowing green raccoon" and that "the line is long" of laureates with questionable ideas like recommending vitamin C for everything, illustrating that even human experts aren't infallible.
**答：** Speaker A通过指出幻觉随时间显著改善且现在发生频率降低，来回答Claude自己关于幻觉和验证的怀疑问题。使用自动驾驶汽车作为类比，Speaker A认为"我们永远不会有一个永不产生幻觉的AI模型"，这是由于AI推理的概率性质，它反映了人类认知。这种概率方法创造了固有的"创造力和幻觉之间的二元性"——要有创造力需要"在编造和产生好想法之间走钢丝"。像聪明但有时错误或持有误解的人类一样，AI模型将改进区分创造力和虚构的过滤器，但永远无法达到完美。Speaker A幽默地引用了有古怪信念的诺贝尔奖获得者，指出Kary Mullis与"发光的绿色浣熊"交谈，"这个队伍很长"，有许多获奖者有可疑的想法，比如建议所有人服用维生素C，说明即使是人类专家也不是绝对可靠的。

### Topic 16: Creativity vs Hallucination Tradeoff in Discovery
创造力与幻觉之间的权衡
_[29:06]_

**Q:** How does the balance between creativity and hallucination manifest in scientific discovery?
**问：** 创造力和幻觉在科学发现中如何平衡?

**A:** The speakers use Kary Mullis as an illustrative example of how brilliant scientific breakthroughs can coexist with high rates of erroneous thinking. Mullis, who developed groundbreaking ideas, also had a "pretty high" hallucination rate, partly due to chemical substances. This example demonstrates the inherent tension between the creative thinking needed for innovation and the risk of generating incorrect or unreliable outputs—a dynamic relevant to both human cognition and AI systems.
**答：** 讲者以Kary Mullis为例说明了突破性科学发现如何与高错误率思维共存。Mullis提出了卓越的想法,但他的"幻觉率相当高",部分原因与化学物质有关。这个例子展示了创新所需的创造性思维与产生错误或不可靠输出风险之间的内在张力——这种动态既适用于人类认知,也适用于AI系统。

### Topic 17: Industry Skepticism: AI Doesn't Work for My Field
行业怀疑论:AI在我的领域不适用
_[29:30]_

**Q:** What is the most common skeptical question about AI in scientific fields?
**问：** 关于AI在科学领域应用最常见的质疑是什么?

**A:** The most frequent skeptical response the speaker encounters is "AI doesn't work for my field." The speaker, identifying as a scientist, addresses this by asserting that while "AI is not going to replace scientists," the real risk is that "scientists who are not using AI are going to get replaced." This formulation can be expanded to "every possible field and every organization." The speaker advocates for informed skepticism combined with active engagement, arguing that leaning into AI capabilities is essential for creating more solutions rather than being displaced by those who do.
**答：** 讲者遇到的最常见质疑是"AI在我的领域不适用"。作为科学家,讲者认为"AI不会取代科学家",但真正的风险是"不使用AI的科学家会被取代"。这一论断可以扩展到"每个领域和每个组织"。讲者倡导有根据的怀疑态度结合积极参与,认为深入利用AI能力对于创造更多解决方案至关重要,否则会被那些使用AI的人所取代。

### Topic 18: Productive AI Use: Avoiding Endless Debates with LLMs
高效使用AI:避免与LLM无休止争论
_[30:14]_

**Q:** How can researchers use AI productively without spending all day arguing with language models?
**问：** 研究人员如何高效使用AI而不至于整天与语言模型争论?

**A:** The speaker emphasizes maintaining "a really strong focus on what it actually is that you're trying to do" and being informed about different models and usage methods. Using Anthropic's terminology, the speaker notes that while many understand basic chat capabilities—"even my mother, who's 92, understands that"—most people stop there without grasping the distinctions between chat, code, and "Claude design." Understanding when to deploy agents, how to connect brain banks to agents for continuous learning, and maintaining goal focus are critical. The example of GLP-1 drugs illustrates this: focusing on "the health" benefits rather than getting "distracted by the hype of the weight loss" represents the kind of strategic clarity needed.
**答：** 讲者强调要"明确聚焦于实际试图完成的任务",并了解不同模型和使用方法。用Anthropic的术语说,虽然很多人理解基本聊天功能——"连我92岁的母亲都明白"——但大多数人止步于此,不理解chat、code和Claude design之间的区别。理解何时部署agents、如何将brain bank连接到agents以实现持续学习、保持目标聚焦都很关键。GLP-1药物的例子说明了这一点:专注于"健康"益处而不是"被减肥炒作分散注意力",代表了所需的战略清晰度。

### Topic 19: Hope: AI Success in Target Discovery
期望:AI在靶点发现中取得成功
_[31:42]_

**Q:** What positive development in AI for life sciences would you most like to see in the next year?
**问：** 你最希望在明年看到AI在生命科学领域的哪个积极进展?

**A:** Daria identifies target discovery as the critical bottleneck and expresses hope for AI breakthroughs in this area, noting that "the models are just knocking on the door" of being able to help significantly. She emphasizes the exponential nature of progress—"exponentials really catch you off guard"—and stresses the importance of scientists maintaining foresight to recognize rapid improvements. The key behavioral shift needed is for scientists to continually reassess AI capabilities rather than assuming current limitations are permanent: "the AI model I had three months ago couldn't help me at all with this but the one I just got today actually is helping me a lot." This requires attention and willingness to "keep pace with the technology" through regular re-evaluation.
**答：** Daria指出靶点发现是关键瓶颈,并希望AI在这一领域取得突破,她提到"模型正敲响大门",即将能够提供重大帮助。她强调进步的指数性质——"指数增长真的会让你措手不及"——并强调科学家保持前瞻性以识别快速改进的重要性。所需的关键行为转变是科学家持续重新评估AI能力,而不是假设当前局限是永久的:"三个月前的AI模型完全帮不上忙,但今天刚拿到的模型实际上帮了很大的忙。"这需要关注并愿意通过定期重新评估来"跟上技术步伐"。

### Topic 20: Fear: Reflexive Skepticism Delaying Benefits
担忧:本能怀疑延迟收益实现
_[33:04]_

**Q:** What negative pattern do you fear could slow AI adoption in life sciences?
**问：** 你担心什么负面模式可能减缓AI在生命科学领域的采用?

**A:** The speaker warns against "reflexive skepticism" based on observing AI's progression across multiple fields including Go, coding, mathematics, and writing. The consistent pattern is that models remain "useless, useless, useless" until they suddenly reach a threshold where they help ordinary practitioners, yet advanced experts continue dismissing them as only useful for "the median person" rather than field advancement. The exponential nature of improvement means this transition happens "so fast" that people "get set in their ways" after seeing "12 generations of AI models that don't help them at all," only to be caught off-guard when capabilities suddenly cross the utility threshold. The speaker cites recent examples like "Mythos and Cyber" and acknowledges biology is "not as easily verifiable," but fears that "scientists, pharmaceutical companies, the regulatory system" being "incredibly slow to recognize it" will delay benefits "years later than we would otherwise" receive them.
**答：** 讲者警告不要基于观察AI在多个领域(包括Go、编程、数学和写作)的进展而产生"本能怀疑"。一致的模式是模型保持"无用、无用、无用",直到突然达到帮助普通从业者的阈值,但高级专家继续将其视为只对"普通人"有用而非推动领域进步。指数级改进的本质意味着这种转变发生得"非常快",以至于人们在看到"12代完全帮不上忙的AI模型"后"固步自封",结果在能力突然跨越实用阈值时措手不及。讲者引用了"Mythos和Cyber"等近期例子,承认生物学"不那么容易验证",但担心"科学家、制药公司、监管系统""识别速度极慢"会将收益延迟"数年"。

### Topic 21: Fear: Loss of Global Scientific Collaboration
担忧:失去全球科学合作
_[35:00]_

**Q:** What geopolitical concern could undermine AI progress in life sciences?
**问：** 什么地缘政治问题可能破坏AI在生命科学领域的进步?

**A:** The speaker expresses concern about the fragmentation of global scientific collaboration as different regions develop separate AI ecosystems. Having "witnessed an absolute revolution in science" becoming "a global sport" that creates "wonderful variability of opinions" enabling the solving of "greater questions together," the speaker fears this collaboration will erode. The specific concern is that "US will have their models, China will have their models, Europe not so much yet," leading to siloed development rather than the collective problem-solving that has characterized recent scientific progress. This balkanization could undermine the collaborative advantage that has made science increasingly effective.
**答：** 讲者担心随着不同地区开发独立的AI生态系统,全球科学合作会分裂。见证了科学"绝对革命"成为"全球运动",创造了"精彩的观点多样性",使我们能够"共同解决更大的问题",讲者担心这种合作会被侵蚀。具体担忧是"美国有自己的模型,中国有自己的模型,欧洲还没那么多",导致孤立开发而非近期科学进步所特有的集体问题解决。这种分裂可能会削弱使科学变得越来越有效的协作优势。

### Topic 22: Hope: Bilingual Teams Bridging Science and AI
期望:架起科学与AI桥梁的双语团队
_[35:24]_

**Q:** What organizational change would most accelerate AI adoption in research institutions?
**问：** 什么组织变革能最有效加速研究机构采用AI?

**A:** The speaker calls for institutions and companies to "help their people to understand the full implications of how they can be helped with AI" by embedding "bilingual people in all teams everywhere." This does not mean people who speak multiple languages, but rather individuals "who are completely fluent in some scientific topic as well as in digital and AI." The emphasis is on embedded expertise: "You cannot just say to people 'use AI'"—instead, having "one person in each team can do wonders." This represents a structural solution where domain expertise and AI literacy are combined within working teams rather than treated as separate functions, enabling organizations to "truly accomplish great things together for the benefit of the world."
**答：** 讲者呼吁机构和公司"帮助员工理解AI如何帮助他们的全部含义",方法是在"所有团队中嵌入双语人才"。这不是指会说多种语言的人,而是"在某个科学主题以及数字和AI方面都完全流利"的人。强调的是嵌入式专业知识:"不能只是对人们说'使用AI'"——相反,"每个团队有一个这样的人就能创造奇迹"。这代表了一种结构性解决方案,将领域专业知识和AI素养结合在工作团队内部,而不是作为独立职能对待,使组织能够"真正共同完成伟大事业,造福世界"。

### Topic 23: Introduction: Eric Carter Abrams and Life Sciences Track
引入:Eric Carter Abrams与生命科学轨道
_[36:19]_

**Q:** Who is leading Anthropic's life sciences initiatives?
**问：** 谁在领导Anthropic的生命科学项目?

**A:** The transition introduces Eric Carter Abrams as Anthropic's life sciences leader. The framing emphasizes two key insights from the preceding conversation: first, that "we can't neatly reverse engineer our way" to the ambitious vision outlined in "Machines of Loving Grace," and second, that this uncertainty "shouldn't inhibit us from simply starting." The introduction reinforces that as "AI continue to grow exponentially," what matters is not "where you start" but "the fact that you simply do start." This sets the stage for Carter Abrams to detail Anthropic's specific approach to starting their journey in life sciences, emphasizing action over perfect planning.
**答：** 过渡部分介绍了Eric Carter Abrams作为Anthropic的生命科学负责人。框架强调了前面对话的两个关键见解:首先,"我们无法简单地逆向工程"到"Machines of Loving Grace"中概述的宏伟愿景;其次,这种不确定性"不应阻止我们简单地开始"。介绍强化了随着"AI继续指数级增长",重要的不是"从哪里开始"而是"你确实开始了"这一事实。这为Carter Abrams详细说明Anthropic在生命科学领域开始旅程的具体方法奠定了基础,强调行动胜于完美规划。

### Topic 24: Exponential Progress Across Disciplines: Coding vs Life Sciences
跨学科的指数进步:编程vs生命科学
_[37:43]_

**Q:** Why will AI transformation in life sciences take longer than in coding?
**问：** 为什么AI在生命科学的转型比编程需要更长时间?

**A:** Carter Abrams draws a parallel with coding, where AI rapidly progressed from autocomplete to "working perhaps at the level of a junior engineer" and then to "increasingly autonomous and capable" systems. However, he acknowledges that "it's going to take longer for this same change to happen in the life sciences" because "it's a much harder problem." The key differences are that life sciences "feedback loops take longer and they involve running real experiments in the physical world" with "so much uncertainty and noise in all biological data." Despite these challenges, the core message is that "the same change is absolutely coming"—it's a matter of timeline, not feasibility.
**答：** Carter Abrams与编程进行类比,AI在编程中迅速从自动完成进步到"可能达到初级工程师水平",然后到"越来越自主和有能力"的系统。然而,他承认"生命科学中相同变化的发生需要更长时间",因为"这是一个困难得多的问题"。关键区别在于生命科学的"反馈循环更长,涉及在物理世界中运行真实实验",且"所有生物数据都有太多不确定性和噪声"。尽管存在这些挑战,核心信息是"相同的变化绝对会到来"——这是时间线问题,而非可行性问题。

### Topic 25: Anthropic's Life Sciences Objectives and Approach
Anthropic的生命科学目标与方法
_[38:51]_

**Q:** What are Anthropic's primary objectives and strategy for life sciences?
**问：** Anthropic在生命科学领域的主要目标和策略是什么?

**A:** Anthropic pursues two primary objectives: "accelerating scientific discovery as an end in itself, pure pursuit of basic research" and "alleviating the burden of disease and aging." The approach is explicitly "full stack," constructed to pursue these objectives "as fast as we can." This full-stack approach encompasses multiple layers: the foundational layer with Claude models, the product layer focused on "optimizing the product for scientists" to make "model intelligence integrated into workflows and accessible," and partnership work with customers and internal efforts to directly pursue these objectives. The emphasis on product optimization reflects recognition that "even with the most capable models in the world, we still need to have the right product features" for practical scientific use.
**答：** Anthropic追求两个主要目标:"加速科学发现本身,纯粹追求基础研究"和"减轻疾病和衰老的负担"。方法明确为"全栈式",旨在"尽可能快地"追求这些目标。这种全栈方法包含多个层次:以Claude模型为基础的基础层,专注于"为科学家优化产品"以使"模型智能集成到工作流程中并可访问"的产品层,以及与客户的合作工作和内部努力来直接追求这些目标。对产品优化的强调反映了这样的认识:"即使拥有世界上最强大的模型,我们仍然需要正确的产品功能"才能实现实际科学应用。

### Topic 26: Model Progress: Claude Opus Performance on Scientific Benchmarks
模型进展:Claude Opus在科学基准测试中的表现
_[39:54]_

**Q:** How has Claude's performance on scientific tasks progressed?
**问：** Claude在科学任务上的表现如何进步?

**A:** Carter Abrams presents evidence of "rapid progress in the underlying foundation model capabilities" over approximately six months, focusing on the Claude Opus series from Opus 4.5 to Opus 4.8. The benchmarks span organic chemistry, bioinformatics, and structural biology. The progression is characterized as dramatic: models have gone "from being not all that useful to performing at a level that is on par or greater than the average PhD level scientist in these fields." This demonstrates that the fundamental question—"Can AI actually tackle scientific problems?"—has been answered affirmatively through measurable benchmark improvements, establishing the foundation for the product-layer work that follows.
**答：** Carter Abrams展示了"底层基础模型能力的快速进步"证据,时间跨度约六个月,聚焦于Claude Opus系列从Opus 4.5到Opus 4.8。基准测试涵盖有机化学、生物信息学和结构生物学。进步被描述为显著的:模型"从不那么有用到表现达到或超过这些领域平均PhD科学家的水平"。这证明了基本问题——"AI真的能解决科学问题吗?"——已通过可衡量的基准测试改进得到肯定回答,为随后的产品层工作奠定了基础。

### Topic 27: Introduction to Claude Science and Speaker Background
Claude Science 介绍及演讲者背景
_[44:26]_

**Q:** Who is presenting Claude Science and what motivated its development?
**问：** 谁在展示 Claude Science,开发它的动机是什么?

**A:** Alec Terashansky leads product development for Claude Science, bringing firsthand experience from his years as a computational biologist. He describes the painful reality of scientific research: "old databases with poorly documented schemas, pipelines that break when a dependency updates," and the chaos of Jupyter notebooks "executed out of order." The inefficiency extends to visualization work with "horrible hours spent making figures in Matplotlib and Illustrator." This accumulated friction means "entire fields are underexplored because the research cycle is too slow and too tedious," creating the fundamental problem Claude Science aims to solve.
**答：** Alec Terashansky 负责 Claude Science 的产品开发,他曾是计算生物学家,有着切身体会。他描述了科研的痛苦现实:文档不全的旧数据库、依赖更新就会崩溃的流程,以及混乱的 Jupyter notebooks 执行顺序错乱的问题。可视化工作也耗费大量时间。这些累积的摩擦导致"整个研究领域未被充分探索,因为研究周期太慢太繁琐",这正是 Claude Science 要解决的根本问题。

### Topic 28: PKU Drug Discovery Demo Setup
PKU 药物发现演示设置
_[45:43]_

**Q:** What real-world drug discovery problem is being used to demonstrate Claude Science?
**问：** 用什么真实世界的药物发现问题来演示 Claude Science?

**A:** The demonstration focuses on phenylketonuria (PKU), a disease where "one broken enzyme" causes an amino acid to accumulate "until it damages the brain." While an approved small molecule drug exists, "it doesn't work in the most common severe mutation," motivating other groups to pursue new therapies. The presenter emphasizes the traditional timeline burden: "Month one of a program like this is a lot of manual work" including literature review, structure preparation, genetics analysis, screen scoping, and building the business case, requiring "three to six weeks before a single experiment can be run." This sets up the contrast with Claude Science's approach.
**答：** 演示聚焦于苯丙酮尿症(PKU),这是一种由于"一个酶的缺陷"导致氨基酸累积"直到损害大脑"的疾病。虽然已有批准的小分子药物,但"它对最常见的严重突变无效",这促使其他团队寻求新疗法。演讲者强调传统时间线的负担:这类项目的"第一个月需要大量手工工作",包括文献综述、结构准备、遗传学分析、筛选范围界定和商业案例构建,"需要三到六周才能进行第一个实验"。这为对比 Claude Science 的方法做了铺垫。

### Topic 29: Single-Sentence Prompt and Multi-Agent Plan Generation
单句提示词与多智能体计划生成
_[46:26]_

**Q:** How does Claude Science translate a simple request into an executable research plan?
**问：** Claude Science 如何将简单请求转化为可执行的研究计划?

**A:** The presenter demonstrates starting with minimal input: "Find me a stabilizer for PAH variants" and "get me up to speed on PKU as the indication so we can put a program together." Claude autonomously "built a plan" with three phases executed through "parallel sub-agents." Phase one involves landscaping analysis with three sub-agents studying variant biology, structure and pockets assessment, and investment case building. Phase two focuses on library assembly, distributing compounds "across a bunch of GPUs to assess their binding affinity," scoring results, and producing deliverables leading to a "go/no-go verdict." The system provides transparency by showing "its confidence in the scope and feasibility of the plan," and while users would "normally iterate with Claude to refine the plan," the demo proceeds with immediate approval.
**答：** 演讲者展示了从最少输入开始的过程:"找到 PAH 变体的稳定剂"和"让我了解 PKU 适应症以便制定项目"。Claude 自主"构建了计划",通过"并行子智能体"执行三个阶段。第一阶段是全景分析,三个子智能体分别研究变体生物学、结构和口袋评估、投资案例构建。第二阶段专注于库组装,将化合物"分布在多个 GPU 上评估结合亲和力",对结果评分,并产生可交付成果以得出"继续/中止决策"。系统通过显示"对计划范围和可行性的信心"提供透明度,虽然用户"通常会迭代改进计划",但演示中直接批准了。

### Topic 30: Structure Analysis: Mutation Mapping and Druggability Assessment
结构分析:突变映射与成药性评估
_[48:00]_

**Q:** What autonomous scientific analysis did Claude perform before any computational screening?
**问：** Claude 在任何计算筛选之前自主执行了什么科学分析?

**A:** The structure and pockets sub-agent received a detailed brief including "steps you should complete" and "output schema that you should return," emphasizing transparency through sub-agent delegation. Claude independently performed three critical analyses not explicitly requested. First, it "confirmed that R48W is the variant that matters" as the most common severe mutation. Second, it mapped the mutation onto the crystal structure, revealing the active site and mutation site are "over 20 angstroms apart," immediately indicating "this is likely not a problem with a broken active site" but rather "a folding problem," validating that "a stabilizer is the right call." Third, it checked for druggability at the mutation site, finding "essentially a smooth surface" with "zero druggability on a scale of 0 to one," closing "that dead end before we even tried it." Additionally, it built the business case identifying competitors and establishing "separatory precedent."
**答：** 结构和口袋子智能体收到详细简报,包括"应完成的步骤"和"应返回的输出架构",通过子智能体委派强调透明度。Claude 独立执行了三个未明确要求的关键分析。首先,它"确认 R48W 是重要的变体",是最常见的严重突变。其次,它将突变映射到晶体结构上,发现活性位点和突变位点"相距 20 埃以上",立即表明"这可能不是活性位点损坏的问题"而是"折叠问题",验证了"稳定剂是正确的选择"。第三,它检查了突变位点的成药性,发现"基本上是光滑表面","成药性评分为零(满分一分)",在"我们尝试之前就关闭了那条死胡同"。此外,它还构建了商业案例,确定竞争对手并建立"分离先例"。

### Topic 31: Product Primitives: Built-in Scientific Capabilities and Customization
产品原语:内置科学能力与定制化
_[50:15]_

**Q:** What scientific capabilities come built into Claude Science and how can users extend them?
**问：** Claude Science 内置了哪些科学能力,用户如何扩展它们?

**A:** Claude Science "ships with capabilities in many different domains" including "proteomics, structural biology, chemistry, genomics, literature review—altogether more than 60 plus databases and scientific resources" accessible through built-in skills and connectors. The system is designed to be "ready for your domain out of the box," but recognizes that pre-built capabilities won't cover every use case. The platform offers "extraordinarily customizable" extension mechanisms: users can "add a skill simply by chatting with Claude," write one from scratch, upload from a zip file, or "import it from your favorite GitHub repository." For Model Context Protocol (MCP) integration, "any local or remote MCP that you have access to, you can add them," providing flexible infrastructure for domain-specific tooling.
**答：** Claude Science "内置多个领域的能力",包括"蛋白质组学、结构生物学、化学、基因组学、文献综述——总共 60 多个数据库和科学资源",可通过内置技能和连接器访问。系统设计为"开箱即用适配你的领域",但认识到预构建能力无法覆盖所有用例。平台提供"极其可定制"的扩展机制:用户可以"通过与 Claude 聊天添加技能",从头编写,从 zip 文件上传,或"从你喜欢的 GitHub 仓库导入"。对于 Model Context Protocol(MCP)集成,"任何你能访问的本地或远程 MCP,都可以添加",为特定领域工具提供灵活的基础设施。

### Topic 32: Artifact System: Provenance, Versioning, and Reproducibility
Artifact 系统:溯源、版本控制与可重现性
_[51:22]_

**Q:** How does Claude Science ensure reproducibility and track the history of research outputs?
**问：** Claude Science 如何确保可重现性并跟踪研究输出的历史?

**A:** The artifact system represents a fundamental shift in scientific reproducibility. "Every output that Claude produces comes with its full history attached," including the code that produced it, input artifacts it depends on, "the full execution log of all the cells executed in that session," the conversation context, and "the exact environmental snapshot that produced this artifact." This comprehensive provenance means "you can come back to the product six months, a year, two years later, every artifact is reproducible by construction." The system enables powerful interaction patterns because "Claude knows how every artifact was made." Users can annotate figures using vision capabilities without explicit references—simply stating "this label is hard to see" triggers Claude to examine the code provenance and make "the precise edit." Every artifact is versioned with "versions are immutable," and each version maintains its own provenance, creating a permanent record of iterative refinements.
**答：** Artifact 系统代表了科学可重现性的根本转变。"Claude 产生的每个输出都附带完整历史",包括生成它的代码、依赖的输入 artifacts、"该会话中执行的所有单元格的完整执行日志"、对话上下文和"产生此 artifact 的精确环境快照"。这种全面的溯源意味着"六个月、一年、两年后回来,每个 artifact 在构造上都是可重现的"。系统支持强大的交互模式,因为"Claude 知道每个 artifact 是如何制作的"。用户可以使用视觉能力标注图表而无需明确引用——只需说"这个标签很难看清",就会触发 Claude 检查代码溯源并进行"精确编辑"。每个 artifact 都有版本控制,"版本是不可变的",每个版本都保持自己的溯源,创建迭代改进的永久记录。

### Topic 33: Built-in Reviewer and Error Correction
内置审查者与错误纠正
_[53:36]_

**Q:** How does Claude Science validate the accuracy of agent outputs?
**问：** Claude Science 如何验证智能体输出的准确性?

**A:** The system implements quality control through an architectural pattern where "underneath every agent is a built-in reviewer that is assessing the accuracy of every claim that the agent is making and every artifact it produces." This creates a two-layer verification system. When the structure and pockets sub-agent "wrote a brief," a markdown document, "the reviewer caught a mistake" and "injected a notice into the agent thread." The agent then "corrected it" with "both versions on the record," allowing users to see the diff between versions. This automatic verification layer operates continuously without user intervention, catching errors before they propagate through the research workflow.
**答：** 系统通过架构模式实现质量控制,"每个智能体下面都有一个内置审查者,评估智能体所做的每个声明和产生的每个 artifact 的准确性"。这创建了双层验证系统。当结构和口袋子智能体"写了一份简报"(markdown 文档)时,"审查者发现了一个错误"并"向智能体线程注入通知"。智能体随后"纠正了它","两个版本都记录在案",允许用户查看版本之间的差异。这个自动验证层持续运行,无需用户干预,在错误传播到研究工作流之前就将其捕获。

### Topic 34: End-to-End Reproducibility Example and Science Future
端到端可重现性示例与科学未来
_[54:10]_

**Q:** What does comprehensive reproducibility look like in a real research project using Claude Science?
**问：** 在使用 Claude Science 的真实研究项目中,全面的可重现性是什么样的?

**A:** The presenter shares a compelling personal example: "in my own personal research project where I continued my own PhD, at this point, Claude has produced thousands and thousands of artifacts all leading up to one output, a manuscript." The significance lies not just in the volume but in the quality of reproducibility—"a manuscript that is by construction fully reproducible end to end." This represents a paradigm shift from traditional scientific publishing where methods sections are often insufficient for true replication. The presenter's conviction is clear: "I honestly think that that is the future of science," envisioning a world where every published result carries its complete computational provenance, making scientific claims verifiable by construction rather than by trust.
**答：** 演讲者分享了一个有力的个人例子:"在我继续攻读博士学位的个人研究项目中,到目前为止,Claude 已经产生了成千上万个 artifacts,最终汇聚成一个输出:一份手稿"。重要的不仅是数量,还有可重现性的质量——"一份在构造上完全端到端可重现的手稿"。这代表了从传统科学出版的范式转变,传统方法部分通常不足以实现真正的复制。演讲者的信念很明确:"我真的认为这就是科学的未来",设想一个世界,其中每个发表的结果都带有完整的计算溯源,使科学声明通过构造而非信任来验证。

### Topic 35: Computational Screening and Distributed Compute
计算筛选与分布式计算
_[54:36]_

**Q:** How does Claude Science handle large-scale computational screening tasks?
**问：** Claude Science 如何处理大规模计算筛选任务?

**A:** Following the landscaping analysis, Claude autonomously executed the computational screening phase by compiling "a focused library of 2,200 compounds" and distributing them "across 80 GPUs." This represents the third core product pillar: Compute. The system demonstrates the ability to coordinate large-scale parallel processing without explicit user configuration, translating the high-level goal of screening compounds into a distributed computational workflow. The scale—2,200 compounds across 80 GPUs—suggests the platform abstracts away infrastructure complexity, allowing researchers to focus on scientific questions rather than cluster management.
**答：** 在全景分析之后,Claude 自主执行计算筛选阶段,编译了"一个包含 2,200 个化合物的专注库"并将它们"分布在 80 个 GPU 上"。这代表了第三个核心产品支柱:Compute。系统展示了协调大规模并行处理的能力,无需明确的用户配置,将筛选化合物的高级目标转化为分布式计算工作流。规模——80 个 GPU 上的 2,200 个化合物——表明平台抽象了基础设施复杂性,使研究人员能够专注于科学问题而不是集群管理。

### Topic 36: Cloud Science Portability and Infrastructure Support
Cloud Science 的可移植性和基础设施支持
_[55:05]_

**Q:** How does Cloud Science handle deployment and infrastructure requirements for computational biologists?
**问：** Cloud Science 如何处理计算生物学家的部署和基础设施需求?

**A:** Speaker A emphasizes that Cloud Science was designed as a "drop-in replacement for Jupyter notebooks" to minimize friction for computational biologists. The platform's portability means it can run on laptops, cloud VMs, Linux workstations, and connect to any SSH-accessible host including lab clusters and EC2 instances. Beyond basic SSH connectivity, Cloud Science offers "built-in integrations with Modal as a cloud provider" and can access model endpoints through Nvidia, providing researchers flexibility to work wherever their data resides without worrying about compatibility issues.
**答：** Speaker A 强调 Cloud Science 被设计为 "Jupyter notebooks 的直接替代品",以最大程度减少计算生物学家的使用障碍。该平台的可移植性意味着它可以在笔记本电脑、云虚拟机、Linux 工作站上运行,并连接到任何可通过 SSH 访问的主机,包括实验室集群和 EC2 实例。除了基本的 SSH 连接外,Cloud Science 还提供了 "与 Modal 作为云提供商的内置集成",并可以通过 Nvidia 访问模型端点,让研究人员可以灵活地在数据所在的任何地方工作,无需担心兼容性问题。

### Topic 37: Computational Drug Screening Results and Filtering Pipeline
计算药物筛选结果和过滤流程
_[55:42]_

**Q:** What were the results of the automated computational screening campaign that Claude executed?
**问：** Claude 执行的自动化计算筛选活动的结果是什么?

**A:** Speaker A walks through the filtering pipeline results, starting with Claude collecting "all 80 jobs" with two failures that didn't matter. Of the 2,200 compounds that went into the screening, approximately 2,100 were successfully folded and assessed for binding affinity with the target enzyme. The pipeline applied increasingly stringent filters: 723 compounds passed the initial binding affinity thresholds, and then "four survived a check with a second independent model." This multi-stage validation approach demonstrates how the system narrows from thousands of candidates to a handful of high-confidence leads through computational validation.
**答：** Speaker A 详细介绍了过滤流程的结果,从 Claude 收集 "所有 80 个任务" 开始,其中两个失败但不影响整体。在进入筛选的 2,200 个化合物中,约 2,100 个成功折叠并评估了与目标酶的结合亲和力。流程应用了越来越严格的过滤器:723 个化合物通过了初始结合亲和力阈值,然后 "四个通过了第二个独立模型的检查"。这种多阶段验证方法展示了系统如何通过计算验证从数千个候选化合物缩小到少数几个高置信度的先导化合物。

### Topic 38: Visual Artifact Rendering Capabilities in Cloud Science
Cloud Science 中的可视化输出渲染能力
_[56:14]_

**Q:** What types of scientific visualizations and artifacts can Cloud Science render natively?
**问：** Cloud Science 可以原生渲染哪些类型的科学可视化和输出?

**A:** Speaker A highlights that "Cloud Science is a very visual product" with built-in artifact renderers that go far beyond simple images. The platform supports markdown documents, HTML dashboards, chemistry structures, structure viewers, and more specialized scientific data types. Recognizing that "the tail of data types in this space is heavy," the team is committed to continuously expanding renderer support. The demo shows compounds on the left, the protein structure with compounds overlaid, and interactive elements where users can toggle views and even iterate on HTML documents with Claude to refine visualizations based on feedback like "the legend is covering it."
**答：** Speaker A 强调 "Cloud Science 是一个非常可视化的产品",拥有远超简单图像的内置输出渲染器。该平台支持 markdown 文档、HTML 仪表板、化学结构、结构查看器以及更多专业的科学数据类型。认识到 "该领域的数据类型长尾效应很重",团队致力于持续扩展渲染器支持。演示显示了左侧的化合物、叠加了化合物的蛋白质结构,以及交互元素,用户可以切换视图,甚至可以与 Claude 在 HTML 文档上迭代以根据反馈(如 "图例挡住了")优化可视化。

### Topic 39: Drug Candidate Ranking and Comparison to Approved Drug
药物候选物排名及与已批准药物的比较
_[56:59]_

**Q:** How did the computational screening results compare the novel candidates to the existing approved drug?
**问：** 计算筛选结果如何将新候选物与现有已批准药物进行比较?

**A:** Speaker A reveals a striking finding in the ranked results: sertraline, the approved drug for this indication, was "ranked dead last for binding affinity to the pocket." In contrast, the top four candidates identified through the computational screen showed superior binding profiles, with visible "stabilizing arms of the compound" that suggest better interaction with the target. Claude also generated a comprehensive ranked list with all metadata attached, where the top four survivors of the second model check are clearly distinguished from the 700 trailing candidates. This comparison demonstrates how computational screening can identify potentially superior candidates compared to existing approved therapies.
**答：** Speaker A 揭示了排名结果中的一个惊人发现:sertraline(该适应症的已批准药物)在 "与口袋的结合亲和力排名中垫底"。相比之下,通过计算筛选确定的前四名候选物显示出更优越的结合特性,可见 "化合物的稳定臂",表明与靶标有更好的相互作用。Claude 还生成了一个附带所有元数据的综合排名列表,其中通过第二个模型检查的前四名幸存者与后面的 700 个候选物明确区分。这种比较展示了计算筛选如何能够识别出可能优于现有已批准疗法的候选物。

### Topic 40: AI-Generated Go/No-Go Decision Framework and Kill Criteria
AI 生成的 Go/No-Go 决策框架和终止标准
_[57:46]_

**Q:** What decision-making framework did Claude provide for advancing or terminating the drug program?
**问：** Claude 为推进或终止药物项目提供了什么决策框架?

**A:** Speaker A emphasizes the business intelligence aspect of the output, noting that Claude produced a "go no-go memo" that provided strategic guidance beyond just scientific results. Claude gave a "conditional go" recommendation, specifying "the first decisive experiment we would need to do to fund a larger experimental campaign." Critically, it also established "kill criteria for that experiment" - defining clear thresholds for when to abandon this approach and look for alternatives. This demonstrates how the AI can bridge "from the bench to the boardroom in a single session," providing not just scientific analysis but strategic decision frameworks that executives need for resource allocation.
**答：** Speaker A 强调了输出的商业智能方面,指出 Claude 生成了一份 "go no-go 备忘录",提供了超越科学结果的战略指导。Claude 给出了 "有条件的 go" 建议,明确了 "我们需要进行的第一个决定性实验以资助更大规模的实验活动"。关键的是,它还建立了 "该实验的终止标准" - 定义了何时放弃该方法并寻找替代方案的明确阈值。这展示了 AI 如何能够 "在一个会话中从实验台连接到董事会",不仅提供科学分析,还提供高管进行资源配置所需的战略决策框架。

### Topic 41: Parallel Rare Disease Screening at Scale with Sub-Agents
使用子代理进行大规模并行罕见病筛选
_[58:22]_

**Q:** How did the system scale to evaluate multiple rare diseases simultaneously?
**问：** 系统如何扩展以同时评估多种罕见病?

**A:** Speaker A demonstrates the scalability of the approach by describing a parallel campaign where Claude was asked to "run a month one stabilizer assessment for 100 rare monogenic diseases." Using "100 parallel sub-agents," each sub-agent independently performed landscaping analysis for one disease, identifying which variant matters, analyzing structure implications, and assessing whether a computational screen was worthwhile. The results showed that "out of 100 rare diseases, 32 were worth a full computational screen," all completed "under an hour." Combined with the previous full screening session taking under two hours, Speaker A argues that "scale is no longer an issue" and asks rhetorically why stop at 100 when 1,000, 5,000, or 10,000 diseases could be evaluated with similar efficiency.
**答：** Speaker A 通过描述一个并行活动展示了该方法的可扩展性,其中要求 Claude "为 100 种罕见单基因疾病运行第一个月的稳定剂评估"。使用 "100 个并行子代理",每个子代理独立地对一种疾病进行景观分析,识别哪个变体重要,分析结构影响,并评估是否值得进行计算筛选。结果显示 "100 种罕见病中有 32 种值得进行全面计算筛选",所有这些都在 "不到一小时" 内完成。结合之前不到两小时的全面筛选会话,Speaker A 认为 "规模不再是问题",并反问为什么要停留在 100 种,而不是用类似的效率评估 1,000、5,000 或 10,000 种疾病。

### Topic 42: Transforming the Economics of Scientific Exploration
改变科学探索的经济学
_[59:17]_

**Q:** How does this technology change what's possible for individual researchers and underexplored fields?
**问：** 这项技术如何改变个人研究人员和未充分探索领域的可能性?

