The Briefing: AI for Science

Hey hey hey.

嘿嘿嘿。

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。

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。

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。

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。

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。

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。

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。

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。

Coding has irreversibly changed.

编程已经发生了不可逆转的改变。

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。

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。

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。

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。

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。

The experiment happens at the bench.

实验在实验台上进行。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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

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?

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。

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。

Yeah. So, first of all, thank you both

是的。首先,感谢你们二位

For coming.

感谢你来。

We're so thrilled.

我们非常激动。

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。

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。

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。

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。

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。

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。

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——

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。

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,

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。

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,

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。

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?

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。

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。

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

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?

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,

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?

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?

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。

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。

And another area is recruiting patients for clinical trials.

另一个领域是为临床试验招募患者。

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。

So it must be possible to do that faster.

所以一定有办法加快这个速度。

I also think the regulatory...

我还认为监管方面……

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。

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。

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?

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。

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。

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——

What AI can do here?

AI在这方面能做什么呢?

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。

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。

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。

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。

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。

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?

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。

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。

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?

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...

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?

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。

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。

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。

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。

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。

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?

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。

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。

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。

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?

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?

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。

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?

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。

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。

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。

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。

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。

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。

I just wonder if you have a thought there, L.

我想知道你对此有什么想法,L。

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。

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?

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。

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。

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。

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...

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。

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。

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。

How do you guard against that, Daria?

Daria，你们如何防范这种情况？

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。

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？

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。

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。

It is not yet turned on for biology.

但生物学领域的窗口期还没有开启。

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。

Biology—

生物学——

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。

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。

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。

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?

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——

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？

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？

And they have their own protocols for it.

他们有自己的一套处理规程。

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。

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。」

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。

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。

How do you

你如何

Think about these risk problems?

如何看待这些风险问题?

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。

Who would that person be?

那会是什么样的人呢?

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...

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。

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。

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。

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。

So, Dario, I asked Claude what I should ask you.

那么 Dario,我问了 Claude 我应该问你什么。

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。

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?

I thought it would actually—

我觉得这个问题确实——

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。

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。

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。

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——

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?

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。

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。

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?

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。

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」。

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。

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。

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?

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...

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。

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。

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。

Yeah, yeah.

是啊,是啊。

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。

That we haven't asked him yet.

我们还没问过他的问题。

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。

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……

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?

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,

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。

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。

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。

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?

I wanted to start with Daria.

我想先从 Daria 开始。

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。

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。

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。

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。

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。

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。

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……

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。

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。」

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。

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。

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。

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。

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。

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。

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——

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?

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。

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。

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。

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。

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

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。

It's great to be here with you all today.

很高兴今天能和大家在这里相聚。

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,

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。

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。

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。

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。

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。

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。

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。

Everything that we're doing is constructed to build

我们所做的一切都是为了构建

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。

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。

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。

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。

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?

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。

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。

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。

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

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。

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。

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……

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。

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。

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。

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。

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。

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。

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。

So every time that you create a figure...

所以每次你创建一个图表时……

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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……

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。

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。

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。

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。

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。

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。

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。

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。

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。

The disease is phenylketonuria, PKU.

这种疾病是苯丙酮尿症,简称 PKU。

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。

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。

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。

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。

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。

So to start, I just gave Claude one sentence.

所以一开始,我只给 Claude 一句话。

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。

Claude went ahead and built a plan.

Claude 随即制定了一个计划。

So this plan it built has three phases.

它制定的这个计划分为三个阶段。

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。

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。

First, it'll study the variant biology.

首先,它会研究变体的生物学特性。

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。

And then it'll build the investment case.

接着它会构建投资论证。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

You are the structure and product sub.

你是结构和产品分支。

agent for a PAH stabilizer discovery

用于 PAH 稳定剂发现项目的子智能体。

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...

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

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。

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。

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。

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。

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。

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?

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。

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。

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。

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。

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。

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。

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

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。

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。

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。

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。

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。

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。

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。

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。

What this means is that you can come

这意味着你可以

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

And artifacts are checked.

而且生成物都经过了检查。

Going back to the structure and pocket sub-agent,

回到结构和 pocket 子代理,

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。

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。

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。

You can see the diff here.

你可以在这里看到差异对比。

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。

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。

Okay, going back to the campaign.

好,回到这个研究项目上。

So, what's next?

那么,接下来是什么?

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。

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。

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。

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。

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。

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。

And then it did the scoring.

然后它进行了打分。

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。

And that leads us to the deliverables.

这就引出了我们的交付成果。

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。

Ultimately Claude produced a dashboard.

最终 Claude 生成了一个仪表板。

Let me zoom out a little bit here.

让我在这里缩小一点。

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。

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。

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。

And you can iterate with Claude on HTML

而且你可以与 Claude 在 HTML 上进行迭代

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。

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。

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。

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。

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?

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。

Now, why stop there?

那么,为什么要止步于此呢?

Why do it just for one disease?

为什么只针对一种疾病呢?

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。

At the end of it, we found out of 100

最终,我们从 100 种疾病中发现

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。

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。

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?

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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]。

Workflows that include the computational part as well.

包含计算部分在内的工作流程。

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。

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。

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。

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。

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。

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

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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?

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。

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。

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。

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。

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。

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。

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

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。

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。

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。

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。

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。

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。

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。

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。

Please welcome Chris, Aviv, and Vas.

欢迎 Chris、Aviv 和 Vas。

Great. Thank you all for joining us.

很好。感谢大家的参与。

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。

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——

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。

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。

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。

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。

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……

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

It's not the same as it used to be.

这已经不像过去那样了。

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。

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……

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。

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。

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……

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。

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。

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。

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。

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。

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。

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。

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。

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...

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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

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。

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。

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。

Information latency, operational latency, and biological latency.

信息延迟、运营延迟和生物学延迟。

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%。

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。

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。

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%。

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......

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。

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。

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。

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。

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。

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。

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%。

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。

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。

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。

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。

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。

All the steps of this process are very difficult.

这个过程的每一步都非常困难。

Very difficult.

非常困难。

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...

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。

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。

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。

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...

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,

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。

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。

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

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?

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。

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。

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。

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。

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。

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。

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。

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?

Where can you change the processes leveraging this technology?

哪里可以利用这项技术来改变流程?

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。

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。

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。

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。

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。

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。

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?

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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...

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。

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。

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。

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

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。

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。

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。

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。

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。

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。

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。

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。

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。

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……

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。

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。

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。

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。

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

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。

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。

Uh-huh.

嗯哼。

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。

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。

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。

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。

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。

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。

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。

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...

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。

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。

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。

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。

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。

Instilling this discernment.

培养这种辨别力。

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。

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。

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。

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。

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。

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……

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。

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?

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。

Um,

嗯,

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

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。

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。

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。

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,

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。

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。

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

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。

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。

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。

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。

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。

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。

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。

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。

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。

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...

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。

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。

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。

Thank you all.

谢谢大家。

Great. Thank you.

很好。谢谢。

Please welcome back to the stage Zoubin Ghahramani.

请欢迎 Zoubin Ghahramani 重返舞台。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

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。

Hey hey hey hey hey hey hey.

嘿嘿嘿嘿嘿嘿嘿。