**A:** Speaker A reflects on the broader implications, noting that "entire fields sit underexplored because the cycle is too slow and expensive" and that this technology represents what happens "when it isn't." Speaking from personal experience, they note it's "a dream to be able to build this product because if I had this when I was in the lab, the scale at which I could have worked and the ideas I could have afforded to try would have been completely different." This democratization of capability means individual researchers can now pursue ambitious ideas that previously would have required large teams and substantial funding, fundamentally changing who can contribute to scientific progress and which problems get explored.
**答：** Speaker A 反思了更广泛的影响,指出 "整个领域因为周期太慢和成本太高而未得到充分探索",而这项技术代表了 "当它不是这样时" 的情况。从个人经验出发,他们指出 "能够构建这个产品是一个梦想,因为如果我在实验室时有这个,我能够工作的规模和我能够尝试的想法会完全不同"。这种能力的民主化意味着个人研究人员现在可以追求以前需要大团队和大量资金的雄心勃勃的想法,从根本上改变了谁可以为科学进步做出贡献以及探索哪些问题。

### Topic 43: Cloud Science Ecosystem and Integration Architecture
Cloud Science 生态系统和集成架构
_[01:00:19]_

**Q:** What is the partnership and integration strategy behind Cloud Science?
**问：** Cloud Science 背后的合作伙伴关系和集成策略是什么?

**A:** Speaker C (Eric) explains that everything demonstrated is "powered by an open ecosystem of connectors" to partners throughout the scientific world, specifically naming "Benchling, Bolts, Latch Bio" as key partners. The architecture was deliberately "designed to be easy to add connectors to all the tools that scientists need to use every day," suggesting a plugin-based approach that allows Cloud Science to integrate with the existing software infrastructure researchers rely on. This open ecosystem strategy positions Cloud Science as a hub that connects to specialized tools rather than attempting to replace them, enabling researchers to maintain their existing workflows while gaining AI capabilities.
**答：** Speaker C (Eric) 解释说,所展示的一切都是 "由一个开放的连接器生态系统驱动的",连接到整个科学界的合作伙伴,特别提到了 "Benchling、Bolts、Latch Bio" 作为关键合作伙伴。该架构被有意 "设计为易于添加连接器到科学家每天需要使用的所有工具",表明采用了基于插件的方法,允许 Cloud Science 与研究人员依赖的现有软件基础设施集成。这种开放生态系统策略将 Cloud Science 定位为连接到专业工具的中心,而不是试图取代它们,使研究人员能够在获得 AI 能力的同时保持现有工作流程。

### Topic 44: UCSF RNA Sequencing Case: Virus Contamination Detection
UCSF RNA 测序案例:病毒污染检测
_[01:00:46]_

**Q:** How did Cloud Science demonstrate its value in detecting data quality issues that eluded human researchers?
**问：** Cloud Science 如何展示其在检测人类研究人员未发现的数据质量问题方面的价值?

**A:** Speaker A recounts a compelling case study from UCSF involving RNA sequencing data analysis where "something wasn't quite right with the data" and it took the research team "over the course of a year" to eventually discover that "a virus contaminating the sample" explained the unusual results. When the same data was provided to Cloud Science, the system identified the virus contaminant "on the first go, over the course of a few minutes." Speaker A emphasizes that "these sort of experiences are critical for building confidence in using AI so thoroughly in scientific workflows," as the ability to catch issues that human experts miss over long periods demonstrates both the system's analytical depth and its potential to prevent costly delays in research programs.
**答：** Speaker A 讲述了一个来自 UCSF 的引人注目的案例研究,涉及 RNA 测序数据分析,其中 "数据有些不太对劲",研究团队花了 "一年多时间" 最终发现 "样本被病毒污染" 解释了异常结果。当相同数据提供给 Cloud Science 时,系统 "在第一次尝试时,在几分钟内" 识别出了病毒污染物。Speaker A 强调 "这类体验对于建立在科学工作流程中全面使用 AI 的信心至关重要",因为能够捕获人类专家在很长时间内错过的问题,既展示了系统的分析深度,也展示了其防止研究项目代价高昂延误的潜力。

### Topic 45: Manifold Bio Case: Raw Data to Publication Figures
Manifold Bio 案例:从原始数据到出版图表
_[01:01:19]_

**Q:** How does Cloud Science accelerate the path from raw experimental data to publishable results?
**问：** Cloud Science 如何加速从原始实验数据到可发表结果的路径?

**A:** Speaker A presents a use case from Manifold Bio, a drug development company, where scientists are "able to go all the way from raw data to publication quality figures in a single session." The workflow encompasses "all of the analyses, all of the visual iterations on the figures" while maintaining "the full history of the changes available" throughout the process. This end-to-end capability eliminates the traditional fragmentation where data analysis, visualization, and figure preparation happen in separate tools or sessions, often requiring manual tracking of versions and changes. The ability to iterate on publication-quality outputs within a unified environment dramatically reduces the time from experimental results to manuscript submission.
**答：** Speaker A 展示了来自 Manifold Bio(一家药物开发公司)的用例,科学家们 "能够在单个会话中从原始数据一路走到出版质量的图表"。工作流程包括 "所有分析、图表的所有可视化迭代",同时在整个过程中保持 "变更的完整历史可用"。这种端到端的能力消除了传统的碎片化,即数据分析、可视化和图表准备在单独的工具或会话中进行,通常需要手动跟踪版本和变更。在统一环境中迭代出版质量输出的能力大大减少了从实验结果到手稿提交的时间。

### Topic 46: Whitehead Institute Case: Democratizing Computational Skills
Whitehead Institute 案例:计算技能的民主化
_[01:01:59]_

**Q:** How does Cloud Science change who can perform computational biology workflows?
**问：** Cloud Science 如何改变谁可以执行计算生物学工作流程?

**A:** Speaker A highlights what they consider "one of the most important themes of Cloud Science" through a testimonial from a Whitehead Institute professor. The platform enables scientists with "primarily experimental biology backgrounds" to independently perform "entire workflows that include the computational part as well." This capability breakdown represents a fundamental shift in scientific productivity, where researchers no longer need to wait for computational specialists or learn programming themselves to execute analyses. The key themes across all customer examples are "accelerating workflows by an order of magnitude in some cases and enabling small teams of scientists to do what previously would have taken much larger teams including a larger array of different skill sets."
**答：** Speaker A 通过来自 Whitehead Institute 教授的推荐强调了他们认为的 "Cloud Science 最重要的主题之一"。该平台使 "主要具有实验生物学背景" 的科学家能够独立执行 "包括计算部分的整个工作流程"。这种能力突破代表了科学生产力的根本转变,研究人员不再需要等待计算专家或自己学习编程来执行分析。所有客户示例的关键主题是 "在某些情况下将工作流程加速一个数量级,并使小团队科学家能够完成以前需要更大团队(包括更大范围的不同技能集)才能完成的工作"。

### Topic 47: Claude Product Suite for Life Sciences Value Chain
用于生命科学价值链的 Claude 产品套件
_[01:02:46]_

**Q:** How does Anthropic's product portfolio address different parts of the drug development value chain?
**问：** Anthropic 的产品组合如何解决药物开发价值链的不同部分?

**A:** Speaker A emphasizes that while the presentation focused on R&D, drug development involves much more, including "clinical and regulatory, commercial, manufacturing and operations." Anthropic's strategy is to offer "cloud offerings from the models and products" that are "designed to address all of it." Rather than treating Cloud Science as the only solution, the company positions different products for different needs across the value chain. To illustrate this breadth, they transition to demonstrating Cloud Code for a biostats use case, showing how the same underlying Claude capabilities can be packaged differently for software development tasks versus scientific research workflows.
**答：** Speaker A 强调,虽然演示专注于 R&D,但药物开发涉及更多方面,包括 "临床和监管、商业、制造和运营"。Anthropic 的策略是提供 "来自模型和产品的云产品",这些产品 "旨在解决所有问题"。公司不是将 Cloud Science 视为唯一解决方案,而是针对价值链上的不同需求定位不同的产品。为了说明这种广度,他们转向演示用于生物统计学用例的 Cloud Code,展示相同的底层 Claude 能力如何针对软件开发任务与科学研究工作流程进行不同的打包。

### Topic 48: Partnerships and Deployment: Bringing AI Models to Scientists
合作伙伴关系与部署:将AI模型带给科学家
_[01:10:19]_

**Q:** What is the role of the partnerships and deployment group, and how do scientists integrate new AI technologies into their work?
**问：** 合作伙伴关系和部署团队的角色是什么?科学家如何将新的AI技术整合到他们的工作中?

**A:** The partnerships and deployment group acts as beneficiaries of AI models and capabilities, asking two fundamental questions: where can they have a positive impact on the frontiers of science, and who are the right partners to bring that possibility to reality. The speaker emphasizes wanting to see these capabilities "in as many scientists' hands as possible" across academia, biotech, pharma, and the wider ecosystem. However, they acknowledge that when scientists receive new technologies—whether CRISPR, next generation sequencing, or a "computational collaborator"—that's actually "where many experiments begin." The integration challenge isn't just technical but organizational and conceptual: scientists must figure out how to integrate these tools into their work, their organizations, and "start to rethink the questions that are possible."
**答：** 合作伙伴关系和部署团队作为AI模型和能力的受益者,提出两个基本问题:在哪些科学前沿可以产生积极影响,以及谁是将这种可能性变为现实的合适合作伙伴。发言人强调希望让这些能力"进入尽可能多的科学家手中",覆盖学术界、生物技术、制药和更广泛的生态系统。然而,他们承认当科学家获得新技术时——无论是CRISPR、下一代测序还是"计算协作者"——那实际上是"许多实验开始的地方"。整合挑战不仅是技术性的,还涉及组织和概念层面:科学家必须弄清楚如何将这些工具整合到他们的工作中、他们的组织中,并"开始重新思考可能的问题"。

### Topic 49: Why Biology is Fundamentally Difficult: Scale, Complexity, and Measurement Limitations
为什么生物学从根本上很困难:规模、复杂性和测量限制
_[01:12:32]_

**Q:** What makes biology and drug R&D inherently difficult from a computational and scientific perspective?
**问：** 从计算和科学角度来看,是什么让生物学和药物研发从本质上很困难?

**A:** Aviv Regev explains that biology isn't just complex—"it's actually just huge, like huge, huge." All the numbers in biology are so large they exceed experimental capacity, "bigger than number of cells on the planet, definitely people on the planet, stars in the universe, atoms in the universe." Beyond scale, biology is "multiscale," requiring thinking at levels from "atoms, and then molecules, and then cells, and then tissues, and then patients, and then populations of patients." Additionally, researchers face severe "measurement limitations"—they don't measure one thing for all its properties but get "a lot of separate views of any one entity." Finally, even similar questions can't be automated in standard ways because the approach isn't "perfectly the same," requiring humans to "somehow operate through it."
**答：** Aviv Regev解释说生物学不仅仅是复杂——"它实际上非常庞大,非常非常庞大"。生物学中的所有数字都大到超出实验能力,"比地球上的细胞数量大,肯定比地球上的人口大,比宇宙中的星星大,比宇宙中的原子大"。除了规模之外,生物学是"多尺度的",需要在从"原子,然后分子,然后细胞,然后组织,然后患者,然后患者群体"的层面上思考。此外,研究人员面临严重的"测量限制"——他们不会测量一个事物的所有属性,而是获得"任何一个实体的许多不同视角"。最后,即使是类似的问题也无法以标准方式自动化,因为方法不是"完全相同的",需要人类"以某种方式去操作"。

### Topic 50: Why AI is Well-Suited for Biology's Challenges
为什么AI非常适合应对生物学的挑战
_[01:14:30]_

**Q:** How do AI's capabilities align with the fundamental challenges of biology?
**问：** AI的能力如何与生物学的基本挑战相匹配?

**A:** Regev systematically matches AI strengths to biology's challenges. For huge spaces with lower actual dimensionality, "AI is great at that." For multiscale problems with nonlinear transformations between scales, "AI has proven itself being really great at that." For integrating multiple views of the same thing—like "language and video, video and audio"—"AI is great at that." And for navigating through worlds "that is not exactly specified, although there's kind of a playbook, but it's not precise," AI agents are "actually great at that." She summarizes this alignment as "great, great, great, and great," emphasizing that combining these capabilities together "changes" the potential. However, she immediately cautions that "AI is not magic" and requires specific conditions to succeed.
**答：** Regev系统地将AI的优势与生物学的挑战相匹配。对于实际维度较低的巨大空间,"AI很擅长"。对于尺度之间存在非线性转换的多尺度问题,"AI已经证明自己真的很擅长"。对于整合同一事物的多个视角——比如"语言和视频,视频和音频"——"AI很擅长"。对于在"没有精确指定,虽然有一种操作手册,但不精确"的世界中导航,AI智能体"实际上很擅长"。她将这种匹配总结为"很好,很好,很好,还有很好",强调将这些能力结合在一起"会改变"潜力。然而,她立即警告说"AI不是魔法",需要特定条件才能成功。

### Topic 51: Lab-in-the-Loop: AI Requires Data, Feedback, and Iteration
Lab-in-the-Loop:AI需要数据、反馈和迭代
_[01:15:26]_

**Q:** What conditions must be met for AI to work effectively in biology, and what is the lab-in-the-loop concept?
**问：** AI要在生物学中有效工作必须满足什么条件?什么是lab-in-the-loop概念?

**A:** Beyond algorithms, models, and GPUs, AI "needs a lot of data and it needs really the ability to understand whether what it did means anything, and that means that it needs iterations." This leads to "lab in the loop or a clinic in the loop"—the faster you iterate in the clinic, the same principle applies. Regev emphasizes a crucial worldview shift: "Your data are the basis really for learning your models. Your models hold the answer, not the data, but the models hold the answer." Because the biological space is so enormous, "we're not going to measure all of it, but we are going to get to a general model." The model guides scientists to the next experimental step, allowing iteration at "the scale of biology that before was beyond us." She notes this "actually does actually move the needle" and "it's not the same as it used to be."
**答：** 除了算法、模型和GPU之外,AI"需要大量数据,并且真正需要能够理解它所做的事情是否有意义,这意味着它需要迭代"。这导致了"lab in the loop或clinic in the loop"的概念——在临床中迭代得越快,同样的原理就适用。Regev强调了一个关键的世界观转变:"你的数据实际上是学习模型的基础。你的模型持有答案,而不是数据,但模型持有答案。"因为生物学空间如此巨大,"我们不会测量所有的东西,但我们会得到一个通用模型"。模型引导科学家进行下一步实验,允许在"以前超出我们能力的生物学规模"上进行迭代。她指出这"实际上确实推动了进展",并且"与过去不一样了"。

### Topic 52: Current Limitations: The Long Tail Problem and Where AI Falls Short
当前的局限性:长尾问题以及AI的不足之处
_[01:16:31]_

**Q:** What are the current limitations and challenges in applying AI to biology and drug discovery?
**问：** 在将AI应用于生物学和药物发现方面,目前的局限性和挑战是什么?

**A:** Regev cautions that "it's not enough on its own today" because "biology and chemistry and so on, they have a huge long tail." People focus on areas "where data abound and models perform well," but even when AI produces promising starting points—she mentions an oncology example she calls "an alien" that "no human would have thought of"—"that was just the starting point." Understanding mechanism of action required "experimental science and thinking that AI cannot help you with right now" because "there's no data like that." These experiments are "small scale, they're very bespoke, very specific, very imaginative." AI doesn't take you "all the way," though combining AI starting points with human experimental science yields answers. The key advantage is working "in a comprehensive way"—"the answer is in the model, not in the data" whereas "before, it was only in the data." This operational shift "is also quite difficult."
**答：** Regev警告说"它今天本身还不够",因为"生物学和化学等等,它们有一个巨大的长尾"。人们关注"数据丰富且模型表现良好"的领域,但即使AI产生有希望的起点——她提到一个她称为"外星人"的肿瘤学例子,"没有人会想到"——"那只是起点"。理解作用机制需要"AI现在无法帮助你的实验科学和思考",因为"没有那样的数据"。这些实验是"小规模的,非常定制化的,非常具体的,非常富有想象力的"。AI不会带你走"全程",尽管将AI起点与人类实验科学相结合会产生答案。关键优势是以"全面的方式"工作——"答案在模型中,而不在数据中",而"以前,它只在数据中"。这种操作转变"也是相当困难的"。

### Topic 53: Bristol Myers Squibb's AI Vision: Transformation Over Acceleration
Bristol Myers Squibb的AI愿景:转型而非加速
_[01:17:54]_

**Q:** How is Bristol Myers Squibb approaching AI implementation, and what are their strategic bets?
**问：** Bristol Myers Squibb如何实施AI,他们的战略押注是什么?

**A:** Chris Boerner emphasizes that BMS doesn't want AI to "simply accelerate the current processes, but actually transform how you operate and do work." At a macro level, they believe "this technology is ultimately going to transform every piece of the value chain," and while "still in the early innings," they're seeing promise across all areas. BMS is making "three big bets." First, that "AI and machine learning can really help us identify those hidden patterns in biology" to drug previously undruggable targets. They've made progress: "all of our small molecules and a large percentage of our large molecules go through an AI screening and validation process before they ever get into the wet lab." Second, AI will "fundamentally change drug development" by reducing times, costs, and improving probability of success—they've set an internal target to "reduce cycle times by 30%" and are "well on our way" to likely beating that target.
**答：** Chris Boerner强调BMS不希望AI"只是加速当前流程,而是实际上转变你的运营和工作方式"。在宏观层面上,他们相信"这项技术最终将转变价值链的每一个环节",虽然"仍处于早期阶段",但他们在所有领域都看到了希望。BMS正在进行"三大押注"。首先,"AI和机器学习确实可以帮助我们识别生物学中的那些隐藏模式",以药物化以前无法成药的靶点。他们取得了进展:"我们所有的小分子和很大比例的大分子在进入湿实验室之前都要经过AI筛选和验证过程"。其次,AI将通过减少时间、成本和提高成功概率"从根本上改变药物开发"——他们设定了"将周期时间减少30%"的内部目标,并且"正在顺利实现"可能会超过这个目标。

### Topic 54: BMS's Broad Deployment: 30,000 Employees and Productivity Gains
BMS的广泛部署:30,000名员工和生产力提升
_[01:19:40]_

**Q:** What is the scale of AI deployment at BMS, and what productivity impacts are they seeing?
**问：** BMS的AI部署规模如何?他们看到了什么样的生产力影响?

**A:** BMS's third major bet is that AI "will create productivity tailwinds really across the organization." They were early adopters, rolling out "the first ChatGPT models in early January, February of 2023, very early on." Today, "over 30,000 employees have access to a whole suite of AI tools," generating "thousands and thousands of use cases." Boerner believes this will show "a 5 to 10% at a minimum increase in productivity across various parts of the organization." He acknowledges they're "in the early part of the journey, but it's a journey this industry ultimately needs to take" because "the business model of this industry will change over time, and we think AI will enable that." The ultimate goal is mission-driven: "make us better at delivering on our mission, which is bring more medicines to patients faster and ultimately change patient outcomes."
**答：** BMS的第三个主要押注是AI"将在整个组织中创造生产力顺风"。他们是早期采用者,在"2023年1月、2月初就推出了第一批ChatGPT模型,非常早"。今天,"超过30,000名员工可以访问一整套AI工具",产生了"成千上万的用例"。Boerner相信这将显示"各个部门的生产力至少提高5%到10%"。他承认他们"处于旅程的早期阶段,但这是这个行业最终需要进行的旅程",因为"这个行业的商业模式将随着时间的推移而改变,我们认为AI将促成这一点"。最终目标是使命驱动的:"让我们更好地实现我们的使命,即更快地为患者带来更多药物,并最终改变患者的结果"。

### Topic 55: Historical Context: 120 Years, Only 800-1,000 Medicines
历史背景:120年,仅有800-1,000种药物
_[01:20:50]_

**Q:** What does the historical track record of drug discovery reveal about the challenge, and what is Novartis's perspective on AI's potential?
**问：** 药物发现的历史记录揭示了什么挑战?Novartis对AI潜力的看法是什么?

**A:** Vas Narasimhan provides sobering historical perspective: "When you look at the 120 years of this sector, we've really only discovered around 800 to 1,000 medicines." Even more striking, "when you look at actual mechanisms of action that are notable, it's actually quite small." Despite being "an industry where we spend 150 to 200 billion a year in drug R&D amongst the bigger companies," the output has been limited to "only a handful of medicines." Narasimhan argues this demonstrates "how hard this is—unpacking billions of years of evolution." The "tools now that you guys are talking about here today can hopefully get us to another level," but he emphasizes the importance of "break[ing] down how that's going to work" rather than assuming AI will magically solve these challenges.
**答：** Vas Narasimhan提供了发人深省的历史视角:"当你看这个行业的120年时,我们实际上只发现了大约800到1,000种药物。"更引人注目的是,"当你看实际的显著作用机制时,实际上相当少"。尽管是"一个行业,大公司每年在药物研发上花费1500亿到2000亿美元",但产出仅限于"只有少数几种药物"。Narasimhan认为这表明"这有多困难——解开数十亿年的进化"。"你们今天在这里谈论的工具有希望让我们达到另一个水平",但他强调"分解这将如何工作"的重要性,而不是假设AI会神奇地解决这些挑战。

### Topic 56: Three Categories of Latency in Drug Development
药物开发中的三类延迟
_[01:21:45]_

**Q:** How does Novartis break down the sources of delay in drug development, and which can AI address?
**问：** Novartis如何分解药物开发中的延迟来源?AI可以解决哪些问题?

**A:** Narasimhan categorizes drug development delays into three types: "information latency, operational latency, and biological latency." Information and operational latency together account for "about 40%" of development time. He believes "with the tools you saw today, we can bring information latency down almost to zero"—information will be "at scientists' fingertips." Operational latency, which involves "organizing trials, organizing experiments, getting all of the work to happen in a large organization," can "probably reduce significantly." However, "biological latency we're stuck with"—this is "actually having to run the experiment in an animal model and a cellular model or in humans, and that's about 60% of this timeline." This fundamental constraint means AI cannot eliminate all delays, only compress the addressable portions.
**答：** Narasimhan将药物开发延迟分为三类:"信息延迟、操作延迟和生物学延迟"。信息延迟和操作延迟合计占开发时间的"约40%"。他相信"有了你今天看到的工具,我们可以将信息延迟降低到几乎为零"——信息将"在科学家的指尖"。操作延迟涉及"组织试验、组织实验、让所有工作在大型组织中发生","可能会显著减少"。然而,"生物学延迟我们无法摆脱"——这是"实际上必须在动物模型和细胞模型或人类中运行实验,这占了约60%的时间线"。这一基本约束意味着AI无法消除所有延迟,只能压缩可解决的部分。

### Topic 57: AI's Potential Impact on Drug Development Success Rates
AI对药物开发成功率的潜在影响
_[01:23:14]_

**Q:** How much can AI realistically improve drug development success rates and timelines?
**问：** AI能在多大程度上现实地提升药物开发的成功率和时间线?

**A:** Speaker A presents a measured assessment of AI's potential impact across different phases of drug development. While acknowledging that identifying "the right drug target for this disease" remains fundamentally difficult, they outline three areas where models can help: improving drug design through historical knowledge, enabling better patient selection for the right indication, and optimizing manufacturability. The speaker suggests AI could potentially double success rates from "8% to 16%" and reduce development time from "12 years to seven years." They emphasize that while these improvements "sound like small moves," when "compounded over the size of our industry pipelines," the public health impact would be "massive," potentially enabling treatment of more diseases and drugging previously "undruggable targets."
**答：** Speaker A对AI在药物开发不同阶段的潜在影响给出了审慎的评估。虽然承认识别"正确的药物靶点"仍然从根本上很困难,但他们指出了模型可以帮助的三个领域:通过历史知识改进药物设计、实现更好的患者选择以找到正确的适应症、优化可制造性。Speaker A认为AI可能将成功率从8%提升到16%,将开发时间从12年缩短到7年。他们强调虽然这些改进"听起来是小幅度的变化",但当"在整个行业管线规模上复合计算"时,对公共健康的影响将是"巨大的",可能使更多疾病得到治疗,以前"不可成药"的靶点被成药。

### Topic 58: GLP-1 Development as Case Study for AI Acceleration
GLP-1开发作为AI加速的案例研究
_[01:24:20]_

**Q:** How does the GLP-1/semaglutide development example illustrate AI's potential for acceleration?
**问：** GLP-1/semaglutide的开发案例如何说明AI的加速潜力?

**A:** Speaker B references a conversation with Lotus about the semaglutide development process, highlighting that it took "four years and several thousand compounds" just to achieve proper stability for GLP-1. They propose that even a "25%" improvement in efficiency—reducing the testing from "a thousand compounds in a year" or potentially more aggressively—would represent "an incredible acceleration and impact." This concrete historical example demonstrates how AI-driven compound screening and optimization could meaningfully compress timelines in drug development, even without achieving perfection. Speaker A agrees that "all the steps of this process are very difficult," reinforcing the value of incremental improvements.
**答：** Speaker B引用了与Lotus的对话,提到GLP-1/semaglutide的开发过程,强调仅仅为了获得GLP-1的适当稳定性就花费了"四年时间和数千个化合物"。他们提出即使是25%的效率提升——将测试从"一年一千个化合物"减少,甚至更激进——都将代表"令人难以置信的加速和影响"。这个具体的历史案例展示了AI驱动的化合物筛选和优化如何能够有意义地压缩药物开发的时间线,即使没有达到完美。Speaker A同意"这个过程的所有步骤都非常困难",强化了渐进式改进的价值。

### Topic 59: No Silver Bullet: Incremental Gains Across All Steps
没有银弹:所有步骤的渐进式收益
_[01:24:44]_

**Q:** Why is an 'all of the above' approach necessary rather than focusing on a single breakthrough?
**问：** 为什么需要"全方位"的方法,而不是专注于单一突破?

**A:** Speaker A emphasizes that drug development has no single solution: "there is not one that if you just solve that one, everything else would be solved." Even with a "perfect oracle for targets," teams would still need to drug them, find the right patients, and execute clinical trials. The advantage of the "all of the above approach" is that "any benefit that you get is a real benefit as it compounds." They highlight that models enable more end-to-end thinking, allowing teams to "think about safety way earlier" by virtualizing steps and considering biological latency. This holistic capability means teams can explore a broader range of targets beyond the "very small sliver" currently examined due to experimental tracking difficulties. When "biological latency can start becoming virtualized," biology becomes "a lot more accessible in a much broader sense."
**答：** Speaker A强调药物开发没有单一解决方案:"没有哪一个问题,如果你只解决了那一个,其他所有问题就都解决了"。即使有"完美的靶点预言机",团队仍然需要对它们成药、找到合适的患者、执行临床试验。"全方位方法"的优势在于"你获得的任何收益都是真实的收益,因为它会复合"。他们强调模型使端到端思考成为可能,允许团队通过虚拟化步骤和考虑生物学潜伏期来"更早地考虑安全性"。这种整体能力意味着团队可以探索更广泛的靶点,超越目前由于实验追踪困难而只能检查的"非常小的一部分"。当"生物学潜伏期可以开始被虚拟化"时,生物学在"更广泛的意义上变得更容易获取"。

### Topic 60: Managing Expectations: Progress vs. Curing Cancer
管理期望:进步 vs. 治愈癌症
_[01:25:53]_

**Q:** How should the industry manage expectations about what AI can accomplish in drug development?
**问：** 行业应该如何管理对AI在药物开发中能取得什么成就的期望?

**A:** Speaker B emphasizes the need for realistic expectation-setting despite the "lot of possibility and opportunity" with AI technology. They caution against overpromising outcomes "that we simply can't deliver on," specifically addressing grandiose claims like "we're going to cure cancer in our lifetime." Instead, the appropriate framing is "we're going to make a lot of progress on cancer in our lifetime, but let's—we don't want to get over our skis." This measured approach acknowledges genuine potential while avoiding the hype that could undermine credibility when results fall short of unrealistic promises. The speaker positions this caution in the context of their organization starting to pursue preclinical work internally to "learn these lessons" with "skin in the game."
**答：** Speaker B强调尽管AI技术有"很多可能性和机会",但需要设定现实的期望。他们警告不要过度承诺"我们根本无法兑现"的成果,特别针对诸如"我们将在有生之年治愈癌症"这样宏大的声明。相反,适当的表述是"我们将在有生之年在癌症方面取得很大进展,但让我们——我们不想过于乐观"。这种审慎的方法承认真正的潜力,同时避免了当结果达不到不切实际的承诺时可能损害可信度的炒作。Speaker B在他们组织开始内部追求临床前工作以"学习这些经验教训"并"真正投入"的背景下提出这一警告。

### Topic 61: Lessons from Scaling AI: Beyond Narrow Use Cases
扩展AI的经验教训:超越狭窄的用例
_[01:26:44]_

**Q:** What has not worked when trying to scale AI in pharmaceutical organizations?
**问：** 在制药组织中尝试扩展AI时,什么方法没有奏效?

**A:** Speaker B describes their initial "let a thousand flowers bloom" approach with thousands of use cases, allowing people to use AI technology freely to improve productivity. While this worked initially, they discovered the use cases "tended to be quite narrow and were very difficult to scale." The fundamental problem was that tools were being "put on top of processes, but you're not changing the process itself." They provide a concrete example: AI could forecast business better than "battalions of people" doing forecasting two years ago, but couldn't scale because "changing the forecast and planning process" requires coordinating finance, commercial, manufacturing, and more—"not a business analytics exercise." The solution required supplementing "bottom-up innovation" with a "very robust and rigorous top-down approach" through an AI accelerator, with small teams given "six to eight weeks to prove out the concept" before scaling, resulting in 40-50 incubator projects and 30 ongoing scaled efforts.
**答：** Speaker B描述了他们最初的"百花齐放"方法,有数千个用例,允许人们自由使用AI技术来提高生产力。虽然这最初奏效,但他们发现用例"往往非常狭窄且很难扩展"。根本问题是工具被"放在流程之上,但你并没有改变流程本身"。他们提供了一个具体例子:两年前AI可以比"大批人员"更好地预测业务,但无法扩展,因为"改变预测和规划流程"需要协调财务、商业、制造等部门——"这不是一个业务分析练习"。解决方案需要用"非常稳健和严格的自上而下方法"来补充"自下而上的创新",通过AI加速器,小团队获得"六到八周来证明概念",然后再扩展,最终产生了40-50个孵化器项目和30个正在进行的规模化工作。

### Topic 62: Research Side Challenges: From Data Generation to Model Focus
研究端挑战:从数据生成到模型聚焦
_[01:28:59]_

**Q:** What unexpected challenges emerged when implementing the 'lab in the loop' hypothesis on the research side?
**问：** 在研究端实施"实验室在环"假设时出现了哪些意想不到的挑战?

**A:** Speaker A explains their research approach started in 2020 with a specific "lab in the loop" hypothesis rather than letting many flowers bloom, requiring investment in data generation capacity suited for AI needs. However, an unexpected problem emerged: once data were generated, people—including computational scientists—"forgot that the goal was to end up with an AI model" and started examining the data themselves because "there's like real results there." They had to pivot to "foundational datasets for foundation models" as an initiative to remind people the goal was model training. Initially met with resistance ("why are you making me do this?"), within months people shifted to asking "why is the foundation model not trained yet?" as they realized "how much more is in a model than in data." A second challenge arose around individual projects: scientists naturally want "that molecule to succeed" for patients, creating tension with the broader goal of making "the model general" and "learning lessons." The question became "why should I spend synthesis money just to make the model better?" when projects need to move forward, requiring careful balance to ensure all projects advance while still generalizing learnings.
**答：** Speaker A解释他们的研究方法从2020年开始就有一个具体的"实验室在环"假设,而不是百花齐放,需要投资适合AI需求的数据生成能力。然而,出现了一个意想不到的问题:一旦数据生成,人们——包括计算科学家——"忘记了目标是最终得到一个AI模型",开始自己检查数据,因为"那里有真实的结果"。他们不得不转向"为foundation模型建立基础数据集"作为一个倡议,提醒人们目标是模型训练。最初遇到阻力("为什么要让我做这个?"),但几个月内人们转而询问"foundation模型为什么还没训练好?",因为他们意识到"模型中的内容比数据多得多"。第二个挑战围绕单个项目出现:科学家自然希望"那个分子成功"以帮助患者,这与"使模型通用化"和"学习经验"的更广泛目标产生了紧张关系。问题变成了"当项目需要推进时,为什么我要花合成经费只是为了让模型更好?",需要仔细平衡以确保所有项目都推进的同时仍然能泛化学习成果。

### Topic 63: Decision-Making Criteria: Kill Early, Persist Late
决策标准:早期淘汰、后期坚持
_[01:31:11]_

**Q:** How should decision-making criteria change as AI dramatically increases molecule generation capacity?
**问：** 随着AI大幅提高分子生成能力,决策标准应该如何改变?

**A:** Speaker B emphasizes that "all the math shows" the importance of setting "high bars" before phase 2A and being "willing to let things fail." With AI's ability to "generate molecules going to increase dramatically," rigorous "decision-making criteria is going to be really important." The strategy is to "set really good experiments whether it's in pre-clinical or in early clinical to really kill early based on hard data." However, the approach flips in late stage: "once you find an active drug" that hits the target with pharmacodynamic effect, "you need to find its use case in late stage and that's where you don't want to give up too soon." They note industry has "many" examples of eventually finding "the right use case in late stage for a drug" after years. This bifurcated approach—rigorous early filtering paired with persistent late-stage exploration—becomes essential as AI expands the pipeline.
**答：** Speaker B强调"所有的数学分析表明"在2A期之前设定"高标准"和"愿意让事情失败"的重要性。随着AI"生成分子的能力将大幅增加",严格的"决策标准将变得非常重要"。策略是"设置真正好的实验,无论是在临床前还是早期临床,基于确凿数据真正做到早期淘汰"。然而,在后期阶段方法会翻转:"一旦你找到一个活性药物"击中靶点并具有药效学效应,"你需要在后期找到它的用例,这时你不想过早放弃"。他们指出行业有"许多"例子,经过多年最终为药物找到"后期的正确用例"。这种两分法——严格的早期筛选与坚持的后期探索相结合——随着AI扩展管线变得至关重要。

### Topic 64: Managing Portfolio Explosion: 130% Growth Without More Biologists
管理组合爆炸:在没有增加生物学家的情况下130%的增长
_[01:32:34]_

**Q:** How has AI's impact on portfolio volume changed organizational processes?
**问：** AI对组合容量的影响如何改变了组织流程?

**A:** Speaker A reveals that "entries into the research portfolio more than doubled in two years"—specifically "130% or so"—without adding "any more biologists during that process." This dramatic expansion came not just from AI but also from improved "data and data generation capacity." The explosion in throughput forced changes to "how you manage the process of vetting and entering" because "the old process was just not made for this type of approach." Looking industry-wide, Speaker A anticipates tools will enable similar volume increases across the sector, raising the total system volume substantially. However, they warn of "the risk that people will do the same thing again and again and again, which is not actually a very desirable outcome." This concern about redundant work rather than genuine innovation highlights the need to rethink decision-making processes: "there's a lot of room for thinking about how we make our decisions in science."
**答：** Speaker A透露"进入研究组合的项目在两年内增长了一倍多"——具体来说"大约130%"——而在此过程中没有"增加任何生物学家"。这种戏剧性的扩张不仅来自AI,还来自改进的"数据和数据生成能力"。吞吐量的爆炸式增长迫使改变"如何管理审核和进入的流程",因为"旧流程根本不适合这种类型的方法"。从整个行业来看,Speaker A预计工具将在整个行业实现类似的容量增长,大幅提高总系统容量。然而,他们警告"人们一遍又一遍做同样事情的风险,这实际上不是一个非常理想的结果"。这种对冗余工作而非真正创新的担忧突显了重新思考决策过程的必要性:"在科学中我们如何做决策,还有很大的思考空间"。

### Topic 65: The Risk of Mediocrity: Autopilot and Loss of Rigor
平庸的风险:自动驾驶与严谨性的丧失
_[01:33:34]_

**Q:** What is the biggest risk when employees use AI tools that can create, edit, and take action on their behalf?
**问：** 当员工使用可以创建、编辑和代表他们采取行动的AI工具时,最大的风险是什么?

**A:** Speaker B identifies mediocrity as "probably the biggest thing I worry about," more than malfeasance or other concerns. The risk is that "employees and society in general goes on autopilot," where people with tools that "can create content, edit content, summarize, take action on your behalf" may "phone it in" and "stop checking my homework or interrogating the output." In pharma, this is "existential" because the industry hires "the best and the brightest" specifically because "the problems that we have to deal with are the most complex problems." The moment people stop "bringing our intelligence and our creativity and our discipline and the approach that we take towards science to the table, that's a big problem." Critically, Speaker B argues this cannot be solved through "governance or guardrails"—instead, "you solve that with a culture of how you engage this technology," which is "much harder to do" and requires thoughtful approach.
**答：** Speaker B认为平庸是"可能是我最担心的事情",超过恶意使用或其他问题。风险在于"员工和整个社会进入自动驾驶状态",人们拥有"可以创建内容、编辑内容、总结、代表你采取行动"的工具,可能会"敷衍了事"并"停止检查我的工作或质疑输出"。在制药行业,这是"生死攸关的",因为该行业雇用"最优秀和最聪明的人"正是因为"我们必须处理的问题是最复杂的问题"。当人们停止"将我们的智慧、创造力、纪律以及我们对待科学的方法带到桌面上,这就是一个大问题"。关键的是,Speaker B认为这不能通过"治理或护栏"来解决——相反,"你通过如何使用这项技术的文化来解决这个问题",这"要困难得多"并需要深思熟虑的方法。

### Topic 66: The Narcissus Problem: AI Evaluation and Novel Discovery
Narcissus问题:AI评估与新颖性发现
_[01:37:24]_

**Q:** How can AI systems avoid the trap of self-reinforcement and maintain their ability to discover truly novel insights?
**问：** AI系统如何避免自我强化的陷阱,保持发现真正新颖洞见的能力?

**A:** Speaker A identifies a critical risk in AI development where models excel at existing evaluations but those "evals are actually where it gets an edge, but they're not the right evals for novelty." The solution involves using "the real world, which is the lab" to escape current limitations. Without this approach, there's a genuine danger of systems "regurgitating the same thing again and again," which A compares to the Greek myth of Narcissus—becoming fixated on one's own reflection. This represents a fundamental challenge in ensuring AI systems can break new ground rather than simply optimizing within existing paradigms.
**答：** Speaker A指出AI发展中的一个关键风险:模型在现有评估中表现出色,但这些评估"并不适合衡量新颖性"。解决方案是利用"真实世界,也就是实验室"来突破当前限制。如果不采取这种方法,系统很可能会"一遍又一遍地重复同样的东西",A将此比作希腊神话中的Narcissus——沉迷于自己的倒影。这代表了确保AI系统能够开辟新领域而不是仅在现有范式内优化的根本挑战。

### Topic 67: Dual Policy Challenge: AI Regulation and Drug Development Modernization
双重政策挑战:AI监管与药物开发现代化
_[01:38:00]_

**Q:** What parallel regulatory changes are needed to fully leverage AI in drug development?
**问：** 为了充分利用AI进行药物开发,需要哪些并行的监管改革?

**A:** Speaker A argues for a two-pronged regulatory approach. First, AI systems themselves need regulation to "mitigate a lot of the threats," with Anthropic and Dario positioned as leaders in this area. Second, and equally important, drug regulatory systems must be "modernized" to capture AI's full potential. This includes streamlining animal models through better preclinical safety prediction, implementing "synthetic control arms" that have historically faced FDA acceptance challenges, and improving dosing models to reduce dose range-finding studies. The core insight is that AI capabilities alone aren't sufficient—regulatory frameworks must evolve in parallel to unlock "the full benefits of speed and probability of success."
**答：** Speaker A主张采取双管齐下的监管策略。首先,AI系统本身需要监管来"减轻许多威胁",Anthropic和Dario在这方面处于领先地位。其次,同样重要的是,药品监管系统必须"现代化"以发挥AI的全部潜力。这包括通过更好的临床前安全预测来简化动物模型,实施历史上面临FDA接受度挑战的"合成对照组",以及改进剂量模型以减少剂量范围探索研究。核心洞见是:仅有AI能力是不够的,监管框架必须同步演进才能释放"速度和成功概率的全部优势"。

### Topic 68: One-Year Accountability: Demonstrating Real-World Drug R&D Impact
一年问责:展示真实世界的药物研发影响
_[01:39:00]_

**Q:** What does success look like in one year, and what failure mode should be avoided?
**问：** 一年后的成功是什么样子,应该避免什么失败模式?

**A:** When asked what he wants to be held accountable for, Speaker A (Moss) commits to "demonstrate that these technologies are starting to have an impact on the speed of our drug R&D process." He acknowledges the industry has made "a lot of bold proclamations" and now must deliver "real results" for patients. His hope for being proven wrong centers on regulatory dynamics—specifically that "it's not going to take a crisis to actually get good regulation to happen." This reflects concern that the current path may require a catastrophic event to trigger appropriate AI regulation, which "would be a shame" given the opportunity for proactive policy-making.
**答：** 当被问及希望对什么负责时,Speaker A(Moss)承诺要"展示这些技术开始对我们药物研发流程的速度产生影响"。他承认行业已经做出了"许多大胆的宣言",现在必须为患者提供"真实的结果"。他希望被证明错误的地方集中在监管动态上——特别是"不需要危机才能实现良好的监管"。这反映了他的担忧:当前的路径可能需要灾难性事件才能触发适当的AI监管,考虑到主动制定政策的机会,这"将是一种遗憾"。

### Topic 69: End-to-End Integration: Moving Beyond Point Solutions
端到端整合:超越单点解决方案
_[01:39:54]_

**Q:** What architectural shift in AI drug development would represent meaningful progress?
**问：** AI药物开发中什么样的架构转变代表着有意义的进展?

**A:** Speaker C identifies a fragmentation problem in current AI drug development where solutions exist as disconnected point tools: "here's my solution for target discovery and here's my solution for a small molecule and for this step in the process and for that step." While acknowledging that "many of the pieces, we already have had them in place now for a while, including real deployment and use," the aspiration is to see "more of the end-to-end" where "the model looks further down the road and the pieces are more connected together." This represents a shift from modular, stage-specific tools to integrated systems that can reason across the entire drug development pipeline.
**答：** Speaker C指出当前AI药物开发中存在碎片化问题,解决方案以断开的单点工具形式存在:"这是我的靶点发现解决方案,这是我的小分子解决方案,这是流程中这个步骤的方案,那是那个步骤的方案"。虽然承认"许多部分我们已经部署使用了一段时间,包括真实部署和使用",但期望看到"更多端到端"的整合,其中"模型能看得更远,各个部分更加连接在一起"。这代表了从模块化的、特定阶段的工具向能够跨整个药物开发管线推理的集成系统的转变。

### Topic 70: The Joy Factor: Combating Rising Stress Through AI Empowerment
快乐因素:通过AI赋能对抗不断上升的压力
_[01:40:25]_

**Q:** What human dimension of AI adoption matters most for achieving good outcomes?
**问：** AI采用的哪个人性维度对实现良好结果最重要?

**A:** Speaker C expresses concern about escalating stress levels "in companies, in other scientific communities like in academia, in society as a whole," a trend they consider "reasonable to assume" will continue rising. Their hope for being proven wrong focuses on a psychological shift: seeing "people's joy in the process, their real enjoyment from the fact that something new and exciting is in their hands." The critical distinction is between using AI tools "in new ways, not just in old ways," and experiencing "the arc of that positivity." This emotional and creative engagement is framed as "extremely important for getting a good outcome in the end," suggesting that technical capabilities without human flourishing would constitute a failed transformation.
**答：** Speaker C对"公司、学术界等科学界以及整个社会"不断上升的压力水平表示担忧,认为这一趋势"合理地假设"会继续上升。他们希望被证明错误的地方集中在心理转变上:看到"人们在过程中的快乐,他们对手中有新奇有趣的东西这一事实的真实享受"。关键区别在于以"新方式而非仅以旧方式"使用AI工具,并体验"积极性的弧线"。这种情感和创造性的参与被认为"对最终获得良好结果极其重要",表明没有人类繁荣的技术能力将构成失败的转型。

### Topic 71: ROI Challenge: From Investment Enthusiasm to Demonstrable Returns
ROI挑战:从投资热情到可证明的回报
_[01:41:10]_

**Q:** How can organizations demonstrate concrete returns on their AI investments, and what happened when one company actually measured it?
**问：** 组织如何证明其AI投资的具体回报,当一家公司实际测量时发生了什么?

**A:** Speaker B describes a shift in investor sentiment from "begging us to make more investments" to asking "what's the return?" To illustrate the measurement challenge, they share a concrete case: a Phase 1 program transitioning to Phase 2 encountered "a gnarly side effect." Because the company had "the data curated" with "AI tools sitting on top of it," they identified within one week that "a very small change that had been made in manufacturing" was the culprit. This intervention "probably saved the program and certainly saved the program 6 to 12 months of time." Yet quantifying this value remains "tricky"—how do you define the ROI of crisis prevention? The speaker emphasizes "we got to figure it out" because without clear ROI frameworks, organizations will struggle to justify continued investment.
**答：** Speaker B描述了投资者情绪的转变,从"恳求我们进行更多投资"到询问"回报是什么?"为了说明测量挑战,他们分享了一个具体案例:一个从Phase 1过渡到Phase 2的项目遇到了"棘手的副作用"。因为公司"整理好了数据"并"在上面部署了AI工具",他们在一周内识别出"制造过程中做的一个非常小的改变"是罪魁祸首。这一干预"可能挽救了该项目,确实为项目节省了6到12个月的时间"。然而量化这一价值仍然"棘手"——你如何定义危机预防的ROI?发言人强调"我们必须弄清楚",因为没有清晰的ROI框架,组织将难以证明持续投资的合理性。

### Topic 72: Rationing Risk: Strategic AI Deployment at Scale
配给风险:大规模战略性AI部署
_[01:42:04]_

**Q:** What organizational threat emerges from poorly managed AI tool deployment, and how should companies respond?
**问：** AI工具部署管理不善会带来什么组织威胁,公司应该如何应对?

**A:** Speaker A warns that without clear ROI demonstration, "large organizations are going to become put under pressure to ration the use of this technology." This concern drove their Anthropic partnership strategy: they've "embedded Anthropic across all of our work processes" with the explicit goal of "apply[ing] the right tools to the right problems." The key is avoiding waste—specifically, "not using the most advanced tools to summarize people's emails," which represents an inefficient use of expensive capabilities. The rationing threat creates urgency around "get[ting] more sophisticated around that concept generally," meaning organizations must develop frameworks for matching tool sophistication to problem complexity. Without this discipline, cost pressures will force indiscriminate cuts rather than strategic optimization.
**答：** Speaker A警告说,如果没有清晰的ROI证明,"大型组织将面临配给这项技术使用的压力"。这一担忧推动了他们的Anthropic合作战略:他们已经"在所有工作流程中嵌入了Anthropic",明确目标是"将正确的工具应用于正确的问题"。关键是避免浪费——特别是"不用最先进的工具来总结人们的电子邮件",这代表了对昂贵能力的低效使用。配给威胁使"在这个概念上变得更加成熟"变得紧迫,意味着组织必须开发框架来匹配工具复杂度与问题复杂度。没有这种纪律,成本压力将迫使无差别削减而非战略优化。

### Topic 73: Scientific Rigor and Honest Deployment of AI in Research
AI在研究中的科学严谨性与诚实部署
_[01:42:43]_

**Q:** What scientific principle should guide AI deployment in drug development?
**问：** 什么科学原则应该指导AI在药物开发中的部署?

**A:** Speaker B, identifying as a scientist, emphasizes the imperative to "be both rigorous and really honest with these new technologies and to think about how we're deploying them." Drawing on their experience running a research group, they invoke a foundational principle: "the only bad experiment was the one that we didn't carefully analyze." This framework is applied to evaluate the panel discussion, acknowledging that "AI and Claude is touching all parts of science and drug development as well as the business," while maintaining honest assessment that "there's still plenty of room to improve" and "we're at early stages." Critically, they note "the impacts are real, but the upside is still ahead of us," positioning current achievements as genuine but preliminary. The emphasis on continued experimentation "going forward both on model capabilities but also doing these experiments and how we get them to patients" reflects commitment to iterative, evidence-based development.
**答：** Speaker B以科学家的身份强调,必须"对这些新技术保持严谨和真正诚实,并思考我们如何部署它们"。借鉴他们领导研究小组的经验,他们援引了一个基本原则:"唯一糟糕的实验是我们没有仔细分析的实验"。这一框架被应用于评估小组讨论,承认"AI和Claude正在触及科学和药物开发以及业务的所有部分",同时保持诚实评估"仍有很大改进空间"和"我们处于早期阶段"。关键的是,他们指出"影响是真实的,但优势仍在前方",将当前成就定位为真实但初步的。对"在模型能力以及进行这些实验并如何将它们交付给患者方面"持续实验的强调,反映了对迭代、基于证据的开发的承诺。

### Topic 74: Closing Transition and Event Conclusion
结束过渡与活动总结
_[01:43:32]_

**Q:** How did the panel discussion conclude and transition to the next speaker?
**问：** 小组讨论如何结束并过渡到下一位发言人?

**A:** The panel discussion concluded with mutual thanks between the moderator and panelists, followed by a brief transitional moment. Speaker B thanked the panel, and there was a pause indicated by timestamp markers showing a slight gap in conversation. The moderator then formally introduced the next segment: "Please welcome back to the stage Zoubin Ghahramani," signaling the return of a speaker who had apparently presented earlier in the event. This transition marks the shift from the panel discussion format focused on drug development and AI implementation to what would be the final presentation of the session.
**答：** 小组讨论以主持人和嘉宾之间的相互感谢结束,随后是短暂的过渡时刻。Speaker B感谢了小组,时间戳标记显示对话中有轻微间隙的停顿。然后主持人正式介绍了下一个环节:"请欢迎Zoubin Ghahramani重返舞台",表明之前在活动中演讲过的发言人将返场。这一过渡标志着从关注药物开发和AI实施的小组讨论形式转向会议的最终演讲。

### Topic 75: Event Framing: From Assistance to Autonomy in Life Sciences
活动定位:从辅助到生命科学中的自主性
_[01:43:55]_

**Q:** How did the event's messaging evolve, and what evidence was presented to support the shift?
**问：** 活动的信息传递如何演变,提供了什么证据来支持这种转变?

**A:** Speaker C (likely Zoubin Ghahramani) frames the event's narrative arc by contrasting messaging evolution: "Six months ago, we told you that Claude could help with the digital work of life sciences R&D. This morning, I said something different: that Claude could run that work." This represents a significant capability claim escalation from assistance to autonomous execution. The speaker then validates this claim by reviewing the evidence presented: "Lattice and Dario walked us through their unvarnished views" on compression dynamics and challenges; "Eric and Alec walked you through the basis of our claim" showing "models that meaningfully improve in biology with every release" and "a workbench that runs the analysis your scientists actually need"—specifically systems that "run pipelines, figures, every step reproducible and traceable." This framing positions the entire event as building a systematic case for AI autonomy.
**答：** Speaker C(可能是Zoubin Ghahramani)通过对比信息演变来构建活动叙事:"六个月前,我们告诉你Claude可以帮助处理生命科学研发的数字工作。今天早上,我说了不同的话:Claude可以运行那些工作。"这代表了从辅助到自主执行的重大能力声明升级。然后发言人通过回顾所展示的证据来验证这一声明:"Lattice和Dario向我们展示了他们对压缩动态和挑战的坦率看法";"Eric和Alec向你们展示了我们声明的基础",表明"模型在生物学方面随着每次发布都有显著改进"以及"一个运行你的科学家实际需要的分析的工作台"——特别是"运行管线、图表,每一步都可重现和可追溯"的系统。这一框架将整个活动定位为为AI自主性构建系统性论据。

### Topic 76: Industry Partnership Showcase: BMS, Genentech, and Artis Transformation
行业合作展示:BMS、Genentech和Artis的转型
_[01:44:38]_

**Q:** How were the industry partners' presentations characterized in the event summary?
**问：** 活动总结中如何描述行业合作伙伴的演讲?

**A:** Speaker C summarizes the industry partner presentations as demonstrating how "Voss, Aviv, and Chris just walked us through how they're charting the path of AI transformation at BMS, Genentech, and Artis." The framing acknowledges these three pharmaceutical and biotech organizations as actively navigating—"charting the path"—rather than having completed their transformations, emphasizing the ongoing, exploratory nature of AI adoption in drug development. The speaker then positions this collective evidence as foundational but incomplete: "At the end of the day, this is just the beginning." This sets up the argument that building "an intelligence platform where scientific discovery truly happens is going to be a collaboration with the industry and our partners represented here," explicitly framing AI development in life sciences as requiring deep partnership rather than vendor-customer relationships.
**答：** Speaker C总结行业合作伙伴的演讲展示了"Voss、Aviv和Chris刚刚向我们展示了他们如何在BMS、Genentech和Artis规划AI转型之路"。这一框架承认这三家制药和生物技术组织正在积极导航——"规划路径"——而不是已经完成转型,强调了药物开发中AI采用的持续性、探索性。然后发言人将这些集体证据定位为基础但不完整:"归根结底,这只是一个开始。"这为论点奠定了基础:构建"一个真正发生科学发现的智能平台将是与行业和我们在这里代表的合作伙伴的协作",明确将生命科学中的AI开发框定为需要深度合作而非供应商-客户关系。

---

## Vocabulary (CEFR B2+)

### go-to-market  /ˈɡoʊ tuː ˈmɑːrkɪt/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** relating to the strategy and tactics used to bring a product to customers and achieve competitive advantage  
**CN:** 市场推广的，上市策略的

**Original examples:**
- [01:14] Please welcome head of **go-to-market** for healthcare and life sciences at Anthropic, Zubair Jandali.  
  请欢迎Anthropic医疗保健和生命科学市场推广负责人Zubair Jandali。

**Extra example:**
- Our **go-to-market** strategy focuses on enterprise customers first.  
  我们的市场推广策略首先聚焦于企业客户。

### claim  /kleɪm/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 3

**EN:** a statement that something is true, especially one that is disputed or in doubt  
**CN:** 声称，断言

**Original examples:**
- [01:57] Six months ago on this stage, we made a **claim** that Claude could help with the work of life sciences R&D.  
  六个月前在这个舞台上，我们做出了一个声称，即Claude可以帮助生命科学研发工作。
- [02:06] Today, we're going to build on that **claim** with a new one.  
  今天，我们将在那个声称的基础上提出一个新的。
- [02:06] The **claim** is this: Claude can run the work, not help with it, not accelerate it, but run it.  
  这个声称是：Claude可以运行这项工作，不是帮助它，不是加速它，而是运行它。

**Extra example:**
- The CEO's **claim** that AI would replace human judgment was met with skepticism.  
  CEO关于AI将取代人类判断的声称遭到了质疑。

### irreversibly  /ˌɪrɪˈvɜːrsəbli/
**CEFR:** C1 | **Part of speech:** adv. | **Occurrences:** 1

**EN:** in a way that cannot be undone or altered  
**CN:** 不可逆转地

**Original examples:**
- [02:26] Coding has **irreversibly** changed.  
  编程已经不可逆转地改变了。

**Extra example:**
- Climate change has **irreversibly** altered some ecosystems.  
  气候变化已经不可逆转地改变了一些生态系统。

### stretch  /stretʃ/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a continuous period of time  
**CN:** 一段时间

**Original examples:**
- [02:37] Once AI could sit inside that loop, the loop could keep turning for longer and longer **stretches** of time before an engineer had to step back in.  
  一旦AI可以进入那个循环，这个循环就可以在工程师需要重新介入之前持续越来越长的一段时间。

**Extra example:**
- The team worked for a long **stretch** without taking breaks.  
  团队连续工作了很长一段时间而没有休息。

### stall  /stɔːl/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to stop making progress or stop working effectively  
**CN:** 停滞，陷入停顿

**Original examples:**
- [03:07] Analysis happens at a keyboard, and that is where the loop **stalls**.  
  分析发生在键盘前，而那正是循环停滞的地方。

**Extra example:**
- Negotiations **stalled** when neither side would compromise.  
  当双方都不愿妥协时，谈判陷入了停顿。

### toil  /tɔɪl/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** hard, continuous, and exhausting work that is tedious rather than rewarding  
**CN:** 苦工，繁重乏味的工作

**Original examples:**
- [03:19] Those weeks aren't science. They're the **toil** you endure to get to the science.  
  那些星期不是科学。它们是你为了做科学而忍受的苦工。
- [03:28] It's our belief that that **toil** is collapsing, and as it does, your scientists will get back more of what they trained for: time at the question.  
  我们相信那种苦工正在消失，随着它的消失，你们的科学家将获得更多他们接受训练的目的：在问题上的时间。

**Extra example:**
- Much of software development involves tedious **toil** like manual testing and deployment.  
  很多软件开发工作涉及繁重乏味的工作，比如手动测试和部署。

### endure  /ɪnˈdʊr/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to suffer something painful or difficult patiently  
**CN:** 忍受，承受

**Original examples:**
- [03:19] Those weeks aren't science. They're the toil you **endure** to get to the science.  
  那些星期不是科学。它们是你为了做科学而忍受的苦工。

**Extra example:**
- Startups must **endure** years of uncertainty before achieving success.  
  初创公司在取得成功之前必须忍受多年的不确定性。

### collapse  /kəˈlæps/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to fall down or fail suddenly and completely  
**CN:** 崩溃，瓦解，消失

**Original examples:**
- [03:28] It's our belief that that toil is **collapsing**, and as it does, your scientists will get back more of what they trained for: time at the question.  
  我们相信那种苦工正在消失，随着它的消失，你们的科学家将获得更多他们接受训练的目的：在问题上的时间。

**Extra example:**
- The traditional retail model began to **collapse** with the rise of e-commerce.  
  随着电子商务的兴起，传统零售模式开始瓦解。

### compress  /kəmˈpres/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 6

**EN:** to reduce the time or space required for something  
**CN:** 压缩（时间或空间）

**Original examples:**
- [03:46] First, Dario is going to sit down with the scientist who turned GLP-1 into a medicine about **compressing** timelines in biology.  
  首先，Dario将与把GLP-1转化为药物的科学家坐下来讨论压缩生物学时间线的问题。
- [04:11] Two years ago, our CEO wrote that AI-enabled biology and medicine could **compress** 50 to 100 years of progress into 5 to 10.  
  两年前，我们的CEO写道，AI赋能的生物学和医学可以将50到100年的进展压缩到5到10年。
- [04:58] about **compressing** timelines in biology, moderated by Stat senior writer for medicine, Matt Herper.  
  关于压缩生物学时间线，由Stat医学高级撰稿人Matt Herper主持。
- [12:21] So I think it's hard to put a number on, right? But some of the work that I've experienced, right? So four years to make the right molecule. We still need to make molecules and prove that they are the right ones, but it could probably go to one year. And another area is recruiting patients for clinical trials. You know, some of the stuff we had to do sometimes recruit 20,000 people for a five-year study and it takes two years to recruit the people, yet people have never been more connected than they are now. So it must be possible to do that faster. I also think the regulatory... Process could really be revolutionized by AI. So there's just—I could go on for a long time thinking about it. I think so many areas could really become completely—uh, not completely **compressed**, but really markedly **compressed**.  
  所以我认为很难给出一个数字，对吧？但是我经历过的一些工作，对吧？所以制造正确的分子需要四年。我们仍然需要制造分子并证明它们是正确的，但可能可以缩短到一年。另一个领域是为临床试验招募患者。你知道，我们有时不得不为一项五年研究招募20,000人，招募这些人需要两年时间，然而人们从未像现在这样彼此联系紧密。所以一定有可能更快地做到这一点。我还认为监管...过程真的可以被AI彻底改革。所以只是——我可以花很长时间思考这个问题。我认为很多领域真的可以变得完全——呃，不是完全压缩，而是显著压缩。
- [13:46] So I think we really will see a large **compression**, but we still need the clinical trial data.  
  所以我认为我们真的会看到大量的压缩，但我们仍然需要临床试验数据。
- [01:12:32] Yes, and the earliest stages of discovery and thinking a lot about this notion of **compressing** biology that—We've heard a lot and Dario talk about.  
  是的，在最早的发现阶段，大量思考这个压缩生物学的概念——我们听到很多人和Dario谈论这个。

**Extra example:**
- New tools allow us to **compress** months of analysis into a few days.  
  新工具使我们能够将数月的分析压缩到几天内完成。

### timeline  /ˈtaɪmlaɪn/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 2

**EN:** a plan that shows the time by which different stages of a project should be completed  
**CN:** 时间线，时间表

**Original examples:**
- [03:46] First, Dario is going to sit down with the scientist who turned GLP-1 into a medicine about compressing **timelines** in biology.  
  首先，Dario将与把GLP-1转化为药物的科学家坐下来讨论压缩生物学时间线的问题。
- [04:58] about compressing **timelines** in biology, moderated by Stat senior writer for medicine, Matt Herper.  
  关于压缩生物学时间线，由Stat医学高级撰稿人Matt Herper主持。

**Extra example:**
- The project **timeline** was extended due to unexpected technical challenges.  
  由于意外的技术挑战，项目时间线被延长了。

### arc  /ɑːrk/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the progression or development of something through time  
**CN:** 发展历程，轨迹

**Original examples:**
- [04:29] Now, few people alive have carried the **arc** of a scientific idea all the way through.  
  现在，活着的人中很少有人能够将一个科学想法的整个发展历程进行到底。

**Extra example:**
- The **arc** of her career took her from startup founder to venture capitalist.  
  她的职业发展轨迹使她从初创公司创始人转变为风险投资家。

### revolutionary  /ˌrevəˈluːʃəneri/
**CEFR:** B2 | **Part of speech:** n./adj. | **Occurrences:** 1

**EN:** involving or causing a complete or dramatic change; a person who advocates radical change  
**CN:** 革命性的；革命者

**Original examples:**
- [05:26] So, hello everyone and welcome. I'm Matt Herper. I'm a journalist at Stat. We're an award-winning medical news site. And we're here for a conversation between two people I think we can fairly call **revolutionaries** about what might be a revolution.  
  那么，大家好，欢迎。我是Matt Herper。我是Stat的记者。我们是一个获奖的医学新闻网站。我们在这里进行一场对话，对话双方是两个我认为可以公平地称为革命者的人，讨论可能是一场革命的事情。

**Extra example:**
- The invention of the transistor was **revolutionary** for computing.  
  晶体管的发明对计算机来说是革命性的。

### visionary  /ˈvɪʒəneri/
**CEFR:** C1 | **Part of speech:** n./adj. | **Occurrences:** 1

**EN:** a person with original ideas about what the future will or could be like; thinking about or planning the future with imagination and wisdom  
**CN:** 有远见的人；有远见的

**Original examples:**
- [05:43] Lotte Knudsen, and I mean you've heard about her role in creating GLP-1s. She is a **visionary** who saw the impact these medicines could have not only in diabetes but obesity, a field by the way, and a theme that I think we want to pay attention to, that she joined by accident.  
  Lotte Knudsen，我的意思是你们已经听说过她在创造GLP-1方面的作用。她是一个有远见的人，她看到了这些药物不仅在糖尿病方面而且在肥胖方面可能产生的影响，顺便说一句，这是一个领域，也是一个我认为我们应该关注的主题，她是偶然加入的。

**Extra example:**
- Steve Jobs was considered a **visionary** in the technology industry.  
  乔布斯被认为是科技行业的一位有远见的人。

### brink  /brɪŋk/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a point at which something is about to happen or change  
**CN:** 边缘，临界点

**Original examples:**
- [07:13] Are we on the **brink** of a big change?  
  我们是否处于重大变革的边缘？

**Extra example:**
- The company was on the **brink** of bankruptcy before the acquisition.  
  在被收购之前，公司处于破产的边缘。

### inflection point  /ɪnˈflekʃən pɔɪnt/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 3

**EN:** a time of significant change in a situation; a turning point  
**CN:** 拐点，转折点

**Original examples:**
- [07:13] Yeah, I - and thank you for having me here. It's exciting times. I really do believe we are at a real **inflection point**, right, where there's so much data becoming available.  
  是的，我——感谢你们邀请我来这里。这是激动人心的时刻。我真的相信我们正处于一个真正的拐点，对吧，有如此多的数据正在变得可用。
- [08:10] And I've experienced both **inflection points** as well as hype in my world, right?  
  在我的世界里，我既经历过拐点，也经历过炒作，对吧？
- [21:03] We've kind of entered an **inflection point** or a critical window, you know, as we go along the exponential.  
  你知道，随着我们沿着指数曲线前进，我们已经进入了一个拐点或关键窗口。

**Extra example:**
- The introduction of the iPhone was an **inflection point** for mobile computing.  
  iPhone的推出是移动计算的一个拐点。

### hype  /haɪp/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** extravagant or intensive publicity or promotion; exaggerated claims  
**CN:** 炒作，夸大宣传

**Original examples:**
- [08:10] And I've experienced both inflection points as well as **hype** in my world, right? where the real goal with semaglutide was to actually create a health impact.  
  在我的世界里，我既经历过拐点，也经历过炒作，对吧？司美格鲁肽的真正目标实际上是创造健康影响。

**Extra example:**
- Despite all the **hype**, the product failed to meet market expectations.  
  尽管有所有的炒作，该产品未能达到市场预期。

### exponential  /ˌekspəˈnenʃəl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 6

**EN:** becoming more and more rapid; characterized by very fast growth  
**CN:** 指数级的

**Original examples:**
- [09:09] One, the technology is still getting better. It's on a fast **exponential**.  
  第一，技术仍在变得更好。它处于快速指数级增长中。
- [09:30] But, you know, we still have some ways to go on the **exponential**.  
  但是，你知道，在指数级增长方面我们仍有一段路要走。
- [21:03] Different things in terms of both benefits and economically useful applications and, you know, potentially concerning applications turn on at different times as we go along the **exponential**.  
  在收益、经济上有用的应用以及潜在令人担忧的应用方面，随着我们沿着指数曲线前进，不同的事物在不同的时间点出现。
- [32:31] So you know **exponentials** really catch you off guard.  
  所以你知道指数增长真的会让你措手不及。
- [32:52] And then the **exponential** does its thing, and they actually get to the point where they can help you a lot.  
  然后指数增长发挥了作用，它们实际上达到了可以给你很大帮助的程度。
- [37:49] So returning to the image of the **exponential** and how it affects different disciplines.  
  所以回到指数增长的形象以及它如何影响不同的学科。

**Extra example:**
- The **exponential** growth of data has created new storage challenges.  
  数据的指数级增长带来了新的存储挑战。

### inertia  /ɪˈnɜːrʃə/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** a tendency to do nothing or remain unchanged; resistance to change  
**CN:** 惯性，惰性

**Original examples:**
- [09:30] And the second is, I think, just the **inertia** in the system—both the **inertia** of getting used to using these tools and operating in the new way, and figuring out, you know, how do they help with academic biology research, how do they help with new target discovery, how do they help with running clinical trials faster.  
  第二个是，我认为，只是系统中的惯性——既包括习惯使用这些工具和以新方式运作的惯性，也包括弄清楚，你知道，它们如何帮助学术生物学研究，它们如何帮助新靶点发现，它们如何帮助更快地进行临床试验。
- [14:45] So there's this **inertia**, but I think there's lots of things we can do to work around it or to shorten the cycle time.  
  所以存在这种惯性，但我认为我们可以做很多事情来绕过它或缩短周期时间。

**Extra example:**
- Organizational **inertia** often prevents companies from adopting new technologies quickly.  
  组织惯性常常阻止公司快速采用新技术。

### regulatory  /ˈreɡjələtɔːri/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 2

**EN:** relating to rules or laws that control an activity or process  
**CN:** 监管的，法规的

**Original examples:**
- [09:50] And I think the longer response to the **regulatory** system, which is, you know, it's going to take a decade for all this new—  
  我认为对监管系统的更长期反应是，你知道，所有这些新的——需要十年时间。
- [13:02] I also think the **regulatory**... Process could really be revolutionized by AI.  
  我还认为监管...过程真的可以被AI彻底改革。

**Extra example:**
- The startup struggled to navigate the complex **regulatory** environment.  
  这家初创公司努力应对复杂的监管环境。

### pipeline  /ˈpaɪplaɪn/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 3

**EN:** a series of stages in a process, especially in business or research  
**CN:** 流水线，管线，研发管线

**Original examples:**
- [10:11] with AI, but I think once we get it going, particularly in all the parts of the **pipeline**,  
  使用AI，但我认为一旦我们让它运转起来，特别是在管线的所有部分，
- [01:31:43] Maybe the only thing I'll add in terms of advice, I think all the math shows when you look at **pipelines** that to Viv's point  
  也许我在建议方面唯一要补充的是，我认为所有的数学计算都表明，当你查看管道时，就像Viv所说的
- [01:44:38] Run **pipelines**, figures, every step reproducible and traceable.  
  运行流程、生成图表，每一步都可重现和可追溯。

**Extra example:**
- The company has several promising drugs in its development **pipeline**.  
  该公司的研发管线中有几种有前景的药物。

### revolutionize  /ˌrevəˈluːʃənaɪz/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 2

**EN:** to change something completely and fundamentally  
**CN:** 彻底改革，彻底改变

**Original examples:**
- [12:21] I also think the regulatory... Process could really be **revolutionized** by AI.  
  我还认为监管...过程真的可以被AI彻底改革。
- [12:21] There are already areas of the whole drug discovery and development process that have already been **revolutionized**.  
  整个药物发现和开发过程中已经有一些领域已经被彻底改革了。

**Extra example:**
- Cloud computing **revolutionized** how businesses manage their IT infrastructure.  
  云计算彻底改变了企业管理IT基础设施的方式。

### shoot for  /ʃuːt fɔːr/
**CEFR:** B2 | **Part of speech:** phrasal v. | **Occurrences:** 1

**EN:** to aim for or try to achieve something  
**CN:** 力争，争取

**Original examples:**
- [18:52] That is what we should **shoot for**.  
  那就是我们应该争取的目标。

**Extra example:**
- When setting targets, we should **shoot for** ambitious but achievable goals.  
  在设定目标时，我们应该争取雄心勃勃但可实现的目标。

### optimistic  /ˌɑːp.tɪˈmɪs.tɪk/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** hopeful and confident about the future  
**CN:** 乐观的

**Original examples:**
- [18:52] I am **optimistic**.  
  我是乐观的。

**Extra example:**
- Despite the challenges, investors remain **optimistic** about the company's long-term prospects.  
  尽管面临挑战，投资者仍然对公司的长期前景持乐观态度。

### confident  /ˈkɑːn.fɪ.dənt/
**CEFR:** B1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** feeling certain about something  
**CN:** 有信心的，确信的

**Original examples:**
- [18:52] I can't be **confident** because we don't know the future, but I'm optimistic.  
  我无法确信，因为我们不知道未来会怎样，但我持乐观态度。

**Extra example:**
- I'm **confident** that our team can deliver the project on time.  
  我相信我们的团队能够按时交付这个项目。

### probability  /ˌprɑː.bəˈbɪl.ə.t̬i/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 2

**EN:** the likelihood that something will happen  
**CN:** 可能性，概率

**Original examples:**
- [19:00] So I think we'll get better at improving the **probability** of success of new medicines because we'll understand them better.  
  所以我认为我们会更好地提高新药成功的概率，因为我们会更好地理解它们。
- [19:49] So you will improve your **probability** of success for whether a new medicine actually comes out successful.  
  所以你会提高新药是否真正成功推出的概率。

**Extra example:**
- The **probability** of achieving product-market fit increases with better user research.  
  通过更好的用户研究，实现产品与市场契合的概率会提高。

### foundation  /faʊnˈdeɪ.ʃən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the basic facts or principles on which something is built  
**CN:** 基础

**Original examples:**
- [19:00] We have a better **foundation** for why we picked those targets.  
  我们有更好的基础来解释为什么我们选择了那些目标。

**Extra example:**
- Building a strong technical **foundation** is essential before scaling a product.  
  在扩大产品规模之前，建立坚实的技术基础至关重要。

### target  /ˈtɑːr.ɡɪt/
**CEFR:** B1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** something that is aimed at or intended; in medicine, a molecule or pathway intended for drug action  
**CN:** 目标；（药物的）靶点

**Original examples:**
- [19:00] We have a better foundation for why we picked those **targets**.  
  我们有更好的基础来解释为什么我们选择了那些靶点。
- [32:09] So, um, I think what I'd like to most see, I think Latte mentioned it, is some success of AI with discovering new **targets** because I think that is the bottleneck.  
  所以，嗯，我认为我最想看到的，我想Latte提到过，是AI在发现新靶点方面取得一些成功，因为我认为那是瓶颈。

**Extra example:**
- Our research team identified a promising **target** for treating autoimmune diseases.  
  我们的研究团队确定了一个有希望治疗自身免疫性疾病的靶点。

### clinical trial  /ˈklɪn.ɪ.kəl ˈtraɪ.əl/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** a research study to test how well new medical treatments work in people  
**CN:** 临床试验

**Original examples:**
- [19:00] We can make smarter **clinical trials**.  
  我们可以进行更明智的临床试验。
- [25:33] So we have independent data monitoring committees in large **clinical trials**, maybe that's also something that might already be in place.  
  所以我们在大型临床试验中有独立的数据监测委员会，也许这也是可能已经存在的东西。

**Extra example:**
- The company is recruiting patients for a phase III **clinical trial** of their new cancer drug.  
  该公司正在招募患者参加其新癌症药物的三期临床试验。

### failure  /ˈfeɪl.jɚ/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** lack of success; an unsuccessful result  
**CN:** 失败

**Original examples:**
- [19:00] So I think we will see fewer **failures**.  
  所以我认为我们会看到更少的失败。

**Extra example:**
- Learning from **failure** is crucial for startup success.  
  从失败中学习对创业成功至关重要。

### insightful  /ɪnˈsaɪt.fəl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** showing a deep understanding of a situation or person  
**CN:** 有洞察力的，深刻的

**Original examples:**
- [19:19] And he's very **insightful**, right?  
  而且他非常有洞察力，对吧？

**Extra example:**
- Her **insightful** analysis of user behavior helped us redesign the product.  
  她对用户行为的深刻分析帮助我们重新设计了产品。

### criticism  /ˈkrɪt.ɪ.sɪ.zəm/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the act of expressing disapproval or pointing out faults  
**CN:** 批评

**Original examples:**
- [19:25] So, but there will be **criticism** until some examples have moved forward and people are going to say, oh this is not fully AI designed, but that's again, that's not the point.  
  所以，但在一些例子取得进展之前会有批评，人们会说，哦，这不是完全由AI设计的，但那又不是重点。

**Extra example:**
- The startup faced **criticism** for its data privacy practices.  
  这家初创公司因其数据隐私做法而面临批评。

### pleiotropic  /ˌplaɪ.əˈtrɑː.pɪk/
**CEFR:** C2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** describing a gene or drug that affects multiple traits or systems  
**CN:** 多效性的

**Original examples:**
- [19:57] And a very important learning from GLP-1 is that actually a **pleiotropic** effect is a good one, right?  
  GLP-1的一个非常重要的经验是，实际上多效性作用是好的，对吧？

**Extra example:**
- The **pleiotropic** effects of the compound made it a promising candidate for multiple diseases.  
  该化合物的多效性作用使其成为治疗多种疾病的有希望的候选药物。

### organ  /ˈɔːr.ɡən/
**CEFR:** B1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a part of the body that has a specific function  
**CN:** 器官

**Original examples:**
- [19:57] What's really so fantastic about GLP-1 is that you have all these benefits on multiple different **organs**.  
  GLP-1真正奇妙的地方是它对多个不同器官都有这些益处。

**Extra example:**
- The drug targets specific **organs** to minimize side effects.  
  该药物针对特定器官以最大限度地减少副作用。

### drug discovery  /drʌɡ dɪˈskʌv.ɚ.i/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the process of identifying new candidate medications  
**CN:** 药物发现

**Original examples:**
- [19:57] Yet the whole field of **drug discovery** is still looking for the genetics with the highest window or the mouse model with the highest window.  
  然而整个药物发现领域仍在寻找具有最高窗口期的遗传学或具有最高窗口期的小鼠模型。

**Extra example:**
- AI is accelerating **drug discovery** by predicting molecular interactions.  
  AI通过预测分子相互作用正在加速药物发现。

### broad  /brɔːd/
**CEFR:** B1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** wide-ranging or extensive  
**CN:** 广泛的

**Original examples:**
- [20:25] Finding these **broad** signals and we can help.  
  找到这些广泛的信号，我们可以提供帮助。

**Extra example:**
- We need to take a **broad** view of the problem before focusing on specific solutions.  
  在关注具体解决方案之前，我们需要对问题有一个广泛的看法。

### bioweapon  /ˈbaɪ.oʊˌwep.ən/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a harmful biological agent used as a weapon  
**CN:** 生物武器

**Original examples:**
- [20:34] I mean, the obvious worry is could all these cool technologies help people make **bioweapons** or synthetic organisms that run wild.  
  我是说，显而易见的担忧是，所有这些酷炫的技术会不会帮助人们制造生物武器或失控的合成生物。

**Extra example:**
- International treaties aim to prevent the development of **bioweapons**.  
  国际条约旨在防止生物武器的开发。

### synthetic  /sɪnˈθet̬.ɪk/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** made by combining chemical substances rather than being naturally produced  
**CN:** 合成的

**Original examples:**
- [20:34] Or **synthetic** organisms that run wild, or every science fiction novel we've read that actually could happen.  
  或者失控的合成生物，或者我们读过的每一部科幻小说中可能真正发生的事情。

**Extra example:**
- The lab is developing **synthetic** biology solutions for environmental cleanup.  
  该实验室正在开发用于环境清理的合成生物学解决方案。

### guard against  /ɡɑːrd əˈɡenst/
**CEFR:** B2 | **Part of speech:** phrasal v. | **Occurrences:** 1

**EN:** to take action to prevent something bad from happening  
**CN:** 防范，防止

**Original examples:**
- [20:45] How do you **guard against** that, Daria?  
  你怎么防范这个，Daria？

**Extra example:**
- Companies must **guard against** security vulnerabilities in their systems.  
  公司必须防范其系统中的安全漏洞。

### confront  /kənˈfrʌnt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to face or deal with a difficult situation  
**CN:** 面对，应对

**Original examples:**
- [20:57] Um, you know, we're currently **confronting** one of those sets of issues with kind of the cyber risks of AI, right?  
  嗯，你知道，我们目前正在应对其中一组问题，就是AI的网络风险，对吧？

**Extra example:**
- Every startup must **confront** the challenge of scaling while maintaining quality.  
  每个初创公司都必须面对在保持质量的同时扩大规模的挑战。

### critical  /ˈkrɪt̬.ɪ.kəl/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** extremely important; at a point of crisis or danger  
**CN:** 关键的，重要的

**Original examples:**
- [21:03] We've kind of entered an inflection point or a **critical** window, you know, as we go along the exponential.  
  你知道，随着我们沿着指数曲线前进，我们已经进入了一个拐点或关键窗口。

**Extra example:**
- User feedback is **critical** for improving product features.  
  用户反馈对于改进产品功能至关重要。

### exploit  /ɪkˈsplɔɪt/
**CEFR:** B2 | **Part of speech:** n./v. | **Occurrences:** 1

**EN:** a software vulnerability that can be used to gain unauthorized access; to take advantage of such a vulnerability  
**CN:** 漏洞；利用（漏洞）

**Original examples:**
- [21:25] With cyber, it's like you have the ability to find **exploits**, and then it's the same model can also patch the **exploits**, and you can find the **exploits** in a few seconds, you can patch the **exploits** in a few seconds.  
  对于网络安全，就像你有能力发现漏洞，然后同一个模型也可以修补漏洞，你可以在几秒钟内找到漏洞，也可以在几秒钟内修补漏洞。

**Extra example:**
- Security researchers discovered a critical **exploit** in the authentication system.  
  安全研究人员在认证系统中发现了一个严重的漏洞。

### patch  /pætʃ/
**CEFR:** B2 | **Part of speech:** v./n. | **Occurrences:** 1

**EN:** to fix a software vulnerability or bug; a software update that fixes problems  
**CN:** 修补；补丁

**Original examples:**
- [21:25] And then it's the same model can also **patch** the exploits, and you can find the exploits in a few seconds, you can **patch** the exploits in a few seconds.  
  然后同一个模型也可以修补漏洞，你可以在几秒钟内找到漏洞，也可以在几秒钟内修补漏洞。

**Extra example:**
- The development team released a **patch** to fix the security vulnerability.  
  开发团队发布了一个补丁来修复安全漏洞。

### symmetry  /ˈsɪm.ə.tri/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the quality of being the same or balanced on both sides  
**CN:** 对称性，平衡

**Original examples:**
- [21:42] As we learned during COVID-19, there's not necessarily the same **symmetry**, but there are some lessons we can learn.  
  正如我们在COVID-19期间了解到的，不一定存在同样的对称性，但我们可以学到一些教训。

**Extra example:**
- There's no **symmetry** between the cost of building and maintaining the system.  
  构建和维护系统的成本之间没有对称性。

### safeguard  /ˈseɪf.ɡɑːrd/
**CEFR:** C1 | **Part of speech:** n./v. | **Occurrences:** 3

**EN:** a measure taken to protect against harm or risk  
**CN:** 保护措施；保护

**Original examples:**
- [21:52] For example, the **safeguards** that we put on the models to make sure that they can output beneficial content but can't output worrying content.  
  例如，我们在模型上设置的保护措施，以确保它们可以输出有益的内容，但不能输出令人担忧的内容。
- [22:21] How do you balance putting **safeguards** on the model versus putting the model in the hands of people who can kind of be the white hat hacker and figure out what you can do with it that's harmful?  
  你如何平衡在模型上设置保护措施与将模型交给那些可以成为白帽黑客并找出可以用它做什么有害的事情的人？
- [23:10] You both need to have these **safeguards** and you need to have trusted access programs, right?  
  你既需要这些保护措施，也需要可信访问程序，对吧？

**Extra example:**
- The company implemented multiple **safeguards** to protect user data.  
  该公司实施了多项保护措施来保护用户数据。

### withdraw  /wɪðˈdrɔː/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to remove or take back something, especially officially  
**CN:** 撤回，撤销（尤指正式地）

**Original examples:**
- [24:16] You've both seen cases outside of your work where there were drugs that had real side effects that required they be **withdrawn**.  
  你们都见过工作之外的案例，那些药物有真实的副作用，需要被撤回。

**Extra example:**
- The company had to **withdraw** the product from the market after safety concerns emerged.  
  在出现安全问题后，公司不得不从市场上撤回该产品。

### pan out  /pæn aʊt/
**CEFR:** B2 | **Part of speech:** phrasal v. | **Occurrences:** 1

**EN:** to develop or happen in a particular way, especially successfully  
**CN:** 结果是，证明是（尤指成功地）

**Original examples:**
- [24:16] Also some of the worries about GLP-1s which have mostly not **panned out**.  
  还有一些关于GLP-1的担忧，大多数都没有成为现实。

**Extra example:**
- Their initial concerns about the technology **panned out** to be unfounded.  
  他们对这项技术的最初担忧结果证明是没有根据的。

### bylaw  /ˈbaɪlɔː/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a rule made by an organization to control the actions of its members  
**CN:** 章程，内部规章

**Original examples:**
- [24:40] It's even I guess put in your **bylaws** that how you are thinking about the benefit versus the risk.  
  我想这甚至已经写入你们的章程中，即你们如何考虑收益与风险。

**Extra example:**
- The organization's **bylaws** require board approval for major decisions.  
  该组织的章程要求董事会批准重大决策。

### governance  /ˈɡʌvərnəns/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the way that organizations or countries are managed at the highest level  
**CN:** 治理，管理（指最高层面）

**Original examples:**
- [26:12] I think exactly this kind of independent **governance** makes sense.  
  我认为正是这种独立治理是有意义的。

**Extra example:**
- Effective **governance** structures are essential for managing AI risks.  
  有效的治理结构对于管理人工智能风险至关重要。

### hallucinate  /həˈluːsɪneɪt/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 8

**EN:** in AI context: to generate false or nonsensical information presented as fact  
**CN:** （人工智能）产生幻觉，生成虚假信息

**Original examples:**
- [26:47] Why should pharma trust AI predictions when your models **hallucinate**?  
  当你们的模型会产生幻觉时，制药公司为什么应该相信人工智能的预测？
- [27:03] So I would say on **hallucinations**, actually **hallucinations** have gotten better and better over time.  
  所以我想说关于幻觉问题，实际上幻觉已经随着时间的推移变得越来越好了。
- [27:10] You don't hear as much about **hallucinations** as you used to. They still happen.  
  你不会像以前那样经常听到关于幻觉的消息了。它们仍然会发生。
- [27:25] I don't think we will ever have an AI model that never **hallucinates**.  
  我认为我们永远不会有一个从不产生幻觉的人工智能模型。
- [27:36] It is prone to a duality between creativity and, you know, basically **hallucination**, right?  
  它容易陷入创造力和幻觉之间的二元性，对吧？
- [28:29] So I think **hallucinations** are going to go down and down.  
  所以我认为幻觉会越来越少。
- [29:06] I think that's maybe another illustration of this balance between creativity and **hallucination**, right?  
  我认为这可能是创造力和幻觉之间平衡的另一个例证，对吧？
- [29:16] It's like, you know, he came up with brilliant, brilliant ideas, and yet his **hallucination** rate is pretty high, isn't it?  
  就像，你知道，他提出了非常出色的想法，但他的幻觉率相当高，不是吗？

**Extra example:**
- The model occasionally **hallucinates** citations that don't exist.  
  该模型偶尔会虚构不存在的引用文献。

### validation  /ˌvælɪˈdeɪʃən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the process of proving that something is correct or true using evidence  
**CN:** 验证，确认

**Original examples:**
- [26:47] Where's the **validation** data for claims that AI accelerates timelines?  
  关于人工智能加速时间线的说法，验证数据在哪里？

**Extra example:**
- Clinical trials require rigorous **validation** before a drug can be approved.  
  临床试验需要严格的验证，然后药物才能获得批准。

### probabilistic  /ˌprɑːbəbəˈlɪstɪk/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** based on or relating to probability; involving uncertainty  
**CN:** 概率的，基于概率的

**Original examples:**
- [27:25] I think just the **probabilistic** way in which these models reason, which I suspect is the same as the **probabilistic** way in which humans—  
  我认为这些模型进行推理的概率方式，我怀疑与人类的概率方式是一样的——

**Extra example:**
- The system uses **probabilistic** algorithms to predict outcomes.  
  该系统使用概率算法来预测结果。

### duality  /duːˈæləti/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the quality or state of having two parts or aspects  
**CN:** 二元性，双重性

**Original examples:**
- [27:36] It is prone to a **duality** between creativity and, you know, basically hallucination, right?  
  它容易陷入创造力和幻觉之间的二元性，对吧？

**Extra example:**
- There's a **duality** in leadership between being decisive and being collaborative.  
  领导力存在一种二元性，既要果断又要善于合作。

### straddle  /ˈstrædəl/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to exist in or deal with two different situations or states at the same time  
**CN:** 横跨，兼顾（两种不同的状态）

**Original examples:**
- [27:48] In order to be creative, you're often **straddling** the boundary between making things up and having good ideas.  
  为了有创造力，你经常需要在编造事物和产生好想法之间取得平衡。

**Extra example:**
- The company is **straddling** the line between innovation and regulatory compliance.  
  这家公司正在创新和监管合规之间寻求平衡。

### dogmatically  /dɔːɡˈmætɪkli/
**CEFR:** C1 | **Part of speech:** adv. | **Occurrences:** 1

**EN:** in a way that shows an unwillingness to consider other opinions  
**CN:** 教条地，固执己见地

**Original examples:**
- [28:47] Sometimes, you know, like humans, I imagine the AI models may get **dogmatically** attached.  
  有时候，你知道，就像人类一样，我想人工智能模型可能会变得教条地执着。

**Extra example:**
- He **dogmatically** refused to consider alternative approaches to the problem.  
  他固执地拒绝考虑解决问题的其他方法。

### skeptic  /ˈskeptɪk/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 2

**EN:** a person who doubts the truth or value of an idea or belief  
**CN:** 怀疑论者，持怀疑态度的人

**Original examples:**
- [29:30] What question do you hear from **skeptics** in the industry?  
  你从行业中的怀疑论者那里听到什么问题？
- [29:36] So people should be **skeptical**, absolutely, but they should lean in because that is how we can create more solutions for the work.  
  所以人们应该持怀疑态度，绝对应该，但他们应该积极参与，因为这样我们才能为工作创造更多解决方案。

**Extra example:**
- Many **skeptics** changed their minds after seeing the demo.  
  许多怀疑论者在看到演示后改变了他们的想法。

### lean in  /liːn ɪn/
**CEFR:** B2 | **Part of speech:** phrasal v. | **Occurrences:** 1

**EN:** to actively engage with or commit to something, especially when facing uncertainty  
**CN:** 积极参与，主动投入

**Original examples:**
- [29:36] So people should be skeptical, absolutely, but they should **lean in** because that is how we can create more solutions for the work.  
  所以人们应该持怀疑态度，绝对应该，但他们应该积极参与，因为这样我们才能为工作创造更多解决方案。

**Extra example:**
- When facing a difficult challenge, successful teams **lean in** rather than avoid it.  
  面对困难挑战时，成功的团队会主动投入而不是回避。

### foresight  /ˈfɔːrsaɪt/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the ability to predict what will happen or be needed in the future  
**CN:** 远见，先见之明

**Original examples:**
- [32:31] So you know the field rising to meet the moment and scientists having the **foresight** to say well the AI model I had three months ago couldn't help me at all with this but the one I just got today actually is helping me a lot.  
  所以你知道这个领域正在迎接这一时刻，科学家们有远见地说，三个月前我拥有的人工智能模型根本无法帮助我解决这个问题，但我今天刚得到的模型实际上帮了我很多。

**Extra example:**
- Leaders with **foresight** invest in emerging technologies before they become mainstream.  
  有远见的领导者会在新兴技术成为主流之前进行投资。

### pace  /peɪs/
**CEFR:** B2 | **Part of speech:** n./v. | **Occurrences:** 1

**EN:** the speed at which something happens or develops  
**CN:** 速度，节奏

**Original examples:**
- [32:52] Just the attention and the **foresight** to keep **pace** with the technology and just keep revisiting and understanding how fast it's improving.  
  只需要有关注和远见来跟上技术的步伐，并不断重新审视和理解它改进的速度有多快。

**Extra example:**
- Many companies struggle to keep **pace** with rapid changes in consumer behavior.  
  许多公司难以跟上消费者行为快速变化的步伐。

### reflexive  /rɪˈfleksɪv/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** done without thinking, as an automatic reaction  
**CN:** 本能的，条件反射的

**Original examples:**
- [33:04] What I hope I don't see is also something we've alluded to, which is the kind of **reflexive** skepticism.  
  我希望不要看到的也是我们暗示过的东西，即那种条件反射式的怀疑态度。

**Extra example:**
- His **reflexive** rejection of new ideas prevented the team from innovating.  
  他对新想法的本能拒绝阻碍了团队的创新。

### allude  /əˈluːd/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to mention something indirectly or in a way that is not clear  
**CN:** 暗指，间接提到

**Original examples:**
- [33:04] What I hope I don't see is also something we've **alluded** to, which is the kind of reflexive skepticism.  
  我希望不要看到的也是我们暗示过的东西，即那种条件反射式的怀疑态度。

**Extra example:**
- The CEO **alluded** to upcoming changes without revealing specific details.  
  首席执行官暗示即将发生变化，但没有透露具体细节。

### verifiable  /ˈverɪfaɪəbəl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** able to be checked or proven to be true  
**CN:** 可证实的，可核实的

**Original examples:**
- [33:52] Not as easily **verifiable**, and the pattern is always the same within any area.  
  不那么容易验证，而且在任何领域中模式总是相同的。

**Extra example:**
- Scientific claims must be **verifiable** through reproducible experiments.  
  科学主张必须通过可重复的实验来验证。

### dismiss  /dɪsˈmɪs/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 2

**EN:** to decide that something is not important or worth considering  
**CN:** 不予考虑，驳回，轻视

**Original examples:**
- [33:52] The models are useless, useless, useless. They're **dismissed**, and then they get to a point where they can help the ordinary practitioner.  
  这些模型毫无用处，毫无用处，毫无用处。它们被轻视了，然后它们达到了可以帮助普通从业者的程度。
- [34:13] And then those who are the most skilled, the most advanced, still **dismiss** them.  
  然后那些最熟练、最先进的人仍然轻视它们。

**Extra example:**
- Early critics **dismissed** the internet as a passing fad.  
  早期的批评者将互联网视为一时的风潮而不予考虑。

### median  /ˈmiːdiən/
**CEFR:** C1 | **Part of speech:** adj./n. | **Occurrences:** 1

**EN:** relating to the middle value in a set of data; average or typical  
**CN:** 中位数的，中等的

**Original examples:**
- [34:13] This may help the **median** person, but it won't actually advance the field.  
  这可能对中等水平的人有帮助，但实际上不会推进这个领域。

**Extra example:**
- The new tool raised the **median** productivity level across the team.  
  新工具提高了整个团队的中等生产力水平。

### overnight  /ˌoʊvərˈnaɪt/
**CEFR:** B2 | **Part of speech:** adv. | **Occurrences:** 1

**EN:** very quickly or suddenly  
**CN:** 突然地，一夜之间

**Original examples:**
- [34:29] And then suddenly **overnight** the thing—you know, again, the last six months, the thing that's fresh in my mind is cyber.  
  然后突然之间——你知道，再说一次，最近六个月，我记忆犹新的是网络安全。

**Extra example:**
- The company's valuation changed **overnight** after the product launch.  
  产品发布后，公司的估值一夜之间发生了变化。

### analogy  /əˈnæləʤi/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a comparison between things that have similar features  
**CN:** 类比，比拟

**Original examples:**
- [34:39] It's more **verifiable** than biology. It won't be an exact **analogy**, but I think it's going to happen.  
  它比生物学更容易验证。这不会是一个完全精确的类比，但我认为它会发生。

**Extra example:**
- The speaker used an **analogy** to explain the complex algorithm.  
  演讲者使用类比来解释这个复杂的算法。

### feedback loop  /FEED-bak loop/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a process where the output of a system is fed back as input, creating a cycle of cause and effect  
**CN:** 反馈循环；反馈回路

**Original examples:**
- [38:26] In the life sciences, our **feedback loops** take longer and they involve running real experiments in the physical world.  
  在生命科学中，我们的反馈循环需要更长时间，并且涉及在物理世界中运行真实实验。

**Extra example:**
- Creating tight **feedback loops** between engineers and users accelerates product iteration.  
  在工程师和用户之间建立紧密的反馈循环可以加速产品迭代。

### uncertainty  /ʌn-SUR-tn-tee/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the state of being uncertain; lack of certainty or predictability  
**CN:** 不确定性

**Original examples:**
- [38:26] There's so much **uncertainty** and noise in all biological data.  
  所有生物数据中都存在大量的不确定性和噪声。

**Extra example:**
- Market **uncertainty** has made investors more cautious about funding early-stage startups.  
  市场的不确定性使得投资者对资助早期创业公司更加谨慎。

### objective  /əb-JEK-tiv/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 3

**EN:** a goal or aim that you are trying to achieve  
**CN:** 目标；目的

**Original examples:**
- [38:51] For our life science efforts, we have two primary **objectives** that we're pursuing.  
  对于我们的生命科学工作，我们正在追求两个主要目标。
- [39:07] Everything that we're doing is constructed to build a full stack approach to pursue these **objectives** as fast as we can.  
  我们所做的一切都是为了构建一个全栈方法，以尽可能快地实现这些目标。
- [39:41] There's all the work that we're doing with all of our partners and customers and internally to directly pursue these **objectives** that we have.  
  我们与所有合作伙伴、客户以及内部所做的所有工作都是为了直接实现我们的这些目标。

**Extra example:**
- The team failed to meet its quarterly **objectives** due to unexpected technical challenges.  
  由于意外的技术挑战，团队未能完成季度目标。

### pursue  /pər-SOO/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 3

**EN:** to follow or chase something in order to catch or achieve it  
**CN:** 追求；从事

**Original examples:**
- [38:51] For our life science efforts, we have two primary objectives that we're **pursuing**.  
  对于我们的生命科学工作，我们正在追求两个主要目标。
- [39:12] A full stack approach to **pursue** these objectives as fast as we can.  
  一种全栈方法来尽快实现这些目标。
- [39:41] There's all the work that we're doing with all of our partners and customers and internally to directly **pursue** these objectives that we have.  
  我们与所有合作伙伴、客户以及内部所做的所有工作都是为了直接实现我们的这些目标。

**Extra example:**
- She decided to **pursue** a PhD in machine learning after working in industry for five years.  
  在业界工作五年后，她决定攻读机器学习博士学位。

### accelerate  /ak-SEL-ə-rayt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 4

**EN:** to make something happen faster or earlier  
**CN:** 加速；促进

**Original examples:**
- [38:56] The first is **accelerating** scientific discovery as an end in itself, pure pursuit of basic research.  
  第一个目标是加速科学发现本身，纯粹追求基础研究。
- [45:24] So how can we **accelerate** science progress overall and for biology in particular?  
  那么我们如何加速整体科学进步，特别是生物学的进步？
- [01:18:02] You've been outspoken and talked a lot about how at BMS you don't want AI simply to **accelerate** the current processes, but actually transform how you operate and do work.  
  你一直直言不讳地谈论在BMS，你们不希望AI只是加速当前的流程，而是真正改变你们的运营和工作方式。
- [01:24:20] And yesterday chatting with Lotus, she gave this great example of the four years and several thousand compounds that it took to test and get the stability right of GLP-1 and semaglutide. And again, 25%, you know, if you go to a thousand compounds in a year, let alone maybe even more aggressive, you know, that's an incredible **acceleration** and impact.  
  昨天与Lotus聊天时，她给出了一个很好的例子，测试GLP-1和司美格鲁肽的稳定性花了四年时间和数千种化合物。再说一次，25%，你知道，如果一年内测试一千种化合物，更不用说可能更激进，那是一个令人难以置信的加速和影响。

**Extra example:**
- The new automation tools **accelerated** the testing process by 60%.  
  新的自动化工具将测试过程加速了60%。

### alleviate  /ə-LEE-vee-ayt/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to make something less severe or more bearable  
**CN:** 减轻；缓解

**Original examples:**
- [39:01] And the second is **alleviating** the burden of disease and aging.  
  第二个目标是减轻疾病和衰老的负担。

**Extra example:**
- The new policy aims to **alleviate** housing pressure in major cities.  
  新政策旨在缓解大城市的住房压力。

### burden  /BUR-dən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** something difficult or unpleasant that you have to deal with or worry about  
**CN:** 负担；重担

**Original examples:**
- [39:01] And the second is alleviating the **burden** of disease and aging.  
  第二个目标是减轻疾病和衰老的负担。

**Extra example:**
- The financial **burden** of student loans affects career choices for many graduates.  
  学生贷款的经济负担影响着许多毕业生的职业选择。

### full stack  /fool stak/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** involving or encompassing all layers or components of a system, from foundational to user-facing  
**CN:** 全栈的；全方位的

**Original examples:**
- [39:12] A **full stack** approach to pursue these objectives as fast as we can.  
  一种全栈方法来尽快实现这些目标。

**Extra example:**
- We're looking for a **full stack** engineer who can handle both frontend and backend development.  
  我们正在寻找一位能够处理前端和后端开发的全栈工程师。

### foundational  /fown-DAY-shə-nəl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 2

**EN:** forming the basis or foundation; fundamental  
**CN:** 基础的；根本的

**Original examples:**
- [39:18] It starts at the **foundational** layer with our foundation models Claude.  
  它从我们的基础模型Claude的基础层开始。
- [01:30:13] And we called something **foundational** datasets for foundation models as an initiative  
  我们将某项工作称为基础模型的基础数据集，作为一项倡议

**Extra example:**
- Understanding data structures is **foundational** to becoming a good software engineer.  
  理解数据结构是成为优秀软件工程师的基础。

### layer  /LAY-ər/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 4

**EN:** a level or level of structure in a system  
**CN:** 层；层级

**Original examples:**
- [39:18] It starts at the foundational **layer** with our foundation models Claude.  
  它从我们的基础模型Claude的基础层开始。
- [39:27] It includes the product **layer** of optimizing the product for scientists.  
  它包括为科学家优化产品的产品层。
- [40:01] So we'll start at the model **layer**.  
  所以我们将从模型层开始。
- [41:47] There's so much work to do that should be captured in the product **layer**.  
  有太多工作需要在产品层中完成。

**Extra example:**
- The application **layer** sits on top of the transport and network **layers** in the OSI model.  
  在OSI模型中，应用层位于传输层和网络层之上。

### optimize  /OP-tə-myz/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to make something as effective or functional as possible  
**CN:** 优化；使最优化

**Original examples:**
- [39:27] It includes the product layer of **optimizing** the product for scientists.  
  它包括为科学家优化产品的产品层。

**Extra example:**
- We need to **optimize** the database queries to reduce page load times.  
  我们需要优化数据库查询以减少页面加载时间。

### integrate  /IN-tə-grayt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 3

**EN:** to combine things so that they work together effectively  
**CN:** 整合；集成

**Original examples:**
- [39:27] We still need to have the right product features to make the model intelligence **integrated** into workflows.  
  我们仍然需要拥有正确的产品功能，以将模型智能整合到工作流程中。
- [01:11:15] But it's also this big question, and even an experimental question, of how we **integrate** that into our work, how we **integrate** that into our organizations.  
  但这也是一个重大问题，甚至是一个实验性问题，即我们如何将其整合到工作中，如何将其整合到组织中。
- [01:12:54] Help us maybe understand where this notion of lab-in-the-loop and how you're thinking about early stage discovery and **integrating** AI at Genentech and where that's going.  
  帮助我们理解实验室闭环这个概念，以及你们如何思考早期发现阶段以及在基因泰克整合AI的问题，以及这将走向何方。

**Extra example:**
- The new payment system **integrates** seamlessly with our existing e-commerce platform.  
  新的支付系统与我们现有的电商平台无缝集成。

### workflow  /WURK-floh/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 5

**EN:** the sequence of processes through which a piece of work passes from initiation to completion  
**CN:** 工作流程；工作流

**Original examples:**
- [39:27] We still need to have the right product features to make the model intelligence integrated into **workflows**.  
  我们仍然需要拥有正确的产品功能，以将模型智能整合到工作流程中。
- [41:25] There's so many things that are unique about scientific use cases that aren't well captured by the use cases and the **workflows** of these other fields.  
  科学用例有许多独特之处，这些其他领域的用例和工作流程无法很好地涵盖。
- [42:05] Claude Science is intended to be your AI workbench that drives all of the work that you do across the whole scientific **workflow**.  
  Claude Science旨在成为您的AI工作台，推动您在整个科学工作流程中所做的所有工作。
- [43:04] It's increasingly becoming the case that many scientific **workflows** are leaning more and more on these high-performance scientific computing jobs.  
  越来越多的科学工作流程越来越依赖这些高性能科学计算任务。
- [45:43] To demonstrate its capabilities, I'm going to walk you through one **workflow**, the example of a real drug program end to end.  
  为了展示其功能，我将向您介绍一个工作流程，一个真实药物项目从头到尾的例子。

**Extra example:**
- Automating repetitive tasks in your **workflow** can save hours each week.  
  自动化工作流程中的重复性任务每周可以节省数小时。

### tackle  /TAK-əl/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to make a determined effort to deal with a difficult problem or situation  
**CN:** 处理；解决（难题）

**Original examples:**
- [39:54] Can AI actually **tackle** scientific problems?  
  AI真的能解决科学问题吗？

**Extra example:**
- The new team is ready to **tackle** the legacy codebase refactoring project.  
  新团队准备好处理遗留代码库重构项目了。

### demonstrate  /DEM-ən-strayt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 2

**EN:** to show clearly that something exists or is true through evidence or examples  
**CN:** 证明；展示

**Original examples:**
- [40:01] What we've **demonstrated** over the course of the last six months or so is rapid progress in the underlying foundation model capabilities.  
  在过去六个月左右的时间里，我们证明了底层基础模型能力的快速进步。
- [45:43] To **demonstrate** its capabilities, I'm going to walk you through one workflow.  
  为了展示其功能，我将向您介绍一个工作流程。

**Extra example:**
- The prototype **demonstrates** how the new algorithm performs under real-world conditions.  
  原型展示了新算法在实际条件下的表现。

### underlying  /UN-dər-ly-ing/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 2

**EN:** fundamental or basic; forming the foundation of something  
**CN:** 根本的；潜在的；底层的

**Original examples:**
- [40:01] What we've demonstrated over the course of the last six months or so is rapid progress in the **underlying** foundation model capabilities.  
  在过去六个月左右的时间里，我们证明了底层基础模型能力的快速进步。
- [44:11] So we have our **underlying** frontier models.  
  所以我们有我们底层的前沿模型。

**Extra example:**
- We need to fix the **underlying** infrastructure issues before adding new features.  
  在添加新功能之前，我们需要修复底层基础设施问题。

### capability  /kay-pə-BIL-ə-tee/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 6

**EN:** the ability or power to do something  
**CN:** 能力；性能

**Original examples:**
- [40:01] What we've demonstrated over the course of the last six months or so is rapid progress in the underlying foundation model **capabilities**.  
  在过去六个月左右的时间里，我们证明了底层基础模型能力的快速进步。
- [40:47] As I said before, it's not enough to build great models. We also need to build great products. It's not enough to just have the **capabilities**.  
  正如我之前所说，仅仅构建出色的模型是不够的。我们还需要构建出色的产品。仅仅拥有能力是不够的。
- [43:49] And the next feature that I'll talk about is it comes ready for each domain on day one. And in biology, this means connecting to tons of different databases and specialized tools so that you don't have to go in and manually configure that. But it is rapidly reconfigurable so that if there are additional tools that you want to connect to, you can set those up very quickly as well. It comes with **capabilities** in many different domains.  
  我要谈的下一个功能是它从第一天起就为每个领域做好了准备。在生物学中，这意味着连接到大量不同的数据库和专业工具，这样您就不必手动配置。但它可以快速重新配置，因此如果有其他工具想要连接，您也可以很快设置。它在许多不同领域都具有能力。
- [45:43] To demonstrate its **capabilities**, I'm going to walk you through one workflow.  
  为了展示其功能，我将向您介绍一个工作流程。
- [50:27] First, Claude for science ships with **capabilities** in many different domains.  
  首先，Claude for science在许多不同领域都具有能力。
- [51:03] And in the event that it doesn't, it's missing **capabilities**, the product is extraordinarily customizable.  
  如果它缺少某些能力，该产品具有极强的可定制性。

**Extra example:**
- The new framework's testing **capabilities** have significantly reduced our bug count.  
  新框架的测试能力显著减少了我们的bug数量。

### benchmark  /BENCH-mark/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a standard or point of reference against which things may be compared or assessed  
**CN:** 基准；标准

**Original examples:**
- [40:19] And we're seeing rapid progress in several **benchmarks** here.  
  我们在这里看到了几个基准测试的快速进步。

**Extra example:**
- Industry **benchmarks** show that our API latency is in the top 5% of competitors.  
  行业基准测试显示，我们的API延迟在竞争对手中排名前5%。

### on par with  /on par with/
**CEFR:** C1 | **Part of speech:** phrase | **Occurrences:** 1

**EN:** at the same level or standard as someone or something else  
**CN:** 与...相当；与...同等水平

**Original examples:**
- [40:28] And the models over this time frame have gone from being not all that useful to performing at a level that is **on par or greater than** the average PhD level scientist in these fields.  
  在这段时间内，这些模型已经从不太有用发展到表现水平与这些领域的平均博士级科学家相当或更高。

**Extra example:**
- Our startup's engineering culture is **on par with** that of top tech companies.  
  我们初创公司的工程文化与顶级科技公司相当。

### artifact  /AR-tə-fakt/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 11

**EN:** an object produced or shaped by human workmanship; in computing, an output file or resource produced by a build or development process  
**CN:** 产物；制品；工件

**Original examples:**
- [42:13] The first is that Claude Science has a rich set of scientific **artifacts** that it supports that are fully reproducible.  
  首先，Claude Science支持一套丰富的科学产物，这些产物是完全可重现的。
- [42:36] In addition, it supports a wide array of different types of **artifacts**.  
  此外，它支持各种不同类型的产物。
- [42:39] Science is a very visual affair. You need to deal with protein structures and small molecules and multiple sequence alignments, figures in your papers, whole manuscript drafts, and you need to be able to iterate with Claude in real time and explore these different types of **artifacts** together.  
  科学是一项非常视觉化的工作。您需要处理蛋白质结构、小分子、多序列比对、论文中的图表、整个手稿草稿，您需要能够与Claude实时迭代并一起探索这些不同类型的产物。
- [51:29] Every output that Claude produces comes with its full history attached. So going back to the figure I showed you initially, if you click here and look into the provenance of this **artifact**, you'll see that the **artifact** has the code that produced it, including the input **artifacts** it depends on.  
  Claude生成的每个输出都附带完整的历史记录。因此，回到我最初向您展示的图表，如果您在这里点击并查看这个产物的来源，您会看到该产物有生成它的代码，包括它依赖的输入产物。
- [52:06] What this means is that you can come back to the product six months, a year, two years later, every **artifact** is reproducible by construction. And because Claude knows how every **artifact** was made, this supports a lot of incredibly powerful interaction patterns.  
  这意味着您可以在六个月、一年、两年后回到产品，每个产物在构造上都是可重现的。因为Claude知道每个产物是如何制作的，这支持许多非常强大的交互模式。
- [53:13] Every **artifact** is versioned.  
  每个工件都有版本。
- [53:20] Versions are immutable, and each version has its own provenance attached to it.  
  版本是不可变的，每个版本都附有自己的来源记录。
- [53:36] And **artifacts** are checked.  
  工件会被检查。
- [53:51] Underneath every agent is a built-in reviewer that is assessing the accuracy of every claim that the agent is making and every **artifact** it produces.  
  每个代理下面都有一个内置的审查者，评估代理所做的每个声明和它产生的每个工件的准确性。
- [54:14] In my own personal research project where I continued my own PhD, at this point, Claude has produced thousands and thousands of **artifacts** all leading up to one output, a manuscript.  
  在我自己继续攻读博士学位的个人研究项目中，到目前为止，Claude已经产生了成千上万个工件，最终形成一个输出：一份手稿。
- [56:20] Cloud Science is a very visual product. We have built-in **artifact** renderers for images, but not just images.  
  Cloud Science是一个非常可视化的产品。我们有内置的工件渲染器用于图像，但不仅仅是图像。

**Extra example:**
- The build system generates **artifacts** like compiled binaries and deployment packages.  
  构建系统生成诸如编译二进制文件和部署包之类的产物。

### domain  /doh-MAYN/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 3

**EN:** a specific area of knowledge, activity, or expertise  
**CN:** 领域；范畴

**Original examples:**
- [43:43] And the next feature that I'll talk about is it comes ready for each **domain** on day one.  
  我要谈的下一个功能是它从第一天起就为每个领域做好了准备。
- [50:27] First, Claude for science ships with capabilities in many different **domains**.  
  首先，Claude for science在许多不同领域都具有能力。
- [50:52] Claude Science comes ready for your **domain** out of the box.  
  Claude Science开箱即用，为您的领域做好准备。

**Extra example:**
- She has deep expertise in the **domain** of natural language processing.  
  她在自然语言处理领域拥有深厚的专业知识。

### annotation  /ˌæn.əˈteɪ.ʃən/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** a note added to a text, diagram, or image to provide explanation or comment  
**CN:** 注释，标注

**Original examples:**
- [52:32] Claude sees every **annotation** that you make with its vision capabilities as well as its ability to read text, of course.  
  Claude通过其视觉能力以及文本阅读能力，能看到你做的每一个注释。
- [57:14] But at the bottom of the list is serpterin. This is the approved drug, and is ranked dead last for binding affinity to the pocket. But we can always just **annotate** and say the legend is covering it.  
  但在列表底部是serpterin。这是已批准的药物，在与口袋的结合亲和力方面排名垫底。但我们总是可以标注并说明图例遮住了它。

**Extra example:**
- Students should **annotate** the passage with their observations before class discussion.  
  学生应该在课堂讨论前用观察结果标注这段文字。

### provenance  /ˈprɑː.və.nəns/
**CEFR:** C2 | **Part of speech:** n. | **Occurrences:** 2

**EN:** the place of origin or earliest known history of something; a record of ownership or creation  
**CN:** 出处，来源，起源记录

**Original examples:**
- [52:57] It looks at the code **provenance**.  
  它查看代码的来源记录。
- [53:20] Every artifact is versioned. Versions are immutable, and each version has its own **provenance** attached to it.  
  每个工件都有版本。版本是不可变的，每个版本都附有自己的来源记录。

**Extra example:**
- The museum carefully documents the **provenance** of each artwork to verify its authenticity.  
  博物馆仔细记录每件艺术品的来源以验证其真实性。

### immutable  /ɪˈmjuː.tə.bəl/
**CEFR:** C2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** unchanging over time or unable to be changed  
**CN:** 不可变的，不可更改的

**Original examples:**
- [53:20] Versions are **immutable**, and each version has its own provenance attached to it.  
  版本是不可变的，每个版本都附有自己的来源记录。

**Extra example:**
- Blockchain technology relies on **immutable** records that cannot be altered after creation.  
  区块链技术依赖于创建后无法更改的不可变记录。

### beneficiary  /ˌben.əˈfɪʃ.ə.ri/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a person or organization that receives benefits, advantages, or profits from something  
**CN:** 受益者

**Original examples:**
- [01:10:19] So in the partnerships and deployment group, we are the **beneficiaries** like all of you of the models and capabilities that you've been hearing about.  
  在合作与部署小组，我们和大家一样都是这些模型和功能的受益者。

**Extra example:**
- The primary **beneficiaries** of this policy change will be small businesses.  
  这项政策变化的主要受益者将是小企业。

### frontier  /frʌnˈtɪr/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the extreme limit of understanding or achievement in a particular area  
**CN:** 前沿，边界

**Original examples:**
- [01:10:19] One, where are the **frontiers** of science that we believe we can have a positive impact on?  
  第一，我们认为能够产生积极影响的科学前沿在哪里？

**Extra example:**
- Machine learning is pushing the **frontiers** of what computers can achieve.  
  机器学习正在推动计算机能力的前沿边界。

### unearth  /ʌnˈɜːrθ/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to discover something hidden or previously unknown  
**CN:** 发掘，发现

**Original examples:**
- [01:11:02] We've all probably experienced a new technology, whether it be CRISPR, next generation sequencing, or perhaps even a computational collaborator that **unearths** and opens up our mind to what is possible.  
  我们可能都体验过新技术，无论是CRISPR、下一代测序，还是可能是一个计算协作者，它发掘并开启了我们对可能性的认知。

**Extra example:**
- The investigation **unearthed** evidence of systematic fraud.  
  调查发掘出系统性欺诈的证据。

### expansive  /ɪkˈspæn.sɪv/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** covering a wide area; broad in scope or extent  
**CN:** 广泛的，全面的

**Original examples:**
- [01:11:34] And it's really a privilege to have an incredible group that come at the pharma industry from very different perspectives but have a similarly **expansive** view.  
  能有这样一个了不起的团队真是一种荣幸，他们从非常不同的角度看待制药行业，但都有同样广阔的视野。

**Extra example:**
- The CEO presented an **expansive** vision for the company's future.  
  首席执行官提出了公司未来的宏大愿景。

### inherent  /ɪnˈher.ənt/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** existing as a natural or essential part of something  
**CN:** 固有的，内在的

**Original examples:**
- [01:13:01] So I think it's worth understanding why biology in general is difficult. Drug R&D is worse than the general difficulties of biology, and it comes from certain kind of **inherent** properties of biology, chemistry, and that world.  
  所以我认为值得理解为什么生物学总体上是困难的。药物研发比生物学的一般困难更糟糕，这来自生物学、化学及其领域的某些固有属性。

**Extra example:**
- There are **inherent** risks in any investment strategy.  
  任何投资策略都存在固有风险。

### multiscale  /ˌmʌl.tiˈskeɪl/
**CEFR:** C2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** involving or operating at multiple levels or scales simultaneously  
**CN:** 多尺度的

**Original examples:**
- [01:13:37] So that's problem number one. It's not the only one. Second problem is it's **multiscale**.  
  所以这是第一个问题。但不是唯一的。第二个问题是它是多尺度的。

**Extra example:**
- Climate modeling requires **multiscale** approaches that span from local weather to global patterns.  
  气候建模需要多尺度方法，涵盖从局部天气到全球模式。

### wade  /weɪd/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to move or proceed through something with difficulty  
**CN:** 艰难行进，费力处理

**Original examples:**
- [01:14:09] And then on top of all of those things, that piece that people often call complexity is within this vastness having to **wade** through it.  
  然后在所有这些之上，人们通常称之为复杂性的那部分，就是在这种浩瀚中必须艰难跋涉。

**Extra example:**
- Researchers must **wade** through thousands of pages of data to find relevant patterns.  
  研究人员必须费力地处理数千页数据才能找到相关模式。

### automate  /ˈɔː.tə.meɪt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to make a process or system operate automatically without human intervention  
**CN:** 使自动化

**Original examples:**
- [01:14:17] You can't **automate** that because it's not exactly the same way.  
  你无法将其自动化，因为它并非完全相同的方式。

**Extra example:**
- The factory plans to **automate** 80% of its assembly line processes.  
  工厂计划将80%的装配线流程自动化。

### dimensionality  /dɪˌmen.ʃəˈnæl.ə.ti/
**CEFR:** C2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the number of independent parameters or variables in a mathematical space or dataset  
**CN:** 维度

**Original examples:**
- [01:14:30] Now you look at AI. So when you have really huge spaces kind of nominally, but actually the real **dimensionality** is lower, well AI is great at that.  
  现在看看AI。当你有名义上非常巨大的空间，但实际的维度较低时，AI在这方面表现出色。

**Extra example:**
- Reducing the **dimensionality** of the data helps improve model performance.  
  降低数据的维度有助于提高模型性能。

### nonlinear  /nɑːnˈlɪn.i.ɚ/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** not forming or following a straight line; having relationships that are not proportional  
**CN:** 非线性的

**Original examples:**
- [01:14:42] When you look at things that are multiscale and as you move from one scale to another you have some **nonlinear** transformation, AI has proven itself being really great at that.  
  当你观察多尺度的事物，当你从一个尺度移动到另一个尺度时，会有一些非线性转换，AI已经证明自己在这方面非常出色。

**Extra example:**
- The relationship between input and output in neural networks is **nonlinear**.  
  神经网络中输入和输出之间的关系是非线性的。

### transformation  /ˌtræns.fɚˈmeɪ.ʃən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 4

**EN:** a thorough or dramatic change in form, appearance, or character  
**CN:** 转变，变革

**Original examples:**
- [01:14:42] When you look at things that are multiscale and as you move from one scale to another you have some nonlinear **transformation**, AI has proven itself being really great at that.  
  当你观察多尺度的事物，当你从一个尺度移动到另一个尺度时，会有一些非线性转换，AI已经证明自己在这方面非常出色。
- [01:18:02] You've been outspoken and talked a lot about how at BMS you don't want AI simply to accelerate the current processes, but actually **transform** how you operate and do work.  
  你一直直言不讳地谈论在BMS，你们不希望AI只是加速当前的流程，而是真正改变你们的运营和工作方式。
- [01:18:45] Well look, I mean at a macro level, our view is that this technology is ultimately going to **transform** every piece of the value chain.  
  好吧，我是说从宏观层面来看，我们的观点是这项技术最终将改变价值链的每个环节。
- [01:44:38] And Voss, Aviv, and Chris just walked us through how they're charting the path of AI **transformation** at BMS, Genentech, and Artis.  
  Voss、Aviv和Chris刚刚向我们介绍了他们如何在BMS、Genentech和Artis规划AI转型的路径。

**Extra example:**
- Digital **transformation** requires changes to both technology and organizational culture.  
  数字化转型需要改变技术和组织文化两个方面。

### agent  /ˈeɪ.dʒənt/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** an autonomous entity that perceives its environment and takes actions to achieve goals  
**CN:** 代理，智能体

**Original examples:**
- [01:14:58] And that last piece, which is kind of winging through a world that is not exactly specified, although there's kind of a playbook, but it's not precise. Well, AI **agents** are actually great at that.  
  最后一部分，即在一个没有完全明确定义的世界中摸索前进，虽然有某种行动手册，但并不精确。AI智能体实际上在这方面非常擅长。

**Extra example:**
- AI **agents** can learn from their environment and adapt their behavior over time.  
  AI智能体可以从环境中学习并随着时间调整其行为。

### iteration  /ˌɪt.əˈreɪ.ʃən/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 2

**EN:** the repetition of a process or procedure, typically to improve results or approach a desired outcome  
**CN:** 迭代

**Original examples:**
- [01:15:26] And models and it needs GPUs, but that's not enough. It needs a lot of data and it needs really the ability to understand whether what it did means anything, and that means that it needs **iterations**.  
  还需要模型和GPU，但这还不够。它需要大量数据，而且真正需要能够理解它所做的事情是否有意义，这意味着它需要迭代。
- [01:15:56] So it will allow you to **iterate**, repeat, etc., etc. In some way, it's exactly what biologists and scientists have done all along.  
  所以它将允许你迭代、重复等等。在某种程度上，这正是生物学家和科学家一直在做的事情。

**Extra example:**
- Through multiple **iterations**, the team refined the algorithm's accuracy to 95%.  
  通过多次迭代，团队将算法的准确率提高到95%。

### worldview  /ˈwɜːrld.vjuː/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a particular philosophy or way of viewing and understanding the world  
**CN:** 世界观，观念

**Original examples:**
- [01:15:48] Your models hold the answer, not the data, but the models hold the answer and you have to shift to that **worldview**.  
  你的模型拥有答案，不是数据，而是模型拥有答案，你必须转变到这种世界观。

**Extra example:**
- The scientific revolution fundamentally changed humanity's **worldview**.  
  科学革命从根本上改变了人类的世界观。

### generative  /ˈdʒen.ə.reɪ.tɪv/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** able to produce or create new content, ideas, or solutions  
**CN:** 生成的，创造性的

**Original examples:**
- [01:16:22] And I think Dario touched on this earlier, but like there is a very unique shape of **generative** biology that I think matches this space in biology so well.  
  我认为Dario之前提到过这一点，但生成生物学有一种非常独特的形态，我认为它与生物学领域非常匹配。

**Extra example:**
- **Generative** AI models can create realistic images from text descriptions.  
  生成式AI模型可以根据文本描述创建逼真的图像。

### abound  /əˈbaʊnd/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to exist in large numbers or amounts  
**CN:** 大量存在，充足

**Original examples:**
- [01:16:45] Biology and chemistry and so on, they have a huge long tail. So people love focusing on examples that are in the areas where data **abound** and models perform well.  
  生物学和化学等等，它们有一个巨大的长尾。所以人们喜欢关注数据充足、模型表现良好的领域的例子。

**Extra example:**
- In Silicon Valley, opportunities for tech startups **abound**.  
  在硅谷，科技初创公司的机会比比皆是。

### mechanism  /ˈmek.ə.nɪ.zəm/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 2

**EN:** a system or process by which something operates or is brought about  
**CN:** 机制，机理

**Original examples:**
- [01:17:05] We have a nice example like that in oncology. The outcome was what I would call an alien. No human would have thought of that. Humans say the same, but that was just the starting point. When you needed to figure out the **mechanism** of action, you actually needed the kind of experimental science and thinking that AI cannot help you with right now.  
  我们在肿瘤学中有一个很好的例子。结果是我称之为外星的东西。没有人会想到那个。人类说同样的话，但那只是起点。当你需要弄清楚作用机制时，你实际上需要AI目前无法帮助你的那种实验科学和思考。
- [01:21:13] And actually, when you look at actual **mechanisms** of action that are notable, it's actually quite small.  
  实际上，当你看到真正值得注意的作用机制时，它实际上相当少。

**Extra example:**
- Understanding the **mechanism** of disease transmission is crucial for prevention.  
  理解疾病传播的机制对预防至关重要。

### bespoke  /bɪˈspoʊk/
**CEFR:** C2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** custom-made or tailored to particular requirements  
**CN:** 定制的，量身定做的

**Original examples:**
- [01:17:16] There's no data like that. The experiments are all small scale, they're very **bespoke**, very specific, very imaginative, and that's great.  
  没有那样的数据。实验都是小规模的，非常定制化，非常具体，非常有想象力，这很好。

**Extra example:**
- The consulting firm provides **bespoke** solutions for each client's unique challenges.  
  这家咨询公司为每个客户的独特挑战提供定制化解决方案。

### comprehensive  /ˌkɑːm.prɪˈhen.sɪv/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 3

**EN:** including or dealing with all or nearly all elements or aspects  
**CN:** 全面的，综合的

**Original examples:**
- [01:17:35] I think it can one day help you get much closer, but what it does let you do is it lets you work in a **comprehensive** way.  
  我认为有一天它可以帮助你更接近，但它确实让你能够以全面的方式工作。
- [01:17:54] So on the tone of **comprehensiveness** and also, you know, the breadth of tasks, Chris, I'd like to transition to you.  
  那么在全面性以及任务的广度方面，Chris，我想转向你。
- [01:25:33] We look actually as an industry at a very small sliver of them because of this difficulty to manage the **comprehensiveness**.  
  作为一个行业，我们实际上只关注其中很小的一部分，因为很难管理全面性。

**Extra example:**
- The report provides a **comprehensive** analysis of market trends.  
  该报告对市场趋势进行了全面分析。

### outspoken  /aʊtˈspoʊ.kən/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** frank and direct in expressing one's opinions, especially if controversial  
**CN:** 直言不讳的，坦率的

**Original examples:**
- [01:18:02] You've been **outspoken** and talked a lot about how at BMS you don't want AI simply to accelerate the current processes, but actually transform how you operate and do work.  
  你一直直言不讳地谈论在BMS，你们不希望AI只是加速当前的流程，而是真正改变你们的运营和工作方式。

**Extra example:**
- The CEO is **outspoken** about the need for industry-wide sustainability standards.  
  首席执行官直言不讳地表示需要全行业的可持续性标准。

### impart  /ɪmˈpɑːrt/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to communicate information, knowledge, or a particular quality to someone  
**CN:** 传授，告知，赋予

**Original examples:**
- [01:26:54] Would kind of **impart** and, you know, set those expectations at the right time?  
  会以某种方式传达并在合适的时机设定这些期望吗？

**Extra example:**
- Great teachers **impart** not just knowledge but also critical thinking skills.  
  优秀的教师不仅传授知识，还传授批判性思维技能。

### standpoint  /ˈstændpɔɪnt/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 3

**EN:** a particular perspective or position from which something is considered  
**CN:** 立场，观点

**Original examples:**
- [01:26:57] Well, maybe I'll start from our **standpoint**.  
  好吧，也许我会从我们的立场开始。
- [01:38:00] Maybe the only thing I'd add is from a public policy **standpoint**.  
  也许我唯一要补充的是从公共政策的角度来看。
- [01:41:10] From my **standpoint**, I just build on what Vos said.  
  从我的角度来看，我只是在Vos所说的基础上进行补充。

**Extra example:**
- From a business **standpoint**, this investment makes perfect sense.  
  从商业角度来看，这项投资非常合理。

### intention  /ɪnˈtenʃən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** an aim or plan; what someone means to do  
**CN:** 意图，目的

**Original examples:**
- [01:27:02] When we started rolling out this technology, our **intention** was to let a thousand flowers bloom.  
  当我们开始推出这项技术时，我们的意图是让百花齐放。

**Extra example:**
- The company's **intention** is to expand into Asian markets within two years.  
  公司的意图是在两年内扩展到亚洲市场。

### scale  /skeɪl/
**CEFR:** B2 | **Part of speech:** v./n. | **Occurrences:** 6

**EN:** to grow or expand in size, scope, or capability; the size or extent of something  
**CN:** 扩展，规模化；规模

**Original examples:**
- [01:27:13] But what we found was that the use cases tended to be quite narrow and were very difficult to **scale**.  
  但我们发现，这些用例往往非常狭窄，很难扩展。
- [01:27:34] And so we really were struggling with **scaling**.  
  所以我们在规模化方面确实遇到了困难。
- [01:27:53] but we couldn't **scale** it.  
  但我们无法将其规模化。
- [01:28:09] and you simply couldn't get the **scale** we needed.  
  你根本无法获得我们需要的规模。
- [01:28:29] We put small teams that had six to eight weeks to prove out the concept and then we pushed to **scale** that  
  我们组建了小团队，给他们六到八周的时间来证明概念，然后我们推动将其规模化
- [01:33:43] And with that **scale**, of course, also comes difficulties  
  当然，伴随着这种规模，也会出现困难

**Extra example:**
- The startup struggled to **scale** its operations after securing Series B funding.  
  这家初创公司在获得B轮融资后，在扩展运营方面遇到了困难。

### individualized  /ˌɪndɪˈvɪdʒuəlaɪzd/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** designed or adapted for a particular person or individual needs  
**CN:** 个性化的，针对个人的

**Original examples:**
- [01:27:24] Because you're developing these tools for a very **individualized** case.  
  因为你正在为一个非常个性化的案例开发这些工具。

**Extra example:**
- The platform offers **individualized** learning paths based on each student's performance.  
  该平台根据每个学生的表现提供个性化的学习路径。

### mundane  /mʌnˈdeɪn/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** ordinary, routine, and not interesting or exciting  
**CN:** 平凡的，单调的，世俗的

**Original examples:**
- [01:27:34] I'll give you one sort of somewhat **mundane** example  
  我会给你举一个有点平凡的例子

**Extra example:**
- AI can automate **mundane** tasks, freeing employees to focus on strategic work.  
  人工智能可以自动化单调的任务，让员工专注于战略性工作。

### forecast  /ˈfɔːrkæst/
**CEFR:** B2 | **Part of speech:** v./n. | **Occurrences:** 2

**EN:** to predict or estimate future events or trends; a prediction  
**CN:** 预测；预报

**Original examples:**
- [01:27:34] but we knew two years ago that AI could predict and **forecast** our business  
  但我们两年前就知道人工智能可以预测和预报我们的业务
- [01:27:53] when you talk about changing the **forecast** and planning process  
  当你谈到改变预测和规划流程时

**Extra example:**
- Analysts **forecast** a 15% revenue growth for the tech sector this quarter.  
  分析师预测本季度科技行业的收入将增长15%。

### battalion  /bəˈtæljən/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a large organized group (originally military; here used figuratively for a large number of people)  
**CN:** 大批，大量（原指军队编制单位）

**Original examples:**
- [01:27:34] better than the **battalions** of people that we had actually doing the forecasting  
  比我们实际从事预测工作的大批人员要好

**Extra example:**
- The company deployed **battalions** of engineers to fix the security vulnerability.  
  公司部署了大批工程师来修复安全漏洞。

### supplement  /ˈsʌplɪmənt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to add something extra to improve or complete something  
**CN:** 补充，增补

**Original examples:**
- [01:28:09] So what we realized very quickly is that we needed to **supplement** this sort of bottom-up innovation with a very robust and...  
  所以我们很快意识到，我们需要用非常稳健的……来补充这种自下而上的创新

**Extra example:**
- We need to **supplement** our existing data with customer interviews.  
  我们需要用客户访谈来补充我们现有的数据。

### robust  /roʊˈbʌst/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** strong, effective, and able to withstand difficult conditions  
**CN:** 强健的，稳健的，可靠的

**Original examples:**
- [01:28:09] we needed to supplement this sort of bottom-up innovation with a very **robust** and...  
  我们需要用非常稳健的……来补充这种自下而上的创新

**Extra example:**
- The team developed a **robust** testing framework to catch bugs early.  
  团队开发了一个稳健的测试框架来及早发现错误。

### rigorous  /ˈrɪɡərəs/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 2

**EN:** extremely thorough, careful, and precise  
**CN:** 严格的，严谨的，缜密的

**Original examples:**
- [01:28:17] **rigorous** top-down approach.  
  严格的自上而下的方法。
- [01:42:43] So as a scientist, I always feel compelled and it's most critical to be both **rigorous** and really honest with these new technologies  
  作为一名科学家，我总是感到有责任，最关键的是对这些新技术既要严谨又要诚实

**Extra example:**
- The research underwent **rigorous** peer review before publication.  
  这项研究在发表前经历了严格的同行评审。

### accelerator  /əkˈseləreɪtər/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a program or mechanism that speeds up development or growth  
**CN:** 加速器（指促进发展的项目或机制）

**Original examples:**
- [01:28:17] So we created this AI **accelerator** in the company.  
  所以我们在公司内创建了这个人工智能加速器。

**Extra example:**
- The startup joined a three-month **accelerator** program to refine its product.  
  这家初创公司加入了一个为期三个月的加速器项目来完善其产品。

### vertical  /ˈvɜːrtɪkəl/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a specific industry sector or business area  
**CN:** （行业）垂直领域

**Original examples:**
- [01:28:17] We had my team say where within the key **verticals** in the organization can you get real productivity improvements?  
  我让我的团队说，在组织的关键垂直领域中，哪里可以获得真正的生产力提升？

**Extra example:**
- Our company focuses on three main **verticals**: healthcare, finance, and education.  
  我们公司专注于三个主要垂直领域：医疗保健、金融和教育。

### leverage  /ˈlevərɪdʒ/
**CEFR:** C1 | **Part of speech:** v. | **Occurrences:** 3

**EN:** to use something to maximum advantage; to apply force or influence strategically  
**CN:** 利用，发挥……的作用

**Original examples:**
- [01:28:26] Where can you change the processes **leveraging** this technology?  
  你在哪里可以利用这项技术来改变流程？
- [01:28:48] So this ability to take bottoms up and top down approaches to these problems so that you can overcome the organizational and people issues associated with fully **leveraging** this technology is super important.  
  因此，采取自下而上和自上而下的方法来解决这些问题的能力非常重要，这样你就可以克服与充分利用这项技术相关的组织和人员问题。
- [01:38:28] And at the same time, I think we're going to have to modernize our drug regulatory system to fully **leverage** these technologies.  
  与此同时，我认为我们将不得不使我们的药物监管系统现代化，以充分利用这些技术。

**Extra example:**
- The team plans to **leverage** AI to automate customer support responses.  
  团队计划利用人工智能来自动化客户支持响应。

### incubator  /ˈɪŋkjubeɪtər/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a program or facility that supports the development of new projects or companies  
**CN:** 孵化器

**Original examples:**
- [01:28:29] We put small teams that had six to eight weeks to prove out the concept and then we pushed to scale that and so we have I think 40 or 50 projects in the **incubator** now.  
  我们组建了小团队，给他们六到八周的时间来证明概念，然后我们推动将其规模化，所以我们现在在孵化器中有大约40或50个项目。

**Extra example:**
- The tech **incubator** provides mentorship and seed funding to early-stage startups.  
  这个科技孵化器为早期初创公司提供指导和种子资金。

### complement  /ˈkɑːmplɪment/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to add to something in a way that enhances or improves it  
**CN:** 补充，补足

**Original examples:**
- [01:29:04] So before development just to **complement** what Chris described  
  所以在开发之前，只是为了补充Chris所描述的内容

**Extra example:**
- The new feature will **complement** our existing product line perfectly.  
  新功能将完美地补充我们现有的产品线。

### reshape  /riːˈʃeɪp/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to change the structure, organization, or form of something fundamentally  
**CN:** 重塑，改造

**Original examples:**
- [01:29:04] is actually what we would call **reshaping** of a process.  
  实际上就是我们所说的流程重塑。

**Extra example:**
- Digital transformation is **reshaping** the entire retail industry.  
  数字化转型正在重塑整个零售行业。

### hypothesis  /haɪˈpɑːθəsɪs/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a proposed explanation or assumption that can be tested  
**CN:** 假设，假说

**Original examples:**
- [01:29:19] But on the research side, we didn't come from a million flowers bloom. We came from a very particular **hypothesis** that is now years old.  
  但在研究方面，我们不是从百花齐放开始的。我们来自一个非常特定的假设，这个假设已经有好几年了。

**Extra example:**
- Our **hypothesis** is that users will engage more with personalized content.  
  我们的假设是，用户会更多地参与个性化内容。

### pivot  /ˈpɪvət/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to change direction or strategy, especially in business or projects  
**CN:** 转向，转变策略

**Original examples:**
- [01:30:06] So we had to **pivot** for that  
  所以我们不得不为此转向

**Extra example:**
- After initial user feedback, the startup decided to **pivot** from B2C to B2B.  
  在收到最初的用户反馈后，这家初创公司决定从B2C转向B2B。

### initiative  /ɪˈnɪʃətɪv/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a new plan or program intended to solve a problem or achieve a goal  
**CN:** 倡议，计划，措施

**Original examples:**
- [01:30:13] And we called something foundational datasets for foundation models as an **initiative**  
  我们将某项工作称为基础模型的基础数据集，作为一项倡议

**Extra example:**
- The company launched a sustainability **initiative** to reduce carbon emissions.  
  公司启动了一项可持续发展倡议以减少碳排放。

### generalize  /ˈdʒenərəlaɪz/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 2

**EN:** to make something applicable to a broader range of cases; to derive general principles  
**CN:** 推广，归纳，概括

**Original examples:**
- [01:30:48] And now our goal is not to learn how to do the process in general, and rightfully so, because we want to change the lives of patients in the end. It's not to make the model general. It's not to learn lessons. It's to actually get a project into patients.  
  现在我们的目标不是学习如何泛化地执行流程，这是理所当然的，因为我们最终想要改变患者的生活。不是要让模型通用化。不是要吸取教训。而是要真正让项目惠及患者。
- [01:31:04] How do you still move all your projects but also **generalize** from it?  
  你如何既推进所有项目又从中进行归纳？

**Extra example:**
- The team is working to **generalize** the model so it works across different datasets.  
  团队正在努力推广该模型，使其适用于不同的数据集。

### synthesis  /ˈsɪnθəsɪs/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the combination of components to form a connected whole; in chemistry, creating compounds  
**CN:** 综合，合成

**Original examples:**
- [01:31:04] why should I spend **synthesis** money just to make the model better?  
  为什么我要花合成的钱只是为了让模型变得更好？

**Extra example:**
- The report provides a **synthesis** of findings from multiple research studies.  
  该报告对多项研究的发现进行了综合。

### preclinical  /priːˈklɪnɪkəl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 2

**EN:** relating to the testing phase before human clinical trials  
**CN:** 临床前的

**Original examples:**
- [01:32:04] So setting really good experiments whether it's in **pre-clinical** or in early clinical  
  所以设计真正好的实验，无论是在临床前还是在早期临床
- [01:38:28] If we actually can better predict **preclinical** safety  
  如果我们确实能够更好地预测临床前安全性

**Extra example:**
- The drug is currently in **preclinical** testing with promising results.  
  该药物目前正在进行临床前测试，结果很有希望。

### unvarnished  /ʌnˈvɑːr.nɪʃt/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** presented plainly and honestly, without concealment or embellishment  
**CN:** 未加修饰的，坦率的，直截了当的

**Original examples:**
- [01:44:17] Eric and D— sorry, Lattice and Dario walked us through their **unvarnished** views on how compression may or may not happen, the challenges and opportunity ahead.  
  Eric和D——抱歉，Lattice和Dario向我们介绍了他们关于压缩可能如何发生或不发生的坦率看法，以及未来的挑战和机遇。

**Extra example:**
- The CEO gave an **unvarnished** assessment of the company's financial difficulties.  
  首席执行官对公司的财务困难进行了坦率的评估。

### compression  /kəmˈpreʃ.ən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the process of reducing something in size, scope, or duration; in AI/tech, can refer to model compression or data compression  
**CN:** 压缩；缩减

**Original examples:**
- [01:44:17] Eric and D— sorry, Lattice and Dario walked us through their unvarnished views on how **compression** may or may not happen, the challenges and opportunity ahead.  
  Eric和D——抱歉，Lattice和Dario向我们介绍了他们关于压缩可能如何发生或不发生的坦率看法，以及未来的挑战和机遇。

**Extra example:**
- The new algorithm achieves 50% **compression** without losing data quality.  
  新算法在不损失数据质量的情况下实现了50%的压缩。

### workbench  /ˈwɜːrk.bentʃ/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** a work surface or integrated software environment where specific tasks are performed; in tech, often refers to a development or analysis platform  
**CN:** 工作台；（软件）工作平台

**Original examples:**
- [01:44:25] Eric and Alec walked you through the basis of our claim: models that meaningfully improve in biology with every release, and a **workbench** that runs the analysis your scientists actually need.  
  Eric和Alec向您介绍了我们主张的基础：每次发布都在生物学方面有实质性改进的模型，以及一个能运行科学家实际需要的分析的工作平台。

**Extra example:**
- The data science **workbench** provides all the tools needed for model training and evaluation.  
  数据科学工作平台提供了模型训练和评估所需的所有工具。

### reproducible  /ˌriː.prəˈduː.sə.bəl/
**CEFR:** C1 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** able to be replicated or repeated with consistent results; essential concept in scientific research  
**CN:** 可重现的，可复制的

**Original examples:**
- [01:44:38] Run pipelines, figures, every step **reproducible** and traceable.  
  运行流程、生成图表，每一步都可重现和可追溯。

**Extra example:**
- **Reproducible** research practices are essential for validating scientific findings.  
  可重现的研究实践对于验证科学发现至关重要。

### traceable  /ˈtreɪ.sə.bəl/
**CEFR:** B2 | **Part of speech:** adj. | **Occurrences:** 1

**EN:** able to be tracked or followed through a process; can be traced back to its origin or through its steps  
**CN:** 可追溯的，可追踪的

**Original examples:**
- [01:44:38] Run pipelines, figures, every step reproducible and **traceable**.  
  运行流程、生成图表，每一步都可重现和可追溯。

**Extra example:**
- All changes to the codebase are **traceable** through our version control system.  
  代码库的所有更改都可以通过我们的版本控制系统进行追溯。

### chart  /tʃɑːrt/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to plan or map out a course of action; to navigate or plot a path forward  
**CN:** 规划，制定（路线或计划）

**Original examples:**
- [01:44:38] And Voss, Aviv, and Chris just walked us through how they're **charting** the path of AI transformation at BMS, Genentech, and Artis.  
  Voss、Aviv和Chris刚刚向我们介绍了他们如何在BMS、Genentech和Artis规划AI转型的路径。

**Extra example:**
- The leadership team is **charting** a new strategic direction for the company.  
  领导团队正在为公司规划新的战略方向。

### collaboration  /kəˌlæb.əˈreɪ.ʃən/
**CEFR:** B2 | **Part of speech:** n. | **Occurrences:** 1

**EN:** the action of working together with others toward a shared goal  
**CN:** 合作，协作

**Original examples:**
- [01:44:54] To build an intelligence platform where scientific discovery truly happens is going to be a **collaboration** with the industry and our partners represented here.  
  要构建一个真正实现科学发现的智能平台，需要与行业以及在座的合作伙伴进行合作。

**Extra example:**
- The project's success depended on close **collaboration** between the engineering and design teams.  
  该项目的成功依赖于工程团队和设计团队之间的密切合作。

### hackathon  /ˈhæk.ə.θɑːn/
**CEFR:** C1 | **Part of speech:** n. | **Occurrences:** 1

**EN:** an event where programmers and others collaborate intensively on software projects, typically over a short period  
**CN:** 黑客马拉松，编程马拉松（一种集中式的软件开发活动）

**Original examples:**
- [01:45:23] And secondly, join us for a **hackathon** with the Gladstone Institutes from July 6th to the 12th, our first ever that'll feature Claude Science.  
  其次，请加入我们与Gladstone研究所合作举办的黑客马拉松，从7月6日到12日，这是我们首次以Claude Science为特色的活动。

**Extra example:**
- The **hackathon** attracted over 200 developers who built innovative prototypes in just 48 hours.  
  这次黑客马拉松吸引了200多名开发者，他们在短短48小时内就构建出了创新原型。

### feature  /ˈfiː.tʃər/
**CEFR:** B2 | **Part of speech:** v. | **Occurrences:** 1

**EN:** to include or prominently display something as a special attraction or important element  
**CN:** 以……为特色，重点展示

**Original examples:**
- [01:45:23] And secondly, join us for a hackathon with the Gladstone Institutes from July 6th to the 12th, our first ever that'll **feature** Claude Science.  
  其次，请加入我们与Gladstone研究所合作举办的黑客马拉松，从7月6日到12日，这是我们首次以Claude Science为特色的活动。

**Extra example:**
- The conference will **feature** keynote speakers from leading tech companies.  
  会议将重点邀请来自领先科技公司的主旨演讲者。

### virtually  /ˈvɜːr.tʃu.ə.li/
**CEFR:** B2 | **Part of speech:** adv. | **Occurrences:** 1

**EN:** by means of virtual technology or online platforms; through remote digital connection rather than physical presence  
**CN:** 通过虚拟方式，在线地，远程地

**Original examples:**
- [01:45:46] To everyone who joined us here today, both in person and **virtually**, thank you.  
  感谢今天加入我们的每一位，无论是现场参加还是在线参与。

**Extra example:**
- The team meeting will be held **virtually** to accommodate colleagues in different time zones.  
  团队会议将以虚拟方式举行，以方便不同时区的同事参加。

---

## Useful Phrases

### sit down with
**Type:** phrasal_verb

**EN:** to have a formal meeting or conversation with someone  
**CN:** 与某人坐下来交谈，进行正式会谈

**Original examples:**
- [03:46] First, Dario is going to **sit down with** the scientist who turned GLP-1 into a medicine about compressing timelines in biology.  
  首先，Dario将与把GLP-1转化为药物的科学家坐下来交谈，讨论压缩生物学时间线的问题。

**Extra example:**
- Let's **sit down with** the team next week to discuss the project timeline.  
  我们下周和团队坐下来讨论一下项目时间表吧。

### wrap with
**Type:** phrasal_verb

**EN:** to conclude or finish with something  
**CN:** 以...结束，用...收尾

**Original examples:**
- [04:01] And finally, we'll **wrap with** three leading lights from the industry sharing with us how AI has transformed their organizations.  
  最后，我们将以三位行业领军人物的分享作为收尾，他们将讲述AI如何改变了他们的组织。

**Extra example:**
- We'll **wrap with** a Q&A session at the end of the presentation.  
  我们将以问答环节作为演讲的收尾。

### set off
**Type:** phrasal_verb

**EN:** to cause something to start or begin, especially a chain of events  
**CN:** 引发，开启（尤其指一系列事件）

**Original examples:**
- [06:06] She **set off** one of the biggest medical revolutions I've seen in a 25-year career.  
  她引发了我25年职业生涯中见过的最大的医学革命之一。

**Extra example:**
- The discovery **set off** a wave of research in the field.  
  这一发现引发了该领域的研究热潮。

### on the brink of
**Type:** collocation

**EN:** at the point when something is about to happen  
**CN:** 在...的边缘，即将...

**Original examples:**
- [07:13] You've seen revolutions in medicine. You also know as well as anyone how the existing system works and what slows it down. Are we **on the brink of** a big change?  
  你见证过医学革命。你也比任何人都清楚现有系统如何运作以及什么在拖慢它。我们是否正处在重大变革的边缘？

**Extra example:**
- The company is **on the brink of** bankruptcy after years of losses.  
  经过多年亏损，这家公司正处在破产的边缘。

### inflection point
**Type:** collocation

**EN:** a time of significant change in a situation; a turning point  
**CN:** 拐点，转折点（形势发生重大变化的时刻）

**Original examples:**
- [07:13] I really do believe we are at a real **inflection point**, right, where there's so much data becoming available.  
  我真的相信我们正处在一个真正的拐点，有如此多的数据变得可用。
- [08:10] And I've experienced both **inflection points** as well as hype in my world, right?  
  在我的领域里，我既经历过拐点，也经历过炒作，对吧？

**Extra example:**
- The introduction of smartphones was an **inflection point** for mobile computing.  
  智能手机的推出是移动计算的一个拐点。

### put a number on
**Type:** idiom

**EN:** to quantify or give a specific numerical value to something  
**CN:** 给出具体数字，量化

**Literal:** 把数字放在某物上  
**Figurative EN:** to assign a specific numerical value or estimate to something  
**Figurative CN:** 为某事物给出具体的数值或估计

**Original examples:**
- [12:11] So I think it's hard to **put a number on**, right?  
  所以我认为很难给出具体数字，对吧？

**Extra example:**
- It's difficult to **put a number on** the value of good customer service.  
  很难量化良好客户服务的价值。

### work around
**Type:** phrasal_verb

**EN:** to find a way to avoid or deal with a problem or obstacle  
**CN:** 绕过，找到变通办法

**Original examples:**
- [14:51] I think there are a lot of things we can do to speed it up or to **work around** it, but I think it's always going to be the limiting factor.  
  我认为有很多事情我们可以做来加速它或绕过它，但我认为它始终会是限制因素。
- [15:17] So there's this inertia, but I think there's lots of things we can do to **work around** it or to shorten the cycle time.  
  所以存在这种惯性，但我认为有很多事情我们可以做来绕过它或缩短周期时间。

**Extra example:**
- We found a clever way to **work around** the software bug until the patch is released.  
  在补丁发布之前，我们找到了一个巧妙的方法来绕过这个软件bug。

### out the other end of the pipe
**Type:** idiom

**EN:** as the final result or output of a process  
**CN:** 作为流程的最终产出或结果

**Literal:** 从管道的另一端出来  
**Figurative EN:** as the final output or result after going through a complete process or pipeline  
**Figurative CN:** 经过完整流程或管道后的最终产出或结果

**Original examples:**
- [15:03] We can't get something **out the other end of the pipe**. There's just absolutely no way to do it.  
  我们无法在流程的另一端得到成果。根本没办法做到。

**Extra example:**
- Even with faster development, we need clinical trials before anything comes **out the other end of the pipe**.  
  即使开发速度更快，在任何成果从流程末端产出之前，我们仍需要临床试验。

### work out
**Type:** phrasal_verb

**EN:** to succeed or have a good result  
**CN:** 成功；有好的结果

**Original examples:**
- [18:44] We don't know for sure if that's going to **work out**, but I think we're seeing signs that... we're seeing the beginnings of it.  
  我们不确定这是否会成功，但我认为我们看到了一些迹象……我们看到了它的开端。

**Extra example:**
- I hope the new treatment plan **works out** for the patients.  
  我希望新的治疗方案对患者有效。

### shoot for
**Type:** phrasal_verb

**EN:** to aim for; to try to achieve something  
**CN:** 力争；以…为目标

**Original examples:**
- [18:52] That is what we should **shoot for**.  
  这就是我们应该追求的目标。

**Extra example:**
- We should **shoot for** at least 80% accuracy in our predictions.  
  我们应该力争预测准确率至少达到80%。

### move forward
**Type:** phrasal_verb

**EN:** to progress or advance  
**CN:** 推进；取得进展

**Original examples:**
- [19:25] So, but there will be criticism until some examples have **moved forward** and people are going to say, oh this is not fully AI designed, but that's again, that's not the point.  
  但在一些例子取得进展之前，会一直有批评，人们会说，哦，这不是完全由AI设计的，但那不是重点。

**Extra example:**
- Let's **move forward** with the clinical trial once we have approval.  
  一旦获得批准，我们就推进临床试验。

### improve upon
**Type:** phrasal_verb

**EN:** to make something better than it was before  
**CN:** 改进；改善

**Original examples:**
- [19:36] The point is that there's so many things you need to know in order to choose the right medicines and develop the right medicines, and all the areas of it can be **improved upon**.  
  关键是，为了选择和开发正确的药物，你需要知道很多事情，而所有这些领域都可以改进。

**Extra example:**
- We need to **improve upon** our current drug discovery methods.  
  我们需要改进现有的药物发现方法。

### run wild
**Type:** idiom

**EN:** to spread or grow uncontrollably  
**CN:** 失控；不受控制地蔓延

**Literal:** 疯狂奔跑  
**Figurative EN:** to spread, develop, or behave in an uncontrolled manner  
**Figurative CN:** 不受控制地发展或传播

**Original examples:**
- [20:34] I mean, the obvious worry is could all these cool technologies help people make bioweapons or synthetic organisms that **run wild**, or every science fiction novel we've read that actually could happen.  
  我的意思是，显而易见的担忧是，这些很酷的技术是否会帮助人们制造生物武器或失控的合成生物体，或者我们读过的每一本科幻小说中的情节真的会发生。

**Extra example:**
- If we don't regulate gene editing, modified organisms could **run wild** in the ecosystem.  
  如果我们不监管基因编辑，改良生物可能会在生态系统中失控蔓延。

### turn on
**Type:** phrasal_verb

**EN:** to become activated or start happening  
**CN:** 开始；激活

**Original examples:**
- [21:03] We've kind of entered an inflection point or a critical window, you know, as we go along the exponential. Different things in terms of both benefits and economically useful applications and, you know, potentially concerning applications **turn on** at different times.  
  随着指数级发展，我们进入了一个拐点或关键窗口期。无论是好处、经济上有用的应用，还是潜在令人担忧的应用，都会在不同时间点被激活。
- [21:18] And so we've had the benefit of seeing the window kind of **turn on** for cyber.  
  所以我们有幸看到了网络安全这个窗口期的开启。
- [21:22] It is not yet **turned on** for biology.  
  对于生物学来说，它还没有开启。

**Extra example:**
- Once the model reaches a certain threshold, new capabilities **turn on** rapidly.  
  一旦模型达到某个阈值，新的能力就会迅速开启。

### piggyback on
**Type:** phrasal_verb

**EN:** to use or build upon something that already exists  
**CN:** 依托；借助（现有的东西）

**Original examples:**
- [23:21] So, can we **piggyback on** those protocols where we say, 'Okay, the elements of society that already handle these potentially dangerous things, can we just say, okay, you guys already know how to do this.'  
  那么，我们能否依托这些协议，也就是说，'好的，社会中已经在处理这些潜在危险事物的机构，我们可以说，好的，你们已经知道怎么做了。'

**Extra example:**
- We can **piggyback on** existing safety protocols in pharmaceutical labs.  
  我们可以借助制药实验室现有的安全协议。

### not pan out
**Type:** phrasal_verb

**EN:** to fail to develop or succeed as expected  
**CN:** 未能成功，没有如预期那样发展

**Original examples:**
- [24:16] and also some of the worries about GLP-1s which have mostly **not panned out**.  
  还有一些关于GLP-1药物的担忧，这些担忧大多没有成为现实。

**Extra example:**
- The merger talks **didn't pan out** and both companies moved on.  
  合并谈判没有成功，两家公司各自继续发展。

### pan out
**Type:** phrasal_verb

**EN:** to develop or result in a particular way, especially successfully  
**CN:** 结果是；成功

**Original examples:**
- [24:36] How do you think about these risk problems? Yeah. So I think just so in the pharmaceutical industry, many people already know that we're very much used to balancing risk versus benefit, and I think it's just having a strong focus on it like Anthropic absolutely has, and it's even I guess put in your bylaws that how you are thinking about the benefit versus the risk. And I think just being focused on that, having a really strong view on how to evaluate that, and also having maybe an independent—so we have independent data monitoring committees in large clinical trials, maybe that's also something that might already be in place, but really having someone who is not motivated by money to actually look at that this is being dealt with properly. [Context: discussing worries about GLP-1s which have mostly not **panned out**]  
  [上下文：讨论关于GLP-1的担忧，这些担忧大多没有成真]

**Extra example:**
- The safety concerns about the new vaccine didn't **pan out** - it proved to be very safe.  
  关于新疫苗的安全担忧最终没有成真——它被证明非常安全。

### straddle the boundary
**Type:** collocation

**EN:** to be positioned between two different states or qualities  
**CN:** 处于两种状态或特性之间，在边界徘徊

**Original examples:**
- [27:48] In order to be creative, you're often **straddling the boundary** between making things up and having good ideas.  
  为了有创造力，你经常在虚构和产生好点子之间徘徊。

**Extra example:**
- The new policy **straddles the boundary** between innovation and regulation.  
  这项新政策在创新与监管之间寻求平衡。

### dogmatically attached
**Type:** collocation

**EN:** stubbornly committed to a belief or idea without questioning  
**CN:** 固执地坚持某种信念或观点，不愿质疑

**Original examples:**
- [28:47] Sometimes, you know, like humans, I imagine the AI models may get **dogmatically attached**.  
  有时候，就像人类一样，我想AI模型可能也会变得固执己见。

**Extra example:**
- Scientists shouldn't become **dogmatically attached** to their initial hypotheses.  
  科学家不应该固执地坚持自己最初的假设。

### lean in
**Type:** phrasal_verb

**EN:** to commit oneself fully to something; to engage actively  
**CN:** 全身心投入，积极参与

**Original examples:**
- [29:36] So people should be skeptical, absolutely, but they should **lean in** because that is how we can create more solutions for the work.  
  所以人们应该保持怀疑态度，这完全没问题，但他们应该积极投入，因为这样我们才能为工作创造更多解决方案。

**Extra example:**
- When facing challenges, successful leaders **lean in** rather than retreat.  
  面对挑战时，成功的领导者会积极应对而不是退缩。

### knock on the door
**Type:** idiom

**EN:** to be very close to achieving something  
**CN:** 接近实现某事，即将成功

**Literal:** 敲门  
**Figurative EN:** to be on the verge of achieving or reaching something  
**Figurative CN:** 即将实现或达到某事，处于突破的边缘

**Original examples:**
- [32:24] I think the models are just **knocking on the door** of where they can help a lot with that, and it's on a fast exponential.  
  我认为这些模型正处于能够在这方面提供巨大帮助的边缘，而且正在快速指数级增长。

**Extra example:**
- The team is **knocking on the door** of a major breakthrough in cancer research.  
  该团队即将在癌症研究方面取得重大突破。

### catch someone off guard
**Type:** idiom

**EN:** to surprise someone by happening unexpectedly  
**CN:** 使某人措手不及，出其不意

**Literal:** 使某人失去防备  
**Figurative EN:** to surprise or shock someone when they are not prepared  
**Figurative CN:** 在某人没有准备时让其感到惊讶或震惊

**Original examples:**
- [32:31] So you know exponentials really **catch you off guard**.  
  所以你知道，指数增长真的会让你措手不及。

**Extra example:**
- The sudden policy change **caught everyone off guard** in the industry.  
  这个突然的政策变化让业内所有人都措手不及。

### keep pace with
**Type:** collocation

**EN:** to move or progress at the same rate as something else  
**CN:** 跟上…的步伐，与…保持同步

**Original examples:**
- [32:52] Just the attention and the foresight to **keep pace with** the technology and just keep revisiting and understanding how fast it's improving.  
  只需要有关注度和前瞻性来跟上技术的步伐，不断重新审视并理解它改进的速度有多快。

**Extra example:**
- Small businesses struggle to **keep pace with** rapid digital transformation.  
  小企业难以跟上快速的数字化转型步伐。

### set in one's ways
**Type:** idiom

**EN:** to be fixed in one's habits and resistant to change  
**CN:** 固守习惯，墨守成规

**Literal:** 在某人的方式中固定下来  
**Figurative EN:** to have established habits and be unwilling or unable to change them  
**Figurative CN:** 养成了固定习惯且不愿或无法改变

**Original examples:**
- [34:22] And people get **set in their ways** because they've seen 12 generations of AI models that don't help them at all.  
  人们变得墨守成规，因为他们已经见过12代对他们毫无帮助的AI模型。

**Extra example:**
- After 30 years in the same job, he became **set in his ways** and resisted new methods.  
  在同一份工作干了30年后，他变得墨守成规，抵制新方法。

### to give someone credit
**Type:** collocation

**EN:** to acknowledge someone's merit or achievement  
**CN:** 承认某人的功劳或成就

**Original examples:**
- [38:16] Now, it's going to take longer for this same change to happen in the life sciences because **to give us some credit** for a moment, it's a much harder problem.  
  在生命科学领域，同样的变化需要更长时间才能发生，因为说句公道话，这是一个更难的问题。

**Extra example:**
- **To give him credit**, he never gave up despite facing numerous setbacks.  
  说句公道话，尽管遇到无数挫折，他从未放弃。

### as an end in itself
**Type:** idiom

**EN:** as something valuable for its own sake, not as a means to achieve something else  
**CN:** 本身就是目的，本身就有价值

**Literal:** 作为本身的终点  
**Figurative EN:** something pursued for its intrinsic value rather than for what it can help achieve  
**Figurative CN:** 为了其内在价值而追求的事物，而非作为达到其他目的的手段

**Original examples:**
- [38:56] The first is accelerating scientific discovery **as an end in itself**, pure pursuit of basic research.  
  第一个目标是加速科学发现，这本身就是目的，是对基础研究的纯粹追求。

**Extra example:**
- Art should be appreciated **as an end in itself**, not just for its commercial value.  
  艺术应该因其本身的价值而被欣赏，而不仅仅是因为其商业价值。

### on par with
**Type:** collocation

**EN:** equal to, at the same level as  
**CN:** 与...相当，与...同等水平

**Original examples:**
- [40:28] And the models over this time frame have gone from being not all that useful to performing at a level that is **on par** or greater than the average PhD level scientist in these fields.  
  在这段时间内，这些模型已经从不太有用发展到能够表现出与这些领域的博士级科学家相当甚至更高的水平。

**Extra example:**
- Her performance this quarter was **on par with** the top salespeople in the company.  
  她本季度的业绩与公司顶级销售人员相当。

### spin up
**Type:** phrasal_verb

**EN:** to start or create something quickly, especially computing resources  
**CN:** 快速启动或创建（尤指计算资源）

**Original examples:**
- [43:27] If you don't have the compute, you prefer that Claude handles it, it will **spin up** its own compute and GPUs and get the jobs done.  
  如果你没有算力，又希望Claude来处理，它会启动自己的计算资源和GPU来完成任务。
- [48:10] So it **spun up** three sub-agents.  
  于是它启动了三个子代理。

**Extra example:**
- We can **spin up** a new server instance in just a few minutes.  
  我们可以在几分钟内启动一个新的服务器实例。

### drill into
**Type:** phrasal_verb

**EN:** to examine or investigate something in detail  
**CN:** 深入研究，仔细查看

**Original examples:**
- [48:18] I'll **drill into** them in a second.  
  我马上会深入研究它们。
- [48:34] So **drilling into** the structure and pocket sub-agent, you can see here this is the brief that the parent agent gave.  
  深入研究结构和口袋子代理，你可以看到这是父代理给出的简报。

**Extra example:**
- Let's **drill into** the financial data to understand what went wrong last quarter.  
  让我们深入研究财务数据，弄清楚上季度出了什么问题。

### out of the box
**Type:** idiom

**EN:** immediately available without additional configuration or customization  
**CN:** 开箱即用，无需额外配置

**Literal:** 从盒子里拿出来  
**Figurative EN:** ready to use immediately without needing setup or modification  
**Figurative CN:** 无需设置或修改就能立即使用

**Original examples:**
- [50:34] Cloud Science comes ready for your domain **out of the box**.  
  Claude Science在你的领域中是开箱即用的。

**Extra example:**
- This software works perfectly **out of the box** - no installation needed.  
  这款软件开箱即用——无需安装。

### from scratch
**Type:** idiom

**EN:** starting from the very beginning with nothing prepared  
**CN:** 从零开始，从头开始

**Literal:** 从划痕开始  
**Figurative EN:** to start something from the very beginning without using anything that already exists  
**Figurative CN:** 从最初的起点开始某事，不使用任何现成的东西

**Original examples:**
- [51:03] You can add a skill simply by chatting with Claude. You can write one **from scratch**.  
  你可以通过与Claude聊天来添加技能。你可以从零开始编写一个。

**Extra example:**
- After the fire, they had to rebuild the entire business **from scratch**.  
  火灾之后，他们不得不从零开始重建整个业务。

### by construction
**Type:** collocation

**EN:** by design, as an inherent feature of how something is built  
**CN:** 在设计上，在构造上（本身就具有的特性）

**Original examples:**
- [52:08] What this means is that you can come back to the product six months, a year, two years later, every artifact is reproducible **by construction**.  
  这意味着你可以在六个月、一年、两年后回到这个产品，每个产物在设计上都是可重现的。

**Extra example:**
- The system is secure **by construction** - no additional security layers are needed.  
  这个系统在设计上就是安全的——不需要额外的安全层。

### send off
**Type:** phrasal_verb

**EN:** to submit or dispatch something  
**CN:** 发送，提交

**Original examples:**
- [52:44] So you can **send this off**.  
  所以你可以把这个发送出去。

**Extra example:**
- I'll **send off** the application by Friday.  
  我会在周五前提交申请。

### on the record
**Type:** collocation

**EN:** officially documented or permanently recorded  
**CN:** 记录在案的，正式记录的

**Original examples:**
- [53:36] All four versions are permanently **on the record**.  
  所有四个版本都被永久记录在案。
- [54:00] The agent corrected it, both versions **on the record**.  
  代理修正了它，两个版本都记录在案。

**Extra example:**
- Everything you say here is **on the record** and may be used in court.  
  你在这里说的一切都会被记录在案，可能会在法庭上使用。

### drop-in replacement
**Type:** collocation

**EN:** a substitute that can be used without any modification to existing systems  
**CN:** 直接替换品，无需修改即可替换的产品

**Original examples:**
- [55:05] We wanted to be as much a **drop-in replacement** for Jupyter notebooks as possible.  
  我们希望尽可能成为 Jupyter notebooks 的直接替换品。

**Extra example:**
- This new battery is a **drop-in replacement** for the old one - no tools needed.  
  这款新电池是旧电池的直接替换品——不需要任何工具。

### zoom out
**Type:** phrasal_verb

**EN:** to show a wider view or take a broader perspective  
**CN:** 缩小视图，拓宽视野

**Original examples:**
- [56:51] Let me **zoom out** a little bit here.  
  让我在这里稍微缩小一下视图。

**Extra example:**
- Let's **zoom out** and look at the bigger picture before making this decision.  
  在做决定之前，让我们拓宽视野，看看更大的格局。

### go no-go
**Type:** collocation

**EN:** a decision point to proceed or stop a project  
**CN:** 继续或停止的决策点

**Original examples:**
- [57:46] And most importantly, it led to the **go no-go** memo.  
  最重要的是，它产生了继续/停止决策备忘录。
- [57:52] Claude gave us a conditional go. It told us the first decisive experiment we would need to do to fund a larger experimental campaign and then it gave us the kill criteria for that experiment.  
  Claude 给出了有条件的继续建议。它告诉我们需要做的第一个决定性实验，以资助更大规模的实验活动，然后给出了该实验的终止标准。
- [58:09] One sentence in and a full campaign and a **go no-go** verdict out.  
  输入一句话，输出完整的活动方案和继续/停止的决策。

**Extra example:**
- We have a **go no-go** meeting scheduled for next week to decide if we continue with Phase 2.  
  我们安排了下周的继续/停止会议，以决定是否继续进行第二阶段。

### from the bench to the boardroom
**Type:** idiom

**EN:** from laboratory research to executive decision-making  
**CN:** 从实验室到董事会，从研发到决策层

**Literal:** 从工作台到董事会会议室  
**Figurative EN:** spanning the entire process from basic research to business/strategic decisions  
**Figurative CN:** 涵盖从基础研究到商业/战略决策的整个过程

**Original examples:**
- [58:22] We went all the way **from the bench to the boardroom** in a single session.  
  我们在一次会话中完成了从实验室到董事会的全过程。

**Extra example:**
- This scientist has experience **from the bench to the boardroom**, making her ideal for the CSO position.  
  这位科学家拥有从实验室到董事会的经验，使她成为首席科学官职位的理想人选。

### fire up
**Type:** phrasal_verb

**EN:** to start or launch (a program, application, or system)  
**CN:** 启动（程序、应用或系统）

**Original examples:**
- [01:03:48] So to do this they **fire up** Cloud Code and they give the prompt in that they need to perform this migration.  
  所以为了做到这一点，他们启动 Cloud Code 并输入需要执行此迁移的提示。

**Extra example:**
- Let me **fire up** the development server and show you the new features.  
  让我启动开发服务器，向你展示新功能。

### in the trenches
**Type:** idiom

**EN:** doing the difficult, practical work in challenging conditions  
**CN:** 在一线奋战，做艰苦的实际工作

**Literal:** 在战壕里  
**Figurative EN:** actively engaged in difficult, hands-on work alongside others  
**Figurative CN:** 与他人一起积极参与困难的实际工作

**Original examples:**
- [01:09:18] And there's no substitute for having our own experiences alongside you all **in the trenches** trying to develop drugs.  
  与你们所有人一起在一线努力开发药物的亲身经历是无可替代的。

**Extra example:**
- Our CEO started **in the trenches** as a junior engineer, so she understands the team's challenges.  
  我们的 CEO 从初级工程师在一线做起，所以她理解团队的挑战。

### jump off
**Type:** phrasal_verb

**EN:** to start or begin something, to use as a starting point  
**CN:** 开始，以...作为起点

**Original examples:**
- [01:11:27] And so it's this topic of how that we're going to **jump off** and focus on the panel.  
  所以我们将从这个话题开始，聚焦在小组讨论上。

**Extra example:**
- Let's **jump off** with the main findings from the research.  
  让我们从研究的主要发现开始。

### jump right in
**Type:** phrasal_verb

**EN:** to start doing something immediately without hesitation  
**CN:** 立即开始，马上投入

**Original examples:**
- [01:12:30] We're going to **jump right in**.  
  我们马上就开始。

**Extra example:**
- No need for introductions—let's **jump right in** and start coding.  
  不需要介绍了——我们马上开始编程吧。

### move the needle
**Type:** idiom

**EN:** to make a noticeable or significant impact or change  
**CN:** 产生显著影响，有实质性改变

**Literal:** 移动指针  
**Figurative EN:** to create measurable progress or impact, to make a real difference  
**Figurative CN:** 产生可衡量的进展或影响，真正起到作用

**Original examples:**
- [01:16:20] So it will allow you to iterate, repeat, etc., etc. In some way, it's exactly what biologists and scientists have done all along, except it operates at the scale of biology that before was beyond us, and it actually does actually **move the needle**.  
  所以它能让你迭代、重复等等。在某种程度上，这正是生物学家和科学家一直在做的事情，只不过它在以前超出我们能力的生物学规模上运作，而且它确实能产生实质性的影响。
- [01:25:33] And also as models can let you work more end to end rather than look at each step separately, then they also can be more foreshadowing challenges. Think about safety way earlier than you would normally think because you have so much that's already happening on the virtual side and on the targets. We look actually as an industry at a very small sliver of them because of this difficulty to manage the...  
  而且，随着模型可以让你进行端到端工作，而不是分别看待每个步骤，它们也能更好地预见挑战。因为在虚拟端和靶点上已经有这么多事情在发生，所以你可以比通常更早地考虑安全性。作为一个行业，我们实际上只关注了其中很小的一部分，因为难以管理这种...

**Extra example:**
- This new marketing strategy should really **move the needle** on customer acquisition.  
  这个新的营销策略应该能在客户获取方面产生显著效果。

### long tail
**Type:** collocation

**EN:** the many less common or edge cases that exist beyond the main, common scenarios  
**CN:** 长尾（指大量不常见的边缘情况）

**Original examples:**
- [01:16:45] Biology and chemistry and so on, they have a huge **long tail**.  
  生物学和化学等领域有一个巨大的长尾。

**Extra example:**
- E-commerce platforms profit from the **long tail** of niche products that few retailers stock.  
  电子商务平台从很少零售商库存的小众产品长尾中获利。

### early innings
**Type:** idiom

**EN:** the beginning or early stages of a process or development  
**CN:** 初期阶段，早期

**Literal:** 棒球比赛的早期局数  
**Figurative EN:** the beginning stages of a long-term process or journey, with much more to come  
**Figurative CN:** 一个长期过程或旅程的初始阶段，还有很多工作要做

**Original examples:**
- [01:18:45] And while we're still in the **early innings** of that journey, we think we're already starting to see promise across each of the areas that we're utilizing this technology.  
  虽然我们仍处于这一旅程的早期阶段，但我们认为在我们使用这项技术的每个领域都已经开始看到希望。

**Extra example:**
- We're only in the **early innings** of AI adoption in healthcare—there's so much more to come.  
  我们只是处于人工智能在医疗保健领域应用的早期阶段——还有很多事情要做。

### double click on
**Type:** phrasal_verb

**EN:** to examine something in more detail, to dive deeper into a topic  
**CN:** 深入研究，详细探讨

**Original examples:**
- [01:19:46] And third, we do believe this technology will create productivity. Tailwinds really across the organization, and we've got thousands of use cases that we can point to. If you **double click on** that, we rolled out the first ChatGPT models in early January, February of 2023, very early on.  
  第三，我们确实相信这项技术将在整个组织中创造生产力顺风，我们有成千上万个可以指出的用例。如果你深入了解一下，我们在2023年1月、2月初就推出了第一批ChatGPT模型，非常早。

**Extra example:**
- Let's **double click on** the revenue numbers from Q3 to understand what happened.  
  让我们深入分析一下第三季度的收入数字，看看发生了什么。

### get over our skis
**Type:** idiom

**EN:** to be too confident or ambitious, to get ahead of oneself  
**CN:** 过于自信，好高骛远

**Literal:** 滑雪时身体前倾越过滑雪板  
**Figurative EN:** to become overly ambitious or confident, often leading to loss of control or failure  
**Figurative CN:** 变得过于雄心勃勃或自信，往往导致失控或失败

**Original examples:**
- [01:26:24] So when you hear the sort of 'we're going to cure cancer in our lifetime,' we're going to make a lot of progress on cancer in our lifetime, but let's—we don't want to **get over our skis**.  
  所以当你听到那种'我们将在有生之年治愈癌症'的说法时，我们确实会在有生之年在癌症方面取得很大进展，但是——我们不想过于自信。

**Extra example:**
- I know you're excited about the new product, but let's not **get over our skis** before we test it with users.  
  我知道你对新产品很兴奋，但在用户测试之前，我们不要太过冒进。

### have skin in the game
**Type:** idiom

**EN:** to have personal investment or risk in an outcome, to be personally affected by results  
**CN:** 有切身利益，承担风险

**Literal:** 在游戏中有皮肤（身体）  
**Figurative EN:** to have personal stake, investment, or risk in something, making one personally affected by the outcome  
**Figurative CN:** 在某事中有个人利害关系、投资或风险，使自己受结果的直接影响

**Original examples:**
- [01:26:44] The big goal there is both to focus on neglected diseases, but also for ourselves to learn these lessons and learn alongside you and really **have skin in the game**.  
  这里的主要目标既是关注被忽视的疾病，也是为了让我们自己学习这些经验，与你们一起学习，真正承担风险、有切身利益。

**Extra example:**
- Investors trust founders who **have skin in the game** with their own money invested.  
  投资者信任那些投入自己的钱、有切身利益的创始人。

### let a thousand flowers bloom
**Type:** idiom

**EN:** to allow many different ideas or approaches to develop freely  
**CN:** 让百花齐放，允许多种想法或方法自由发展

**Literal:** 让一千朵花开放  
**Figurative EN:** to encourage experimentation and diversity of approaches without imposing constraints  
**Figurative CN:** 鼓励实验和方法的多样性，不施加限制

**Original examples:**
- [01:27:02] When we started rolling out this technology, our intention was to **let a thousand flowers bloom**.  
  当我们开始推出这项技术时，我们的意图是让百花齐放。

**Extra example:**
- The company decided to **let a thousand flowers bloom** in the innovation phase before narrowing down to the best ideas.  
  公司决定在创新阶段让百花齐放，然后再缩小到最好的想法。

### prove out
**Type:** phrasal_verb

**EN:** to demonstrate or test that something works in practice  
**CN:** 证明、验证（某事物在实践中有效）

**Original examples:**
- [01:28:26] We put small teams that had six to eight weeks to **prove out** the concept.  
  我们派出小团队，给他们六到八周的时间来验证这个概念。

**Extra example:**
- We need to **prove out** this new technology before scaling it across the company.  
  在公司范围内推广之前，我们需要先验证这项新技术。

### phone it in
**Type:** idiom

**EN:** to do something with minimal effort or without genuine engagement  
**CN:** 敷衍了事，不认真对待

**Literal:** 打电话进来（指远程完成工作）  
**Figurative EN:** to perform a task with little effort or enthusiasm, going through the motions without real commitment  
**Figurative CN:** 敷衍了事地完成任务，缺乏真正的投入和热情

**Original examples:**
- [01:34:16] My worry is that employees and society in general goes on autopilot, because when you have tools that can create content... it's very easy to **phone it in**.  
  我担心的是员工和整个社会会进入自动驾驶模式，因为当你有了可以创建内容的工具时……很容易就会敷衍了事。

**Extra example:**
- The team clearly **phoned it in** on this presentation - there's no original thinking here.  
  团队在这个演示上明显敷衍了事——这里没有任何原创思考。

### in spades
**Type:** idiom

**EN:** in large amounts; to a great degree  
**CN:** 大量地；非常充分地

**Literal:** 用黑桃（扑克牌中点数最高的花色）  
**Figurative EN:** to an extreme or abundant degree; in great quantity or intensity  
**Figurative CN:** 极大程度地；大量地；充分地

**Original examples:**
- [01:35:15] The problems that we have to deal with in our industry are the most complex problems... You've heard that **in spades** on this discussion.  
  我们行业必须处理的问题是最复杂的问题……在这次讨论中你们已经充分听到了这一点。

**Extra example:**
- She has talent **in spades** but lacks the discipline to succeed.  
  她非常有天赋，但缺乏成功所需的自律。

### break free
**Type:** phrasal_verb

**EN:** to escape from constraints or established patterns  
**CN:** 挣脱、摆脱（束缚或既定模式）

**Original examples:**
- [01:36:01] That almost infects their brain and they become even more and more ingrained in that. And the reason I worry about that is that this is something that is actually not new... They have like their way of doing things. And then some other upstart comes and disrupts the whole thing because they're able to **break free** from that.  
  这几乎会感染他们的大脑，让他们变得越来越根深蒂固。我之所以担心这一点，是因为这实际上不是什么新鲜事……他们有自己的做事方式。然后一些新兴公司出现并颠覆了整个局面，因为他们能够摆脱这种束缚。
- [01:36:45] So there is a risk that the model kind of takes over our thinking... and is unable to **break free**.  
  所以有一个风险，就是模型会接管我们的思维……而无法挣脱。

**Extra example:**
- The startup was able to **break free** from traditional industry practices and create something truly innovative.  
  这家初创公司能够摆脱传统行业做法，创造出真正创新的东西。

### bring someone along
**Type:** phrasal_verb

**EN:** to help someone progress or develop understanding alongside changes  
**CN:** 带领某人一起前进，帮助某人跟上变化

**Original examples:**
- [01:38:28] We also need to **bring the drug regulator along** with us to get the full benefits of speed and probability of success we've been talking about.  
  我们还需要带领药物监管机构一起前进，以获得我们一直在讨论的速度和成功概率的全部好处。

**Extra example:**
- As we adopt these new technologies, we need to **bring our customers along** so they understand the value.  
  在我们采用这些新技术时，我们需要带领客户一起前进，让他们理解其价值。

### build on
**Type:** phrasal_verb

**EN:** to use something as a foundation to develop further ideas or comments  
**CN:** 在……基础上发展；补充说明

**Original examples:**
- [01:41:10] From my standpoint, I just **build on** what Vos said.  
  从我的角度来看，我只是在Vos所说的基础上补充一下。

**Extra example:**
- I'd like to **build on** Sarah's point about customer engagement and add some data from our recent survey.  
  我想在Sarah关于客户参与度的观点基础上补充一些我们最近调查的数据。

### put under pressure
**Type:** collocation

**EN:** to force someone or something into a difficult or stressful situation where action is required  
**CN:** 施加压力，使处于压力之下

**Original examples:**
- [01:42:04] The thing I worry about... is at some point organizations, certainly large organizations, are going to become **put under pressure** to ration the use of this technology.  
  我担心的是……在某个时候，组织，尤其是大型组织，将被迫限制这项技术的使用。

**Extra example:**
- The company was **put under pressure** by shareholders to improve quarterly results.  
  公司被股东施压要求改善季度业绩。

### walk someone through something
**Type:** phrasal_verb

**EN:** to guide someone step-by-step through a process or explanation  
**CN:** 引导某人逐步了解某事，详细讲解

**Original examples:**
- [01:44:17] Eric and D— sorry, Lattice and Dario **walked us through** their unvarnished views on how compression may or may not happen, the challenges and opportunity ahead.  
  Eric和D——抱歉，Lattice和Dario详细讲解了他们对压缩可能发生或不发生的坦率看法，以及未来的挑战和机遇。
- [01:44:25] Eric and Alec **walked you through** the basis of our claim: models that meaningfully improve in biology with every release, and a workbench that runs the analysis your scientists actually need.  
  Eric和Alec详细讲解了我们主张的基础：每次发布都在生物学方面有显著改进的模型，以及一个能运行科学家实际需要的分析的工作台。
- [01:44:38] And Voss, Aviv, and Chris just **walked us through** how they're charting the path of AI transformation at BMS, Genentech, and Artis.  
  Voss、Aviv和Chris刚刚详细讲解了他们如何在BMS、Genentech和Artis规划AI转型的路径。

**Extra example:**
- Let me **walk you through** the installation process step by step.  
  让我逐步引导你完成安装过程。

### at the end of the day
**Type:** collocation

**EN:** ultimately; when everything is considered; what really matters  
**CN:** 最终，归根结底，说到底

**Original examples:**
- [01:44:54] **At the end of the day**, this is just the beginning.  
  归根结底，这只是个开始。

**Extra example:**
- **At the end of the day**, what matters most is customer satisfaction.  
  归根结底，最重要的是客户满意度。

### put something in someone's hands
**Type:** idiom

**EN:** to give someone direct access to or control of something  
**CN:** 让某人直接使用或掌控某物

**Literal:** 把某物放在某人手中  
**Figurative EN:** to give someone direct access to or control of something  
**Figurative CN:** 让某人直接获得或掌控某物，使某人能亲自使用

**Original examples:**
- [01:45:16] The QR code on this screen will give you access to it and **put it right in your hands**.  
  屏幕上的二维码会让你能够访问它，并直接将它交到你手中。

**Extra example:**
- This new app **puts powerful analytics right in your hands**.  
  这个新应用让你直接掌握强大的分析功能。

### give something back
**Type:** phrasal_verb

**EN:** to return something that was taken or lost  
**CN:** 归还，返还（被拿走或失去的东西）

**Original examples:**
- [01:45:36] In the end, this is about time and the question, **giving your scientists back** what they trained for and you deciding which questions they'll spend it on.  
  最终，这关乎时间和问题，把科学家们接受训练的目标还给他们，由你决定他们将把时间花在哪些问题上。

**Extra example:**
- Automation can **give back** hours of your day for more creative work.  
  自动化可以为你节省出数小时，用于更有创造性的工作。

---

## Complex Sentences

### [02:37]
**Original:** Once AI could sit inside that loop, the loop could keep turning for longer and longer stretches of time before an engineer had to step back in.

**Translation:** 一旦AI能够置身于那个循环之中,这个循环就能够持续运转越来越长的时间,然后工程师才需要重新介入。

**Core structure:**
- Once AI could sit inside the loop, the loop could keep turning before an engineer had to step back in.  
  一旦AI能置身循环中,循环就能持续运转,然后工程师才需介入。

**Structure tree:**
```
时间状语从句: Once AI could sit inside that loop
主句: the loop could keep turning
时间范围: for longer and longer stretches of time
时间状语从句: before an engineer had to step back in
```

**Grammar points:**
- **Once引导的时间状语从句** - 表示"一旦...就..."的条件时间关系
- **before引导的时间状语从句** - 表示主句动作发生在从句动作之前
- **比较级叠用** - longer and longer表示程度递增

### [03:12]
**Original:** Every R&D-er in this room knows the shape of it: the experiment that takes three days to run but three or more weeks to analyze.

**Translation:** 在座的每一位研发人员都知道它的模式:实验运行需要三天,但分析却需要三周或更长时间。

**Core structure:**
- Every R&D-er knows the shape: the experiment that takes three days but three weeks to analyze.  
  每位研发人员都知道这个模式:实验需要三天但分析需要三周。

**Structure tree:**
```
主句: Every R&D-er knows the shape of it
同位语: the experiment that...
定语从句: that takes three days to run but three or more weeks to analyze
并列结构: takes X to run but Y to analyze
```

**Grammar points:**
- **同位语结构** - 冒号后解释说明前面的'the shape of it'
- **定语从句中的并列对比** - but连接两个时间对比,强调反差
- **不定式作状语** - to run和to analyze表示目的

### [03:28]
**Original:** It's our belief that that toil is collapsing, and as it does, your scientists will get back more of what they trained for: time at the question.

**Translation:** 我们相信,这种劳苦正在消解,而随着它的消解,你们的科学家将会重新获得更多他们所接受训练的目标:专注于问题本身的时间。

**Core structure:**
- We believe that the toil is collapsing, and your scientists will get back time at the question.  
  我们相信劳苦正在消解,科学家将重新获得专注问题的时间。

**Structure tree:**
```
主句: It's our belief that...
宾语从句: that toil is collapsing
并列句: and as it does, scientists will get back...
宾语: more of what they trained for
同位语: time at the question
```

**Grammar points:**
- **强调句型** - It's our belief that...强调观点
- **as引导的时间状语从句** - as it does中does代替前面的collapses,避免重复
- **what引导的宾语从句** - what they trained for作get back的宾语

### [04:11]
**Original:** Two years ago, our CEO wrote that AI-enabled biology and medicine could compress 50 to 100 years of progress into 5 to 10.

**Translation:** 两年前,我们的CEO写道,AI赋能的生物学和医学可以将50到100年的进展压缩到5到10年。

**Core structure:**
- Our CEO wrote that AI-enabled biology could compress 50 to 100 years into 5 to 10.  
  我们CEO写道,AI赋能的生物学可以将50到100年压缩到5到10年。

**Structure tree:**
```
主句: our CEO wrote that...
宾语从句: AI-enabled biology and medicine could compress...
动词结构: compress X into Y
数字范围: 50 to 100 years / 5 to 10 (years)
```

**Grammar points:**
- **that引导的宾语从句** - 从句作wrote的宾语
- **复合形容词** - AI-enabled作定语修饰biology and medicine
- **compress...into结构** - 表示'将...压缩到...'的动作

### [04:38]
**Original:** Please welcome the scientist behind the GLP-1 medicines, Lasker Award winner and former chief scientific adviser at Novo Nordisk, Lotte Bjerre Knudsen, in conversation with Dario Amodei, our co-founder and CEO.

**Translation:** 请欢迎GLP-1药物背后的科学家、拉斯克奖得主、诺和诺德前首席科学顾问洛特·比耶尔·克努森,她将与我们的联合创始人兼CEO达里奥·阿莫代进行对话。

**Core structure:**
- Please welcome the scientist Lotte Bjerre Knudsen in conversation with Dario Amodei.  
  请欢迎科学家洛特·比耶尔·克努森与达里奥·阿莫代对话。

**Structure tree:**
```
祈使句: Please welcome...
核心名词: the scientist
同位语1: Lasker Award winner and former chief scientific adviser
人名: Lotte Bjerre Knudsen
介词短语: in conversation with Dario Amodei
同位语2: our co-founder and CEO
```

**Grammar points:**
- **多重同位语** - 多个名词短语修饰说明the scientist的身份
- **介词短语作状语** - in conversation with表示对话关系

### [09:30]
**Original:** And the second is, I think, just the inertia in the system—both the inertia of getting used to using these tools and operating in the new way, and figuring out, you know, how do they help with academic biology research, how do they help with new target discovery, how do they help with running clinical trials faster.

**Translation:** 第二点是，我认为,就是系统中的惰性——既包括习惯使用这些工具和以新方式运作的惰性，也包括弄清楚，你知道，它们如何帮助学术生物学研究、如何帮助新靶点发现、如何帮助更快地进行临床试验。

**Core structure:**
- The second is the inertia in the system.  
  第二点是系统中的惰性。

**Structure tree:**
```
main clause: The second is the inertia
  |-- parenthetical: I think
  |-- appositive expansion: both the inertia of...
      |-- parallel gerund phrases: getting used to... and operating... and figuring out...
      |-- embedded questions: how do they help with X, Y, Z
```

**Grammar points:**
- **同位语扩展** - 破折号后用both...and结构展开解释inertia的具体内容
- **并列动名词短语** - getting used to, operating, and figuring out三个动名词并列作of的宾语
- **间接疑问句** - how do they help...作figuring out的宾语从句，保留疑问词但用陈述语序

### [10:38]
**Original:** Like, there are a small number of these discoveries, and some of them feel like they could have happened decades earlier than they did, right?

**Translation:** 比如说，这类发现的数量很少，而其中一些感觉本可以比实际发生的时间早几十年就出现了，对吧？

**Core structure:**
- There are discoveries, and some could have happened earlier.  
  有一些发现，其中一些本可以更早发生。

**Structure tree:**
```
compound sentence:
  clause 1: there are discoveries
  clause 2: some of them feel like...
    |-- that-clause (省略that): they could have happened...
    |-- comparison: earlier than they did
```

**Grammar points:**
- **虚拟语气（过去推测）** - could have happened表示对过去的推测，意为"本可以发生但没有"
- **比较结构的省略** - than they did中did代替happened，避免重复

### [11:24]
**Original:** And then it's all going to be about taking those new discoveries, which have such broad implications across everything, and kind of using them and translating them, both in terms of the science operationally and in terms of the regulatory system to make everything truly go faster and it's going to be, you know, it's going to be the work of years but I think it can be done.

**Translation:** 然后这一切都将围绕着利用那些新发现——它们对各个领域都有如此广泛的影响——以及某种程度上使用和转化它们，无论是在科学运作层面还是监管体系层面，使一切真正加快，这将是多年的工作，但我认为这是可以做到的。

**Core structure:**
- It's going to be about taking discoveries and using them to make everything go faster.  
  这将围绕着利用发现并使用它们来加快一切。

**Structure tree:**
```
main clause: it's going to be about...
  |-- gerund objects: taking...and using...and translating...
      |-- non-restrictive clause: which have implications
      |-- prepositional phrases: in terms of X and Y
      |-- purpose: to make everything go faster
  compound: and it's going to be the work of years but I think it can be done
```

**Grammar points:**
- **be about + 动名词** - 表示"关于/涉及做某事"，动名词taking, using, translating并列
- **非限制性定语从句** - which have...插入修饰discoveries，补充说明其影响范围
- **使役动词结构** - make everything go faster，make后接宾语+不带to的不定式作宾补

### [13:00]
**Original:** So it must be possible to do that faster.

**Translation:** 所以更快地做到这一点一定是可能的。

**Core structure:**
- It must be possible to do that faster.  
  更快地做到这一点一定是可能的。

**Structure tree:**
```
main clause: it must be possible
  |-- real subject: to do that faster
  |-- it: formal subject
```

**Grammar points:**
- **must表推测** - must be possible表示"一定可能"，对现在情况的肯定推测
- **it作形式主语** - 真正主语是to do that faster，it只是占位

### [14:17]
**Original:** How does having an industry with this cycle time, where very often you find out your drug didn't work 10 years later when you were convinced the whole time it would, right—I've watched people go through this process for 25 years, it's an immensely difficult thing—how do you deal with that?

**Translation:** 拥有一个具有这种周期时间的行业——在这个行业中，你经常会在10年后才发现你的药物不起作用，而这整个过程中你一直确信它会有效，对吧——我看着人们经历这个过程25年了，这是一件极其困难的事情——你如何应对这个问题？

**Core structure:**
- How does having an industry affect what you do? How do you deal with that?  
  拥有这样一个行业如何影响你的工作？你如何应对？

**Structure tree:**
```
question: How does having an industry...how do you deal with that?
  |-- gerund subject: having an industry with this cycle time
      |-- relative clause: where you find out...
          |-- temporal clause: when you were convinced...
              |-- object clause: (that) it would (work)
  |-- interruption: I've watched...it's an immensely difficult thing
```

**Grammar points:**
- **动名词短语作主语** - having an industry with this cycle time整个动名词短语作主语
- **where引导定语从句** - where修饰industry，相当于in which，表示抽象地点
- **嵌套时间状语从句** - when从句嵌套在where从句内，when you were convinced又包含宾语从句it would (work)

### [16:27]
**Original:** If we have to do less clinical trials because things work more of the time, if when things work they have a stronger effect, then we need to recruit less patients.

**Translation:** 如果我们需要做更少的临床试验，因为药物更常有效，如果当药物有效时它们有更强的效果，那么我们需要招募更少的患者。

**Core structure:**
- If we do less trials, if things have stronger effect, then we need less patients.  
  如果我们做更少试验，如果药物效果更强，那么我们需要更少患者。

**Structure tree:**
```
main clause: then we need to recruit less patients
condition 1: If we have to do less clinical trials
reason: because things work more of the time
condition 2: if when things work they have a stronger effect
nested time clause: when things work
```

**Grammar points:**
- **多重条件句嵌套** - 两个 if 条件句 + 一个嵌套的 when 时间状语从句，结构层次复杂
- **原因状语从句** - because 从句解释第一个条件

### [17:36]
**Original:** Yeah, so I mean I remember when Andy Grove was talking about these things, there was a writer, I forget who it was, who coined like the Andy Grove fallacy of like if you try and think... Derek Lowe.

**Translation:** 是的，我记得当安迪·格鲁夫谈论这些事情时，有一位作家，我忘了是谁，他创造了像安迪·格鲁夫谬论这样的说法，就像如果你试图思考... 德里克·洛。

**Core structure:**
- I remember there was a writer who coined the Andy Grove fallacy.  
  我记得有一位作家创造了安迪·格鲁夫谬论这个说法。

**Structure tree:**
```
main clause: I remember when...
time clause: when Andy Grove was talking
main structure: there was a writer
defining clause: who coined the Andy Grove fallacy
interruption: I forget who it was
incomplete clause: if you try and think...
```

**Grammar points:**
- **插入语** - I forget who it was 打断主句流畅性
- **口语化不完整结构** - 句子未完成就转向，like 作填充词，体现口语特征
- **定语从句** - who coined... 修饰 a writer

### [18:10]
**Original:** But the way we as humans have only made progress is we've used our human brains, which are capable of comprehending complexity, to kind of wrestle with that, to make sense of it, to make progress against it.

**Translation:** 但我们人类取得进步的唯一方式是，我们使用了我们人类的大脑——它能够理解复杂性——来与之搏斗，理解它，对抗它取得进步。

**Core structure:**
- The way we have made progress is we've used our brains to wrestle with complexity.  
  我们取得进步的方式是我们用大脑与复杂性搏斗。

**Structure tree:**
```
subject: The way we have made progress
predicative clause: we've used our brains...
defining clause: which are capable of comprehending complexity
parallel infinitives: to wrestle / to make sense / to make progress
```

**Grammar points:**
- **The way 引导的名词性从句** - The way 作主语，后接表语从句
- **非限制性定语从句** - which 从句补充说明 human brains
- **三个并列不定式** - to wrestle, to make sense, to make progress 表目的

### [19:57]
**Original:** And a very important learning from GLP-1 is that actually a pleiotropic effect is a good one, right? What's really so fantastic about GLP-1 is that you have all these benefits on multiple different organs, yet the whole field of drug discovery is still looking for the genetics with the highest window or the mouse model with the highest window.

**Translation:** 从GLP-1得到的一个非常重要的教训是，实际上多效性效应是一件好事，对吧？GLP-1真正如此神奇的地方在于，你在多个不同器官上都有这些益处，然而整个药物发现领域仍在寻找具有最高窗口期的遗传学或具有最高窗口期的小鼠模型。

**Core structure:**
- What's fantastic is that you have benefits on multiple organs, yet the field is still looking for the highest window.  
  神奇之处在于你在多个器官上有益处，然而该领域仍在寻找最高窗口期。

**Structure tree:**
```
What-clause subject: What's really so fantastic about GLP-1
predicative clause: that you have all these benefits...
contrast: yet the whole field is still looking for...
parallel objects: the genetics / the mouse model
modifier: with the highest window
```

**Grammar points:**
- **What 主语从句** - What's fantastic 作主语，强调重点
- **yet 表转折** - 连接对比关系，尽管有多效性益处，研究仍追求单一高窗口期
- **并列结构** - the genetics... or the mouse model 两个并列宾语

### [21:03]
**Original:** We've kind of entered an inflection point or a critical window, you know, as we go along the exponential. Different things in terms of both benefits and economically useful applications and, you know, potentially concerning applications turn on at different times.

**Translation:** 我们已经进入了一个拐点或关键窗口期，你知道，随着我们沿着指数曲线前进。就利益和经济上有用的应用以及，你知道，潜在令人担忧的应用而言，不同的东西在不同时间开启。

**Core structure:**
- Different things turn on at different times as we go along the exponential.  
  随着我们沿着指数曲线前进，不同的东西在不同时间开启。

**Structure tree:**
```
main clause: Different things turn on at different times
complex modifier: in terms of benefits and applications and applications
time clause: as we go along the exponential
filler phrases: you know (x2), kind of
```

**Grammar points:**
- **in terms of 复杂修饰语** - 三个并列名词短语修饰 different things，结构冗长
- **口语化填充词** - you know, kind of 等填充词增加理解难度

### [22:07]
**Original:** So we've spent a lot of time putting effort into, you know, what is a helpful query, what is a dangerous query.

**Translation:** 所以我们花了很多时间努力研究，你知道，什么是有帮助的查询，什么是危险的查询。

**Core structure:**
- We've spent time putting effort into what is helpful and what is dangerous.  
  我们花时间努力研究什么是有帮助的，什么是危险的。

**Structure tree:**
```
main clause: We've spent time putting effort into...
into + object: what is a helpful query, what is a dangerous query
two parallel indirect questions as objects
```

**Grammar points:**
- **spend time doing** - 固定搭配，表示花时间做某事
- **间接疑问句作宾语** - what is... 作介词 into 的宾语，注意陈述语序

### [22:35]
**Original:** And you know, I think this is another thing, not just in cyber, not just in biology, that's going to be the work of years, where you have a model that depending on what you do and what you say to it and how you interact with it, can do lots of wondrous things, can create enormous economic value, but then there's a tiny slice of things that are actually dangerous, and within that a tinier slice of things that are dangerous and you couldn't do without AI, or where AI actually is the limiting factor, right?

**Translation:** 你知道，我认为这是另一件事，不仅在网络安全领域，不仅在生物学领域，这将是需要多年努力的工作，在这种情况下，你有一个模型，取决于你做什么、对它说什么以及如何与它互动，它可以做很多奇妙的事情，可以创造巨大的经济价值，但同时也有一小部分实际上是危险的事情，而在这其中又有更小的一部分是危险的且没有AI你做不到的事情，或者说AI实际上是限制因素的事情，对吧？

**Core structure:**
- This is something that's going to be the work of years, where you have a model that can do wondrous things but there's a tiny slice that are dangerous.  
  这是需要多年努力的事情，你有一个既能做奇妙事情又有一小部分危险的模型。

**Structure tree:**
```
main: this is another thing that's going to be the work of years
where clause: where you have a model that...
relative clause: that depending on... can do..., can create..., but...
parallel contrasts: can do X, can create Y, but there's Z
nested within: things that are dangerous and you couldn't do without AI
```

**Grammar points:**
- **定语从句嵌套** - where 引导定语从句修饰 work of years，其中又包含 that 引导的定语从句修饰 model
- **分词短语作状语** - depending on... 作条件状语，修饰整个从句
- **并列转折结构** - can do..., can create... but then... 多重并列后的转折

### [23:46]
**Original:** So this idea of kind of trusted access, existing elements of society that manage these risks, safeguards on the general models, it's kind of almost a multi-layer cake of how to make sure we get as close as we can to 100% of the benefits while blocking as close as we can to 100% of the real counterfactual actual dangerous harms, which is a very narrow slice, but that we need to make sure that we put a proper buffer around to manage appropriately.

**Translation:** 所以这种可信访问的想法，管理这些风险的现有社会元素，通用模型上的保障措施，这几乎像是一个多层蛋糕，关于如何确保我们尽可能接近获得100%的收益，同时尽可能接近阻止100%真实的反事实的实际危险伤害，这是一个非常狭窄的领域，但我们需要确保在其周围设置适当的缓冲区来妥善管理。

**Core structure:**
- This idea is a multi-layer cake of how to get benefits while blocking harms.  
  这个想法是关于如何获得收益同时阻止伤害的多层蛋糕。

**Structure tree:**
```
subject: this idea of trusted access, existing elements, safeguards (三个并列名词短语)
predicate: is a multi-layer cake of how to...
how clause: how to make sure we get... while blocking...
as...as structure: as close as we can to 100%
relative clause: which is a very narrow slice, but that we need...
```

**Grammar points:**
- **多重并列主语** - 三个名词短语并列作主语，结构复杂
- **as...as 比较结构** - as close as we can 表示尽可能接近
- **while 表对比** - while blocking 表示同时进行的对比动作

### [24:16]
**Original:** I mean, you've dealt with the problem of developing a drug in tens of thousands of people and giving it to millions and with the worries of side effects that come, and you've both seen cases outside of your work where there were drugs that had real side effects that required they be withdrawn, and also some of the worries about GLP-1s which have mostly not panned out.

**Translation:** 我的意思是，你处理过在数万人身上开发药物并将其给予数百万人的问题，以及随之而来的副作用担忧，而且你们都见过工作之外的案例，那里有些药物有真实的副作用，要求它们被撤回，还有一些关于GLP-1的担忧，这些担忧大多没有成真。

**Core structure:**
- You've dealt with the problem of developing drugs and you've seen cases where drugs had side effects.  
  你处理过开发药物的问题，也见过药物有副作用的案例。

**Structure tree:**
```
parallel structure: you've dealt with... and you've seen...
first part: the problem of developing... and giving... and with the worries
second part: cases where there were drugs that... and also worries about GLP-1s which...
nested: drugs that had effects that required they be withdrawn
```

**Grammar points:**
- **虚拟语气 require** - required they be withdrawn，require 后用虚拟语气 (should) be
- **多层定语从句嵌套** - where... that... that... 三层从句嵌套修饰

### [27:25]
**Original:** I don't think we will ever have an AI model that never hallucinates. I think just the probabilistic way in which these models reason, which I suspect is the same as the probabilistic way in which humans—The reason is it is prone to a duality between creativity and, you know, basically hallucination, right?

**Translation:** 我认为我们永远不会有一个从不产生幻觉的AI模型。我认为这些模型推理的概率方式，我怀疑这与人类推理的概率方式相同——原因是它容易产生创造力和幻觉之间的二元性，对吧？

**Core structure:**
- The probabilistic way in which models reason is prone to a duality between creativity and hallucination.  
  模型推理的概率方式容易在创造力和幻觉之间产生二元性。

**Structure tree:**
```
main: the way is prone to a duality
relative: in which these models reason
inserted: which I suspect is the same as...
interrupted structure: humans—The reason is...
between: creativity and hallucination
```

**Grammar points:**
- **介词+关系代词** - in which 引导定语从句，相当于 where
- **be prone to** - 固定搭配，表示易于、倾向于
- **句子中断与重启** - 破折号表示说话中断，然后重新表述，口语特征

### [30:42]
**Original:** Yeah, well, I guess you've got to have a really strong focus on what it actually is that you're trying to do, and then you have to be informed about what the different models are and the ways of using them.

**Translation:** 是的,我想你必须非常明确地专注于你实际上想要做的事情,然后你必须了解不同的模型是什么以及使用它们的方法。

**Core structure:**
- You've got to have a focus on what you're trying to do, and you have to be informed about what the models are.  
  你必须专注于你想做的事情,并了解模型是什么。

**Structure tree:**
```
main: you've got to have focus... and you have to be informed...
├─ object clause 1: what it is that you're trying to do
│  └─ emphasis structure: it is...that
└─ object clause 2: what the different models are and the ways of using them
```

**Grammar points:**
- **强调句型 it is...that** - 嵌套在宾语从句中,强调'你想做的事情'
- **并列宾语从句** - what引导两个并列的宾语从句

### [32:31]
**Original:** So you know the field rising to meet the moment and scientists having the foresight to say well the AI model I had three months ago couldn't help me at all with this but the one I just got today actually is helping me a lot.

**Translation:** 所以你知道,这个领域正在奋起迎接这一时刻,科学家们有远见地说,我三个月前的AI模型根本无法帮助我解决这个问题,但我今天刚得到的这个模型实际上帮了我很多。

**Core structure:**
- The field is rising and scientists are having the foresight to say the model is helping me.  
  这个领域正在崛起,科学家们有远见地说模型在帮助我。

**Structure tree:**
```
main: the field rising... and scientists having foresight...
└─ infinitive: to say...
   └─ indirect speech: the model I had... couldn't help... but the one I got... is helping...
      ├─ relative clause 1: I had three months ago
      └─ relative clause 2: I just got today
```

**Grammar points:**
- **并列动名词短语** - the field rising和scientists having作并列宾语
- **省略关系代词的定语从句** - I had和I got省略了that/which
- **转折对比结构** - couldn't help...but...is helping形成对比

### [33:28]
**Original:** I've seen it run through when AI first beat the world Go champion to how well it's performing on code to what we've seen with Mythos and Cyber over the last few months to how AI is doing on mathematics to the quality of AI's writing to cite something that like biology is not as... Not as easily verifiable.

**Translation:** 我见证了它的历程,从AI首次击败世界围棋冠军,到它在编程上的出色表现,到我们在过去几个月看到的Mythos和Cyber的成果,到AI在数学上的表现,再到AI写作的质量,举这些例子是因为生物学不像它们那样容易验证。

**Core structure:**
- I've seen it run through from when AI beat the champion to how it's performing on code to what we've seen.  
  我见证了它从AI击败冠军到在编程上表现出色到我们所看到的成果。

**Structure tree:**
```
main: I've seen it run through...
└─ parallel prepositional phrases (from...to...to...to...):
   ├─ when AI first beat...
   ├─ how well it's performing...
   ├─ what we've seen...
   ├─ how AI is doing...
   └─ to the quality of AI's writing
      └─ purpose: to cite something that biology is not...
```

**Grammar points:**
- **多重并列结构** - 连续使用from...to...to...to连接多个时间点/领域
- **混合从句类型** - when时间状语从句、how/what名词性从句混合使用

### [34:16]
**Original:** And then the exponential does its thing, and they actually get to the point where they can help you a lot. And it happens so fast, and people get set in their ways because they've seen 12 generations of AI models that don't help them at all.

**Translation:** 然后指数增长发挥作用,它们实际上达到了可以帮助你很多的程度。这发生得如此之快,而人们却固守成规,因为他们已经见过12代对他们毫无帮助的AI模型。

**Core structure:**
- The exponential does its thing, they get to the point, it happens fast, and people get set in their ways because they've seen models that don't help them.  
  指数增长发挥作用,它们达到那个程度,这发生得很快,人们固守成规因为他们见过无用的模型。

**Structure tree:**
```
compound sentence (4 independent clauses):
├─ the exponential does its thing
├─ they get to the point where they can help...
├─ it happens so fast
└─ people get set... because they've seen models
   └─ relative clause: that don't help them at all
```

**Grammar points:**
- **多重并列句** - 四个独立分句用and连接,表达连贯的因果逻辑
- **get set in their ways** - 习语,表示'固守习惯/墨守成规'

### [35:42]
**Original:** So I want more bilingual people in all teams everywhere, and of course I don't mean people who speak two languages. I mean people who are completely fluent in some scientific topic as well as in digital and AI.

**Translation:** 所以我希望所有团队中都有更多'双语'人才,当然我说的不是会说两种语言的人。我指的是那些既精通某个科学主题,又精通数字技术和AI的人。

**Core structure:**
- I want bilingual people in teams, and I don't mean people who speak languages. I mean people who are fluent in scientific topics as well as in AI.  
  我想要团队中有双语人才,我不是指会说语言的人。我指精通科学主题和AI的人。

**Structure tree:**
```
compound sentence:
├─ I want bilingual people...
├─ I don't mean people who speak...
│  └─ relative clause: who speak two languages
└─ I mean people who are fluent...
   └─ relative clause: who are completely fluent in...
      └─ parallel objects: in some topic as well as in digital and AI
```

**Grammar points:**
- **clarification structure** - 用I don't mean...I mean...结构澄清定义
- **as well as并列** - 连接两个介词短语,表示'既...又...'

### [38:26]
**Original:** In the life sciences, our feedback loops take longer and they involve running real experiments in the physical world and there's so much uncertainty and noise in all biological data.

**Translation:** 在生命科学领域,我们的反馈循环需要更长时间,它们涉及在物理世界中进行真实实验,而且所有生物数据中都存在大量的不确定性和噪声。

**Core structure:**
- Feedback loops take longer and involve running experiments and there's uncertainty in data.  
  反馈循环需要更长时间,涉及进行实验,数据中存在不确定性。

**Structure tree:**
```
parallel compound sentence:
- clause 1: feedback loops take longer
- clause 2: they involve running experiments (with gerund object)
- clause 3: there's uncertainty and noise in data
modifier: in the life sciences (prepositional phrase)
```

**Grammar points:**
- **并列复合句** - 三个独立分句用and连接,表达递进关系
- **动名词作宾语** - involve后接running experiments
- **there be结构** - 表示存在,主语是uncertainty and noise

### [39:27]
**Original:** It includes the product layer of optimizing the product for scientists, because even with the most capable models in the world, we still need to have the right product features to make the model intelligence integrated into workflows and accessible to scientists.

**Translation:** 它包括为科学家优化产品的产品层,因为即使拥有世界上最强大的模型,我们仍然需要正确的产品功能,以使模型智能融入工作流程并让科学家能够使用。

**Core structure:**
- It includes the product layer because we need features to make intelligence integrated and accessible.  
  它包括产品层,因为我们需要功能来使智能融入并可访问。

**Structure tree:**
```
main clause: It includes the product layer
reason clause: because we need features...
concessive phrase: even with the most capable models
purpose infinitive: to make intelligence integrated and accessible
compound complement: integrated into workflows AND accessible to scientists
```

**Grammar points:**
- **because引导原因状语从句** - 解释为什么需要产品层
- **make + 宾语 + 过去分词** - 使役结构,表示'使...被融入/变得可访问'
- **不定式表目的** - to make...说明需要功能的目的

### [40:28]
**Original:** And the models over this time frame have gone from being not all that useful to performing at a level that is on par or greater than the average PhD level scientist in these fields.

**Translation:** 在这段时间内,这些模型已经从不太有用发展到在这些领域的表现水平与普通博士级科学家相当甚至更高。

**Core structure:**
- The models have gone from being not useful to performing at a high level.  
  模型已经从不太有用发展到高水平表现。

**Structure tree:**
```
main clause: models have gone from A to B
point A: being not all that useful (gerund phrase)
point B: performing at a level (gerund phrase)
level description: that is on par or greater than PhD scientist
modifiers: over this time frame, in these fields
```

**Grammar points:**
- **from...to...结构** - 表示从一个状态到另一个状态的变化,to后接动名词
- **定语从句修饰level** - that is on par or greater than...描述水平高低

### [41:25]
**Original:** In science, there are all these problems that we have that really have little to do with the model capabilities and everything to do with things like connecting to tens of different databases and perfecting every last iteration of the figures that go into manuscripts.

**Translation:** 在科学领域,我们面临的所有这些问题实际上与模型能力关系不大,而完全关系到诸如连接数十个不同数据库和完善论文中图表的每一次迭代这样的事情。

**Core structure:**
- There are problems that have little to do with capabilities and everything to do with connecting databases and perfecting figures.  
  存在一些问题,与能力关系不大,而与连接数据库和完善图表有关。

**Structure tree:**
```
main clause: there are problems
first relative clause: that we have
second relative clause: that have little to do with X and everything to do with Y
Y includes: connecting to databases AND perfecting iterations
nested relative clause: that go into manuscripts (modifying figures)
```

**Grammar points:**
- **双重定语从句** - 两个that从句连续修饰problems,第二个从句是主要内容
- **have (everything) to do with** - 固定搭配,表示'与...有关/无关'
- **动名词并列作介词宾语** - connecting和perfecting并列,作with的宾语

### [43:04]
**Original:** It's increasingly becoming the case that many scientific workflows are leaning more and more on these high-performance scientific computing jobs.

**Translation:** 越来越多的情况是,许多科学工作流程正越来越依赖这些高性能科学计算任务。

**Core structure:**
- It's becoming the case that workflows are leaning on computing jobs.  
  情况是工作流程正依赖计算任务。

**Structure tree:**
```
main clause: It's becoming the case
real subject: that workflows are leaning on jobs
modifiers:
- increasingly (程度副词)
- more and more (修饰leaning)
- high-performance scientific (修饰computing jobs)
```

**Grammar points:**
- **it作形式主语** - 真正主语是that从句,置于句末
- **lean on** - 依赖、倚靠,比depend on更口语化

### [45:01]
**Original:** Old databases with poorly documented schemas, pipelines that break when a dependency updates, Jupyter notebooks scattered throughout your messy file system with cells that were executed out of order, all those horrible hours spent making figures in Matplotlib and Illustrator.

**Translation:** 文档记录不佳的旧数据库、依赖项更新时就会崩溃的管道、散落在混乱文件系统中且单元格执行顺序混乱的Jupyter笔记本,以及所有那些花在Matplotlib和Illustrator上制作图表的糟糕时光。

**Core structure:**
- Old databases, pipelines, notebooks, all those hours.  
  旧数据库、管道、笔记本,所有那些时光。

**Structure tree:**
```
parallel noun phrases (列举式)
- databases (with modifier)
- pipelines (with subordinate clause)
- notebooks (with multiple modifiers)
- hours (with modifier)
each phrase has embedded modifiers
```

**Grammar points:**
- **平行列举结构** - 多个名词短语并列,每个都带有复杂修饰语,形成累积效果。
- **过去分词作后置定语** - executed out of order 和 spent making 修饰前面的名词。

### [46:11]
**Original:** Month one of a program like this is a lot of manual work: literature review, structure prep, doing the genetics, scoping the screen, building the business case.

**Translation:** 像这样的项目的第一个月是大量的人工工作:文献综述、结构准备、做遗传学分析、确定筛选范围、建立商业案例。

**Core structure:**
- Month one is manual work: review, prep, genetics, scoping, building.  
  第一个月是人工工作:综述、准备、遗传学、确定范围、建立案例。

**Structure tree:**
```
main clause: Month one is manual work
colon introduces list
parallel gerund/noun phrases:
- literature review
- doing the genetics
- building the business case
```

**Grammar points:**
- **冒号引出解释性列举** - 冒号后列举具体工作内容,解释manual work。
- **动名词与名词混合并列** - doing, scoping, building(动名词)与review, prep(名词)并列。

### [47:46]
**Original:** Now normally with plans like this, where it can take an hour or two to even conduct the computational screen or do the landscaping analysis, you would iterate with Claude to refine the plan, and in fact that is, you know, that is encouraged to make sure that it's going to do what you want it to do.

**Translation:** 通常对于这样的计划,仅仅进行计算筛选或做景观分析就可能需要一到两个小时,你会与Claude迭代来完善计划,事实上这是被鼓励的,以确保它会做你想让它做的事情。

**Core structure:**
- Normally you would iterate with Claude to refine the plan, and that is encouraged.  
  通常你会与Claude迭代来完善计划,这是被鼓励的。

**Structure tree:**
```
adverbial: normally with plans
relative clause: where it can take...
main clause: you would iterate
coordinate clause: that is encouraged
purpose clause: to make sure that...
```

**Grammar points:**
- **where引导定语从句** - where修饰plans,表示抽象地点(在这种情况下)。
- **嵌套的that从句** - make sure后跟宾语从句,其中又包含what从句作do的宾语。
- **插入语you know** - 口语化插入成分,增加理解难度。

### [49:02]
**Original:** Now, because the work is being done through sub-agents here in this product in particular, it's extremely important to make sure that you guys have as much transparency and visibility into what's going on at any given time as possible.

**Translation:** 现在,因为这项工作是通过子代理完成的,特别是在这个产品中,所以确保你们在任何给定时间都能尽可能多地了解正在发生的事情的透明度和可见性是极其重要的。

**Core structure:**
- It's important to make sure that you have transparency into what's going on.  
  确保你们能了解正在发生的事情是很重要的。

**Structure tree:**
```
because clause: work is being done
main clause: it's important (formal subject)
real subject: to make sure that...
that clause: you have transparency
into + what clause: what's going on
as...as possible structure
```

**Grammar points:**
- **形式主语it** - it作形式主语,真正主语是to make sure that从句。
- **as...as possible结构** - as much...as possible表示尽可能多,修饰transparency and visibility。

### [49:24]
**Original:** And you can see it's over 20 angstroms apart, so immediately we can tell that this is likely not a problem with a broken active site.

**Translation:** 你可以看到它们相距超过20埃,所以我们可以立即判断这很可能不是活性位点损坏的问题。

**Core structure:**
- It's 20 angstroms apart, so we can tell this is not a problem.  
  它们相距20埃,所以我们可以判断这不是一个问题。

**Structure tree:**
```
compound sentence with so
first clause: you can see (that) it's apart
second clause: we can tell that...
that clause: this is not a problem
with phrase modifying problem
```

**Grammar points:**
- **so连接因果复合句** - so引出结论,前句是观察,后句是推断。
- **双重宾语从句** - see后省略that,tell后显式使用that引导宾语从句。

### [50:34]
**Original:** Proteomics, structural biology, chemistry, genomics, literature review—altogether more than 60 plus databases and scientific resources that it has access to.

**Translation:** 蛋白质组学、结构生物学、化学、基因组学、文献综述——总共超过60多个它可以访问的数据库和科学资源。

**Core structure:**
- More than 60 databases and resources that it has access to.  
  超过60个它可以访问的数据库和资源。

**Structure tree:**
```
fragment: list of domains (Proteomics, structural biology...)
main clause: altogether more than 60 databases...
relative clause: that it has access to
```

**Grammar points:**
- **破折号引出解释说明** - 破折号前是列举，后是总结说明
- **限制性定语从句** - that从句修饰databases and resources

### [51:34]
**Original:** So going back to the figure I showed you initially, if you click here and look into the provenance of this artifact, you'll see that the artifact has the code that produced it, including the input artifacts it depends on, the full execution log of all the cells executed in that session leading up to the production of the artifact, the conversation around the artifact, the exact environmental snapshot that produced this artifact.

**Translation:** 回到我最初给你展示的图表，如果你点击这里并查看这个产出物的来源，你会看到该产出物包含生成它的代码，包括它所依赖的输入产出物、该会话中执行的所有单元的完整执行日志（这些执行导致了该产出物的生成）、围绕该产出物的对话、以及生成该产出物的确切环境快照。

**Core structure:**
- You'll see that the artifact has the code, including logs, conversation, and snapshot.  
  你会看到该产出物包含代码，包括日志、对话和快照。

**Structure tree:**
```
condition: if you click and look into...
main clause: you'll see that...
object clause: the artifact has the code...
nested relative: that produced it
participial phrase: including input artifacts...
nested relative: it depends on
participial phrase: executed in that session
participial phrase: leading up to...
```

**Grammar points:**
- **多层嵌套结构** - 包含条件状语、宾语从句、多个定语从句和分词短语
- **including引导补充信息** - 介词短语扩展主要内容，列举多项元素
- **现在分词作后置定语** - leading up to修饰前面的执行过程

### [52:22]
**Original:** Here, I think the label is hard to see, so I'm going to just click and say this label is hard to see.

**Translation:** 在这里，我认为标签很难看清楚，所以我打算直接点击并说这个标签很难看清楚。

**Core structure:**
- I think the label is hard to see, so I'm going to click and say this.  
  我认为标签很难看清，所以我打算点击并说出来。

**Structure tree:**
```
main clause 1: I think...
object clause: the label is hard to see
main clause 2: so I'm going to click and say...
object clause: this label is hard to see
```

**Grammar points:**
- **宾语从句** - think后接that从句(省略that)
- **be going to表将来** - 表示即将进行的动作

### [52:32]
**Original:** And I want to point out one thing here. I could say this without referring to what I'm labeling specifically, because Claude sees every annotation that you make with its vision capabilities as well as its ability to read text, of course.

**Translation:** 我想在这里指出一点。我可以在不具体指明我在标注什么的情况下说这句话，因为Claude通过其视觉能力以及当然还有文本阅读能力，能看到你做的每一个注释。

**Core structure:**
- I could say this without referring to what I'm labeling because Claude sees every annotation.  
  我可以不指明标注内容而说这句话，因为Claude能看到每个注释。

**Structure tree:**
```
main clause: I could say this
prepositional phrase: without referring to...
object of preposition: what I'm labeling
cause clause: because Claude sees...
object: every annotation that you make
prepositional phrase: with its vision capabilities...
```

**Grammar points:**
- **without + 动名词** - 表示在不做某事的情况下
- **what引导名词性从句** - what作referring to的宾语
- **that定语从句** - 修饰every annotation

### [54:14]
**Original:** To give you a sense of the power here, in my own personal research project where I continued my own PhD, at this point, Claude has produced thousands and thousands of artifacts all leading up to one output, a manuscript.

**Translation:** 为了让你感受一下这里的强大之处，在我自己继续攻读博士学位的个人研究项目中，到目前为止，Claude已经生成了成千上万个产出物，所有这些都通向一个输出——一份手稿。

**Core structure:**
- Claude has produced thousands of artifacts all leading up to one output.  
  Claude已经生成了成千上万个产出物，都通向一个输出。

**Structure tree:**
```
purpose: To give you a sense...
locative phrase: in my research project
relative clause: where I continued my PhD
time marker: at this point
main clause: Claude has produced artifacts
participial phrase: all leading up to...
appositive: a manuscript
```

**Grammar points:**
- **不定式表目的** - To give引导目的状语
- **where引导定语从句** - 修饰research project
- **现在分词短语作定语** - leading up to修饰artifacts
- **同位语** - a manuscript解释one output

### [55:05]
**Original:** We wanted to be as much a drop-in replacement for Jupyter notebooks as possible to make sure that computational biologists can use this without worrying about it not running where their data is.

**Translation:** 我们希望尽可能成为 Jupyter notebooks 的直接替代品,以确保计算生物学家可以使用它而不必担心它无法在他们数据所在的地方运行。

**Core structure:**
- We wanted to be a replacement to make sure that biologists can use this.  
  我们希望成为替代品,以确保生物学家可以使用它。

**Structure tree:**
```
main clause: We wanted to be...
purpose clause: to make sure that...
object clause: that biologists can use this
adverbial clause: without worrying about...
noun clause: it not running where their data is
```

**Grammar points:**
- **as...as possible 结构** - 表示尽可能达到某种程度
- **复合宾语从句** - worry about 后接复杂的名词性从句,包含否定和地点从句

### [55:55]
**Original:** So 2200 went in, 2100 or so were folded and assessed for binding affinity with the compounds of interest with the enzyme.

**Translation:** 所以输入了2200个,大约2100个被折叠并评估了与目标化合物和酶的结合亲和力。

**Core structure:**
- 2200 went in, 2100 were folded and assessed.  
  2200个输入,2100个被折叠和评估。

**Structure tree:**
```
compound sentence: clause1 + clause2
clause1: 2200 went in
clause2: 2100 were folded and assessed
prepositional phrases: for binding affinity / with the compounds / with the enzyme
```

**Grammar points:**
- **被动语态并列** - were folded and assessed 两个被动动作并列
- **多重介词短语修饰** - 连续的 for...with...with 结构易造成理解困难

### [56:20]
**Original:** We have built-in artifact renderers for images, but not just images. As you can see elsewhere here, markdown documents, HTML dashboards, chemistry structures, structure viewers, and more.

**Translation:** 我们内置了图像的工件渲染器,但不仅仅是图像。正如你在这里其他地方可以看到的,还有 markdown 文档、HTML 仪表板、化学结构、结构查看器等等。

**Core structure:**
- We have renderers for images and more.  
  我们有图像和更多内容的渲染器。

**Structure tree:**
```
main clause: We have renderers...
contrast: but not just images
parenthetical: As you can see elsewhere
long list: markdown documents, HTML dashboards... and more
```

**Grammar points:**
- **插入语** - As you can see elsewhere here 打断主要信息流
- **长列举** - 多个并列名词短语需要听者保持注意力追踪

### [57:52]
**Original:** It told us the first decisive experiment we would need to do to fund a larger experimental campaign and then it gave us the kill criteria for that experiment. At what point do we decide to look for another approach?

**Translation:** 它告诉我们需要做的第一个决定性实验,以资助更大规模的实验活动,然后它给了我们该实验的终止标准。在什么时候我们决定寻找另一种方法?

**Core structure:**
- It told us the experiment and gave us the criteria.  
  它告诉我们实验并给了我们标准。

**Structure tree:**
```
compound sentence: It told us... and it gave us...
modifier chain: experiment [we would need to do] [to fund campaign]
question: At what point do we decide...
```

**Grammar points:**
- **嵌套定语从句** - experiment 后跟省略 that 的定语从句,再跟目的状语
- **疑问句作解释** - 独立疑问句补充说明 kill criteria 的含义

### [01:00:54]
**Original:** There is a scientist at UCSF who was working on an RNA sequencing data analysis, and before using Claude Science, over the course of a year, something wasn't quite right with the data, and eventually a year later his team figured out that there was a virus contaminating the sample that explained the unusual result.

**Translation:** 加州大学旧金山分校有一位科学家在做 RNA 测序数据分析,在使用 Claude Science 之前,在一年的时间里,数据有些不太对劲,最终一年后他的团队发现有病毒污染了样本,这解释了异常结果。

**Core structure:**
- A scientist was working on analysis, and his team figured out that a virus explained the result.  
  一位科学家在做分析,他的团队发现病毒解释了结果。

**Structure tree:**
```
main clause: There is a scientist...
relative clause: who was working on...
compound: and something wasn't right, and team figured out...
nested that-clause: that there was a virus...
relative clause: that explained the result
```

**Grammar points:**
- **多层嵌套从句** - 定语从句内含 that 引导的宾语从句,其内又有定语从句
- **时间状语插入** - over the course of a year 和 eventually a year later 打断叙述节奏

### [01:02:24]
**Original:** And so in all of these examples, the important themes here are accelerating workflows by an order of magnitude in some cases and enabling small teams of scientists to do what previously would have taken, you know, much larger teams including a larger array of different skill sets.

**Translation:** 因此在所有这些例子中,重要的主题是在某些情况下将工作流程加速一个数量级,并使小型科学家团队能够完成以前需要更大团队(包括更多不同技能组合)才能完成的工作。

**Core structure:**
- The themes are accelerating workflows and enabling small teams to do what would have taken larger teams.  
  主题是加速工作流程并使小团队能够完成需要大团队才能完成的工作。

**Structure tree:**
```
main clause: the themes are...
parallel structure: accelerating... and enabling...
infinity phrase: to do what...
relative clause: what would have taken...
participle phrase: including a larger array...
```

**Grammar points:**
- **并列不定式结构** - accelerating 和 enabling 并列,共同作表语
- **what 引导宾语从句** - what 在从句中作主语,表示'...的事情'
- **虚拟语气 would have taken** - 表示与过去事实相反的假设

### [01:02:46]
**Original:** So in the life science world and taking drug development as an example, there's a lot else going on besides R&D. So in this presentation we focused mostly on R&D, but of course there is critical work happening in clinical and regulatory, in commercial, in manufacturing and operations, and our cloud offerings from the models and products that we have are designed to address all of it.

**Translation:** 在生命科学领域,以药物开发为例,除了研发之外还有很多其他工作在进行。在本次演示中我们主要关注研发,但当然在临床和监管、商业、制造和运营方面也有关键工作在进行,我们的云产品和模型旨在解决所有这些问题。

**Core structure:**
- There is critical work happening in multiple areas, and our cloud offerings are designed to address all of it.  
  在多个领域有关键工作在进行,我们的云产品旨在解决所有这些问题。

**Structure tree:**
```
compound sentence: but there is work... and offerings are designed...
present participle: happening in...
prepositional phrases: in clinical, in commercial, in manufacturing
relative clause: that we have
```

**Grammar points:**
- **there be 结构 + 现在分词** - happening 作后置定语,表示正在进行的工作
- **并列介词短语** - in clinical and regulatory, in commercial, in manufacturing 并列
- **be designed to do** - 被动语态表示目的和功能

### [01:04:50]
**Original:** So as Claude is working, it's going through the codebase and a key feature of how Claude Code works is that it's set up to flag you and stop if it needs your input on something. And we put a lot of work into making sure that the models have the right judgment and knowing when to grab your attention.

**Translation:** 当Claude工作时,它会检查代码库,Claude Code工作方式的一个关键特性是它被设置为在需要你输入时标记并停止。我们投入了大量工作来确保模型具有正确的判断力,知道何时需要引起你的注意。

**Core structure:**
- A key feature is that it's set up to flag you if it needs your input, and we put work into making sure the models have judgment.  
  关键特性是它被设置为在需要输入时标记你,我们投入工作确保模型有判断力。

**Structure tree:**
```
compound sentence: feature is that... and we put work into...
that clause: it's set up to flag...
conditional clause: if it needs...
gerund phrase: making sure that...
parallel gerunds: having judgment and knowing when...
```

**Grammar points:**
- **be set up to do** - 被动语态,表示'被设置为做某事'
- **put work into doing** - 固定搭配,into 后接动名词
- **并列动名词** - having 和 knowing 并列作宾语

### [01:06:21]
**Original:** So what we've seen here in this a little bit sped up demo is a team of agents through Cloud Code performing a migration from SAS to Python of a production biostats codebase that otherwise would have taken a team of software engineers and clinical scientists perhaps a few months and can be completed here within a single session over the course of a few hours.

**Translation:** 我们在这个稍微加速的演示中看到的是,一个代理团队通过Cloud Code执行了生产生物统计代码库从SAS到Python的迁移,这项工作原本需要一个软件工程师和临床科学家团队花费几个月时间,但在这里可以在几个小时的单次会话中完成。

**Core structure:**
- What we've seen is a team performing a migration that would have taken months and can be completed in hours.  
  我们看到的是一个团队执行迁移,原本需要几个月但可以在几小时内完成。

**Structure tree:**
```
subject clause: What we've seen...
present participle: performing a migration...
relative clause: that would have taken...
compound predicate: would have taken... and can be completed...
prepositional phrases: from SAS to Python, within a single session
```

**Grammar points:**
- **What 引导主语从句** - What 从句作主语,表示'我们看到的东西'
- **otherwise 虚拟语气** - otherwise 引出与事实相反的情况,用 would have done
- **并列复合谓语** - would have taken 和 can be completed 并列,对比时间差异

### [01:09:18]
**Original:** And we're doing this because we believe first and foremost that to build the right models and products and tools that accelerate the whole industry, we need to live it along with all of you. We believe in the power of tight feedback loops. And there's no substitute for having our own experiences alongside you all in the trenches trying to develop drugs.

**Translation:** 我们这样做是因为我们首先相信,为了构建能够加速整个行业的正确模型、产品和工具,我们需要与你们所有人一起身临其境。我们相信紧密反馈循环的力量。在战壕中与你们一起尝试开发药物,拥有我们自己的经验是无可替代的。

**Core structure:**
- We believe that to build the right tools, we need to live it with you, and there's no substitute for having our own experiences.  
  我们相信为了构建正确的工具,我们需要与你们一起体验,拥有自己的经验是无可替代的。

**Structure tree:**
```
reason clause: because we believe that...
infinity of purpose: to build the right...
relative clause: that accelerate...
there be structure: there's no substitute for...
gerund phrase: having our own experiences
present participle: trying to develop...
```

**Grammar points:**
- **不定式表目的** - to build 作目的状语,说明需要做某事的原因
- **there's no substitute for** - 固定句型,'...是无可替代的',for 后接动名词
- **现在分词作状语** - trying to develop 伴随状语,描述 in the trenches 时的动作

### [01:11:02]
**Original:** We've all probably experienced a new technology, whether it be CRISPR, next generation sequencing, or perhaps even a computational collaborator that unearths and opens up our mind to what is possible and how we can pursue that science.

**Translation:** 我们可能都经历过一种新技术，无论是CRISPR、下一代测序，还是甚至可能是一个计算协作者，它揭示并开启我们的思维，让我们了解什么是可能的以及我们如何追求那门科学。

**Core structure:**
- We've all experienced a technology that opens up our mind to what is possible.  
  我们都经历过一种技术，它开启我们的思维，让我们了解什么是可能的。

**Structure tree:**
```
main clause: We've all experienced a technology
whether clause: whether it be CRISPR... (examples)
relative clause: that unearths and opens up our mind
to-clause: to what is possible and how we can pursue
```

**Grammar points:**
- **whether引导让步从句** - 列举多个可能性的例子，使用虚拟语气(be动词)
- **宾语从句嵌套** - to后接what/how引导的两个并列宾语从句

### [01:11:15]
**Original:** But it's also this big question, and even an experimental question, of how we integrate that into our work, how we integrate that into our organizations, and how we start to rethink the questions that are possible.

**Translation:** 但这也是一个大问题，甚至是一个实验性问题，关于我们如何将其整合到我们的工作中，如何将其整合到我们的组织中，以及我们如何开始重新思考那些可能的问题。

**Core structure:**
- It's a question of how we integrate that and rethink the questions.  
  这是一个关于我们如何整合它并重新思考问题的问题。

**Structure tree:**
```
main clause: it's a question
of-phrase: of how we integrate... (3 parallel clauses)
  - how we integrate into work
  - how we integrate into organizations
  - how we rethink questions
```

**Grammar points:**
- **三个并列how从句** - of后接三个平行的间接疑问句作同位语
- **定语从句** - that are possible修饰questions

### [01:13:01]
**Original:** It's not just complex, it's actually just huge, like huge, huge. All the numbers are big no matter what they are, and it's huge enough that it's bigger than our experimental capacity ever.

**Translation:** 它不仅仅是复杂，它实际上非常巨大，就是巨大，巨大。所有的数字都很大，无论它们是什么，而且它巨大到超过了我们的实验能力。

**Core structure:**
- It's huge, and it's bigger than our experimental capacity.  
  它很巨大，而且它超过了我们的实验能力。

**Structure tree:**
```
compound sentence: It's huge, and it's huge enough
subordinate clause: no matter what they are
result clause: that it's bigger than our capacity
```

**Grammar points:**
- **no matter what引导让步状语从句** - 表示无论什么情况都成立
- **enough...that结果状语** - 表示程度足以导致某种结果

### [01:14:09]
**Original:** And then on top of all of those things, that piece that people often call complexity is within this vastness having to wade through it.

**Translation:** 然后在所有这些事情之上，人们常常称之为复杂性的那部分，就是在这种巨大性中必须艰难跋涉穿越它。

**Core structure:**
- That piece is having to wade through the vastness.  
  那部分就是必须艰难穿越这种巨大性。

**Structure tree:**
```
main clause: that piece is having to wade through
relative clause: that people often call complexity
prepositional phrase: within this vastness (location)
```

**Grammar points:**
- **定语从句修饰主语** - that people call complexity修饰piece，嵌入主语中
- **动名词having to作表语** - 表示一种必要的状态或行为

### [01:15:26]
**Original:** And models and it needs GPUs, but that's not enough. It needs a lot of data and it needs really the ability to understand whether what it did means anything, and that means that it needs iterations.

**Translation:** 还有模型，它需要GPU，但这还不够。它需要大量数据，而且它真正需要理解它所做的事情是否有意义的能力，这意味着它需要迭代。

**Core structure:**
- It needs data and the ability to understand whether what it did means anything.  
  它需要数据和理解它所做的事情是否有意义的能力。

**Structure tree:**
```
compound: It needs data and it needs ability
noun clause: the ability to understand whether...
whether clause: whether what it did means anything
  - subject clause: what it did
that clause: that it needs iterations
```

**Grammar points:**
- **多层嵌套从句** - whether从句中嵌套what主语从句，结构复杂
- **that引导表语从句** - 解释前面that(代词)的具体含义

### [01:17:05]
**Original:** Humans say the same, but that was just the starting point. When you needed to figure out the mechanism of action, you actually needed the kind of experimental science and thinking that AI cannot help you with right now.

**Translation:** 人类也这么说，但那只是起点。当你需要弄清楚作用机制时，你实际上需要那种AI目前还无法帮助你的实验科学和思维方式。

**Core structure:**
- When you needed to figure out the mechanism, you needed the science that AI cannot help you with.  
  当你需要弄清楚机制时，你需要AI无法帮助你的科学。

**Structure tree:**
```
main clause: you needed the science and thinking
temporal clause: When you needed to figure out...
relative clause: that AI cannot help you with
end-weight prep phrase: with right now
```

**Grammar points:**
- **When引导时间状语从句** - 从句在主句前，需注意逻辑关系
- **定语从句中介词后置** - that AI cannot help you with，介词with后置到从句末尾

### [01:17:43]
**Original:** That's why I say the answer is in the model, not in the data. Before, it was only in the data, and people have to change how they operate with it, and that shift is also quite difficult.

**Translation:** 这就是为什么我说答案在模型中，而不在数据中。以前，它只在数据中，人们必须改变他们使用它的方式，而这种转变也相当困难。

**Core structure:**
- People have to change how they operate, and that shift is difficult.  
  人们必须改变他们的操作方式，这种转变很困难。

**Structure tree:**
```
compound sentence: clause1 + and + clause2 + and + clause3
clause2: people have to change...
embedded question: how they operate with it
clause3: that shift is difficult
```

**Grammar points:**
- **嵌入式疑问句作宾语** - how they operate作change的宾语，用陈述语序
- **指示代词that指代前文** - that shift指代前面提到的整个改变过程

### [01:18:45]
**Original:** The first bet is that AI and machine learning can really help us identify those hidden patterns in biology that ultimately will enable us to drug previously undruggable targets and hopefully be able to bring the next new medicine to patients.

**Translation:** 第一个赌注是，AI和机器学习真的可以帮助我们识别生物学中那些隐藏的模式，这些模式最终将使我们能够对以前无法成药的靶点进行药物开发，并有望能够将下一个新药带给患者。

**Core structure:**
- AI can help us identify patterns that will enable us to drug targets and bring medicine to patients.  
  AI可以帮助我们识别模式，这些模式将使我们能够开发靶点药物并将药物带给患者。

**Structure tree:**
```
main clause: The first bet is that...
predicative clause: AI can help us identify patterns
relative clause: that will enable us to...
parallel infinitives: to drug targets and to bring medicine
```

**Grammar points:**
- **that引导表语从句** - 说明bet的具体内容
- **定语从句修饰patterns** - that从句解释patterns的作用
- **动词drug的特殊用法** - drug作动词表示'对...进行药物开发'

### [01:21:18]
**Original:** I mean this is an industry where we spend 150 to 200 billion a year in drug R&D amongst the bigger companies, and yet we've only found a handful of medicines, which I think actually shows you how hard this is—unpacking billions of years of evolution.

**Translation:** 我的意思是，这是一个在大公司中每年在药物研发上花费1500到2000亿美元的行业，然而我们只找到了少数几种药物，我认为这实际上向你展示了这件事有多难——解开数十亿年的进化过程。

**Core structure:**
- This is an industry where we spend billions, yet we've only found a handful of medicines.  
  这是一个花费数十亿的行业，但我们只找到了少数药物。

**Structure tree:**
```
main clause: this is an industry
relative clause: where we spend...
contrast clause: and yet we've only found...
non-restrictive clause: which shows...
embedded question: how hard this is
appositive: unpacking evolution
```

**Grammar points:**
- **and yet表示转折对比** - 强调前后对比的意外性
- **破折号引出同位语** - unpacking...进一步解释this指代的内容
- **嵌套从句结构** - which从句中又包含how引导的从句

### [01:22:36]
**Original:** So what does that translate to? That means that I think, as Chris rightfully said, you can get this down from 12 years—from when we actually have a candidate to the end of this journey—down to seven to eight years, which if you compound over this entire industry is massive.

**Translation:** 那么这意味着什么呢？这意味着，我认为，正如Chris正确指出的，你可以把这个时间从12年——从我们实际拥有候选药物到整个旅程结束——降低到7到8年，如果你在整个行业复合计算的话，这是巨大的。

**Core structure:**
- You can get this down from 12 years to 7-8 years, which is massive.  
  你可以把时间从12年降到7-8年，这是巨大的。

**Structure tree:**
```
main clause: That means that you can get this down
parenthetical: as Chris rightfully said
time range: from 12 years down to 7-8 years
appositive dash: from when we have...to the end
relative clause: which is massive
conditional: if you compound
```

**Grammar points:**
- **插入语的使用** - as Chris said作为插入成分，打断主句
- **破折号插入补充说明** - 进一步解释12年的具体含义
- **条件从句嵌套在定语从句中** - which从句中包含if条件从句，结构复杂

### [01:23:52]
**Original:** And if we were actually able to, across all of our pipelines, go from 12 years to seven years and 8% POS to 16%, the impact on public health is massive.

**Translation:** 如果我们真的能够在所有管线中将时间从12年缩短到7年，将成功率从8%提高到16%，那么对公共健康的影响将是巨大的。

**Core structure:**
- If we were able to go from 12 years to seven years and 8% to 16%, the impact is massive.  
  如果我们能够从12年缩短到7年，从8%提高到16%，影响将是巨大的。

**Structure tree:**
```
条件句: If we were able to...
  插入语: across all of our pipelines
  并列结构: go from X to Y and from A to B
主句: the impact on public health is massive
```

**Grammar points:**
- **虚拟语气（与现在事实相反）** - if + were able to 表示假设，主句用现在时表示可能的结果
- **插入语** - across all of our pipelines 插在动词短语中间，增加句子复杂度

### [01:25:13]
**Original:** And also as models can let you work more end to end rather than look at each step separately, then they also can be more foreshadowing challenges.

**Translation:** 而且，由于模型可以让你进行更多端到端的工作，而不是单独查看每个步骤，那么它们也可以更好地预见挑战。

**Core structure:**
- As models let you work end to end, they can foreshadow challenges.  
  由于模型让你端到端工作，它们可以预见挑战。

**Structure tree:**
```
原因状语从句: as models can let you work...
  对比结构: more end to end rather than look at each step
主句: they can be more foreshadowing challenges
```

**Grammar points:**
- **as 引导原因状语从句** - as 表示因果关系，相当于 because
- **rather than 对比结构** - 连接两个平行的动作，表示选择其一而非另一

### [01:25:33]
**Original:** We look actually as an industry at a very small sliver of them because of this difficulty to manage the comprehensiveness and because of the inability to really track experimentally what you call the biological latency.

**Translation:** 作为一个行业，我们实际上只关注其中很小的一部分，这是因为难以管理全面性，也因为无法真正通过实验追踪你所说的生物潜伏期。

**Core structure:**
- We look at a very small sliver of them because of the difficulty and inability.  
  我们只关注其中很小的一部分，因为难度和无能为力。

**Structure tree:**
```
主句: We look at a very small sliver
  插入语: actually as an industry
原因状语: because of...
  并列原因1: difficulty to manage comprehensiveness
  并列原因2: inability to track what you call biological latency
```

**Grammar points:**
- **插入语打断主干** - actually as an industry 插在主谓之间，干扰理解
- **what 引导的名词性从句** - what you call... 作 track 的宾语，意为"你所称的..."

### [01:27:13]
**Original:** And that worked. But what we found was that the use cases tended to be quite narrow and were very difficult to scale.

**Translation:** 这种做法有效。但我们发现使用案例往往相当狭窄，而且很难扩展。

**Core structure:**
- What we found was that the use cases were narrow and difficult to scale.  
  我们发现使用案例很狭窄且难以扩展。

**Structure tree:**
```
主句: What we found was that...
  主语从句: What we found
  表语从句: that use cases tended to be...
    并列谓语1: tended to be narrow
    并列谓语2: were difficult to scale
```

**Grammar points:**
- **What 引导主语从句** - What we found 整体作主语
- **并列谓语结构** - tended to be 和 were 并列，共用同一主语

### [01:27:53]
**Original:** I'll give you one sort of somewhat mundane example, but we knew two years ago that AI could predict and forecast our business and many parts of our businesses anyway better than the battalions of people that we had actually doing the forecasting, but we couldn't scale it.

**Translation:** 我给你举一个有点平凡的例子，但我们两年前就知道，AI能够预测和预报我们的业务以及业务的许多部分，无论如何都比我们实际用来做预测的大批人员做得更好，但我们无法扩展它。

**Core structure:**
- We knew that AI could predict our business better than the people, but we couldn't scale it.  
  我们知道AI能比人员更好地预测业务，但我们无法扩展它。

**Structure tree:**
```
主句1: we knew that...
  宾语从句: AI could predict... better than...
    比较对象: the battalions of people (定语从句修饰)
转折主句2: but we couldn't scale it
```

**Grammar points:**
- **多重修饰结构** - many parts of our businesses anyway 多个修饰成分叠加
- **that 引导定语从句** - 修饰 battalions of people，说明这些人在做什么
- **比较结构** - better than 连接两个对比对象：AI vs. people

### [01:30:13]
**Original:** And within not a long period of time, a few months, people were like, why is the foundation model not trained yet? They were realizing how much more is in a model than in data.

**Translation:** 在不长的时间内，几个月后，人们开始问，为什么基础模型还没训练好？他们意识到模型中包含的东西比数据中多得多。

**Core structure:**
- Within a few months, people asked why the model wasn't trained yet.  
  几个月内，人们问为什么模型还没训练好。

**Structure tree:**
```
time phrase: within not a long period of time
appositive: a few months
main clause 1: people were like...
quoted question: why is the model not trained yet?
main clause 2: They were realizing...
object clause: how much more is in a model than in data
```

**Grammar points:**
- **双重否定表达** - not a long period 实际表示 a short period，口语中常见的委婉表达
- **be like 引入间接引语** - 口语化表达，相当于 say/think
- **比较结构 more...than** - how much more...than 用于对比模型和数据所含信息量

### [01:31:04]
**Original:** And when that happens, they're like, why should I spend synthesis money just to make the model better? Even though that would be good for all projects, I need to move my project forward.

**Translation:** 当这种情况发生时，他们会说，为什么我要花合成费用只是为了让模型变得更好？尽管这对所有项目都有好处，但我需要推进我自己的项目。

**Core structure:**
- When that happens, they ask why they should spend money to improve the model.  
  当这种情况发生时，他们问为什么要花钱改进模型。

**Structure tree:**
```
time clause: when that happens
main clause: they're like...
rhetorical question: why should I spend...?
purpose clause: just to make the model better
concessive clause: Even though that would be good...
contrasting statement: I need to move my project forward
```

**Grammar points:**
- **反问句表达不情愿** - why should I...? 表示质疑或不愿意做某事
- **让步状语从句** - Even though 引导让步，承认某事但不改变主要立场

### [01:33:10]
**Original:** And now if you multiply it across the whole industry, and tools like the ones that we saw today will surely get there, then the volume just in the whole system all of a sudden rises.

**Translation:** 现在如果你把它乘以整个行业，而且像我们今天看到的这些工具肯定会达到那个水平，那么整个系统中的数量会突然上升。

**Core structure:**
- If you multiply it across the industry, then the volume rises.  
  如果你把它乘以整个行业，那么数量就会上升。

**Structure tree:**
```
conditional clause: if you multiply it...
inserted clause: and tools...will surely get there
main clause: then the volume rises
modifiers: just in the whole system, all of a sudden
```

**Grammar points:**
- **条件句插入结构** - 在 if 从句和 then 主句之间插入补充信息，增加句子复杂度
- **multiply across** - 表示扩展应用到更大范围

### [01:34:16]
**Original:** But in my mind, my worry is that employees and society in general goes on autopilot, because when you have tools that can create content, they can edit content, summarize, take action on your behalf, it's very easy to phone it in and effectively say, well, you know what, I'm going to stop checking my homework or interrogating the output.

**Translation:** 但在我看来，我担心的是员工和整个社会会进入自动驾驶模式，因为当你拥有可以创建内容、编辑内容、总结、代表你采取行动的工具时，很容易就敷衍了事，并且有效地说，好吧，你知道吗，我不打算再检查我的作业或质疑输出结果了。

**Core structure:**
- My worry is that people go on autopilot because it's easy to stop checking.  
  我担心的是人们会进入自动模式因为很容易停止检查。

**Structure tree:**
```
main clause: my worry is that...
that-clause: employees and society goes on autopilot
reason clause: because when you have tools...
time clause: when you have tools that...
relative clause: that can create, edit, summarize...
result clause: it's very easy to phone it in
and clause: and say...I'm going to stop...
```

**Grammar points:**
- **多重嵌套从句** - that 表语从句内嵌 because 原因从句，原因从句内又嵌 when 时间从句和 that 定语从句
- **phone it in** - 俚语，表示敷衍了事、不认真对待工作
- **动词并列省略** - tools that can create, edit, summarize... 省略了重复的 can

### [01:36:41]
**Original:** It happens to large and illustrious companies over their many, many decades and more than decades, centuries of existence.

**Translation:** 这种情况发生在大型且杰出的公司身上，在它们存在的许多许多十年，甚至不止十年，几个世纪的时间里。

**Core structure:**
- It happens to companies over centuries of existence.  
  这发生在公司几个世纪的存在过程中。

**Structure tree:**
```
main clause: It happens to companies
modifiers: large and illustrious
time phrase: over their many decades...
gradual expansion: decades → more than decades → centuries
prepositional phrase: of existence
```

**Grammar points:**
- **递进式时间跨度表达** - many decades → more than decades → centuries 层层递进强调时间长度
- **happen to** - 表示某事发生在某对象身上

### [01:37:30]
**Original:** And then you can also mitigate for that, especially in our field, by using the real world, which is the lab, to let you kind of escape free from where you're already at.

**Translation:** 然后你也可以缓解这个问题，尤其是在我们的领域，通过使用现实世界（也就是实验室），让你从你目前所在的位置摆脱出来。

**Core structure:**
- You can mitigate for that by using the real world to let you escape.  
  你可以通过使用现实世界来缓解这个问题，让你摆脱出来。

**Structure tree:**
```
main clause: you can mitigate for that by using...
method: by using the real world
appositive: which is the lab
purpose: to let you escape free from where you're already at
location clause: where you're already at
```

**Grammar points:**
- **非限制性定语从句** - which is the lab 作为插入语解释 the real world
- **不定式表目的** - to let you escape 表示使用实验室的目的
- **where 引导地点状语从句** - where you're already at 描述当前所处的位置

### [01:38:37]
**Original:** If we can better model dosing and not have to do as much dose range finding, I mean, all of these things are possible, but we also need to bring the drug regulator along with us to get the full benefits of speed and probability of success we've been talking about.

**Translation:** 如果我们能更好地模拟剂量，而不必做那么多的剂量范围探索，我的意思是，所有这些事情都是可能的，但我们也需要让药物监管机构与我们同行，以获得我们一直在谈论的速度和成功概率的全部好处。

**Core structure:**
- All of these things are possible, but we need to bring the drug regulator along to get the full benefits.  
  所有这些事情都是可能的，但我们需要让药物监管机构同行以获得全部好处。

**Structure tree:**
```
conditional: If we can better model...
main clause 1: all of these things are possible
conjunction: but
main clause 2: we need to bring the drug regulator along
purpose: to get the full benefits
attribute: of speed and probability of success
relative clause: we've been talking about
```

**Grammar points:**
- **条件状语从句** - If 从句描述前提条件
- **bring...along with us** - 习语，意为'让...与我们同行/一起前进'
- **省略关系代词的定语从句** - we've been talking about 修饰 benefits，省略了 that/which

### [01:40:25]
**Original:** I think it is reasonable to assume that people's stress levels are rising, and increasingly so, in companies, in other scientific communities like in academia, in society as a whole.

**Translation:** 我认为可以合理地假设，人们的压力水平正在上升，而且越来越严重，在公司里、在其他科学社区（比如学术界）、在整个社会都是如此。

**Core structure:**
- It is reasonable to assume that people's stress levels are rising.  
  可以合理地假设人们的压力水平正在上升。

**Structure tree:**
```
main clause: I think it is reasonable to assume that...
formal subject: it
real subject: to assume that...
object clause: that people's stress levels are rising
parallel modifiers: in companies / in other scientific communities / in society
```

**Grammar points:**
- **形式主语 it** - it 作形式主语，真正主语是 to assume that 从句
- **increasingly so** - so 代替前面提到的 rising，避免重复
- **并列介词短语** - 三个 in... 短语并列，列举压力上升的不同场景

### [01:41:37]
**Original:** We started to see this kind of gnarly side effect that fortunately we had the data curated, we had AI tools sitting on top of it, and within a week, we were able to very clearly articulate a very small change that had been made in manufacturing that probably saved the program and certainly saved the program 6 to 12 months of time.

**Translation:** 我们开始看到这种棘手的副作用，幸运的是我们已经整理了数据，我们有人工智能工具在其上运行，而且在一周内，我们能够非常清楚地阐明在制造过程中进行的一个非常小的改变，这个改变很可能挽救了这个项目，并且肯定为该项目节省了6到12个月的时间。

**Core structure:**
- We saw a side effect, we had the data and AI tools, and we were able to articulate a change that saved the program.  
  我们看到了副作用，我们有数据和工具，我们能够阐明一个挽救了项目的改变。

**Structure tree:**
```
clause 1: We started to see this side effect
clause 2: we had the data curated
clause 3: we had AI tools sitting on top of it
clause 4: we were able to articulate a change
relative clause 1: that had been made in manufacturing
relative clause 2: that saved the program
```

**Grammar points:**
- **多重定语从句嵌套** - 第一个 that 修饰 change，第二个 that 也修饰 change，形成并列修饰
- **have + 宾语 + 过去分词** - we had the data curated 表示'让数据被整理'
- **并列句式** - 多个独立分句通过逗号和 and 连接，描述一系列相关事件

### [01:42:04]
**Original:** And if we don't figure it out, the thing I worry about and I hope I'm proven wrong on is at some point organizations, certainly large organizations, are going to become put under pressure to ration the use of this technology.

**Translation:** 如果我们不弄清楚这一点，我担心的事情——我希望我在这一点上被证明是错的——是在某个时刻，组织，尤其是大型组织，将会被迫对这项技术的使用进行限量配给。

**Core structure:**
- If we don't figure it out, the thing I worry about is organizations are going to be put under pressure.  
  如果我们不弄清楚，我担心的事情是组织将会面临压力。

**Structure tree:**
```
condition: if we don't figure it out
main subject: the thing I worry about and I hope I'm proven wrong on
main verb: is
predicative clause: organizations are going to become put under pressure
purpose: to ration the use of this technology
```

**Grammar points:**
- **主语中的双重定语从句** - I worry about 和 I hope I'm proven wrong on 都修饰 the thing
- **become put under pressure** - become + 过去分词，表示'变得被...'
- **不定式表目的** - to ration 说明施加压力的目的

### [01:44:25]
**Original:** Eric and Alec walked you through the basis of our claim: models that meaningfully improve in biology with every release, and a workbench that runs the analysis your scientists actually need.

**Translation:** Eric 和 Alec 向你们介绍了我们主张的基础：每次发布都在生物学领域有实质性改进的模型，以及一个能运行你们的科学家真正需要的分析的工作平台。

**Core structure:**
- Eric and Alec walked you through the basis of our claim.  
  Eric 和 Alec 向你们介绍了我们主张的基础。

**Structure tree:**
```
main clause: Eric and Alec walked you through the basis
appositive: models... and a workbench...
parallel structure: models that... and a workbench that...
relative clauses: that improve / that runs
```

**Grammar points:**
- **冒号引出同位语** - 冒号后解释说明 the basis of our claim 的具体内容
- **并列定语从句** - 两个 that 从句分别修饰 models 和 workbench

### [01:44:38]
**Original:** Run pipelines, figures, every step reproducible and traceable.

**Translation:** 运行流程、生成图表，每一步都可重现和可追踪。

**Core structure:**
- Every step is reproducible and traceable.  
  每一步都可重现和可追踪。

**Structure tree:**
```
imperative clauses: Run pipelines, (run) figures
absolute construction: every step reproducible and traceable
ellipsis: (being) reproducible and traceable
```

**Grammar points:**
- **独立主格结构** - every step + 形容词，省略 being，表示伴随状态
- **祈使句省略** - Run 后并列省略了第二个动词

### [01:44:54]
**Original:** To build an intelligence platform where scientific discovery truly happens is going to be a collaboration with the industry and our partners represented here.

**Translation:** 要构建一个真正发生科学发现的智能平台，将需要与在座的行业和合作伙伴协作。

**Core structure:**
- To build a platform is going to be a collaboration.  
  构建平台将需要协作。

**Structure tree:**
```
subject: To build an intelligence platform...
relative clause: where scientific discovery truly happens
predicate: is going to be
object: a collaboration with...
participle modifier: represented here
```

**Grammar points:**
- **不定式短语作主语** - To build... 整个短语作句子主语
- **过去分词后置定语** - represented here 修饰 partners

### [01:45:36]
**Original:** In the end, this is about time and the question, giving your scientists back what they trained for and you deciding which questions they'll spend it on.

**Translation:** 归根结底，这关乎时间和问题本身，把科学家训练的内容还给他们，并由你决定他们将在哪些问题上花费时间。

**Core structure:**
- This is about time and the question.  
  这关乎时间和问题。

**Structure tree:**
```
main clause: this is about time and the question
parallel participle phrases:
  - giving your scientists back what they trained for
  - you deciding which questions they'll spend it on
embedded clauses: what they trained for / which questions...
```

**Grammar points:**
- **现在分词短语作状语** - giving 和 deciding 并列，表示伴随或补充说明
- **动名词复合结构** - you deciding 中 you 作逻辑主语