本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
我们最重要的指导原则之一就是简单。
One of our big guiding principles is just simplicity.
比如，你看像 Chai-one 这样的模型，我觉得里面大概有 23 个独立的子模块。
So, like, when you look at a model like, let's say, Chai-one, I think there are 23 distinct submodules in Chai-one.
当你想针对这种结构做迭代时，会非常困难，因为你得去理解每个子模块各自的功能。
And, like, when you're trying to iterate on something like that, it gets really hard because you're like, I kinda I need to understand each of these sub modules independently.
你得搞清楚它们所有的行为和动态关系，但这其实很难扩展。
I need to understand all of their behaviors, their dynamics, and, like, that doesn't actually scale that well.
所以，你可以开始思考，好吧。
So, like, you can start to think, okay.
我该怎么简化这个？
How do I simplify this?
我该怎么识别，比如，什么才是真正重要的？
How do I identify, like, what's what's really important?
一旦你有了这些，整个研究过程和识别这类扩展方向就会变得简单得多。
Once you have that, the whole research process and identifying these types of scaling directions becomes a lot simpler.
今天我们请到了 Josh 和 Matt，他们是 Chai Discovery 的两位联合创始人。
We're here with Josh and Matt, two of the co founders of Chai Discovery.
Chai 这家公司是用人工智能来设计分子的。
Chai is engineering molecules with AI.
所以它有点像生物领域的基座模型实验室。
So it's something of a foundation model lab for biology.
听起来是个大主意，不过让我们
Sounds like a big idea, but let
我们就从这儿开始吧。
me just start there.
核心思路是什么？
What is the big idea?
感谢邀请我们来到节目。
Thanks for having us on the show.
我们在 Chai 正在努力做的一件令人兴奋的事，就是让药物研发过程看起来更像工程实践。
One of the exciting things we're trying to do at Chai is to make the drug discovery process look a little bit more like engineering.
我们看到了像语言模型在代码生成方面的应用，效果确实很好，因为代码结构相对简单，能很直观地表达你的意图。
And we've seen all the work happening with the LMs for cogeneration, for instance, right, and that works really well because code's a very simple obstruction, right, of like to kind of get across what you want to do.
但现在的生物学可不是这样。
Biology doesn't look like that today.
它更多是靠不断的试错。
It's a lot of like trial and error.
回想一下现代生物技术刚起步的时候，如今很多药物其实都是偶然发现的。
I think if we go back to the early days of just like modern biotech, right, a lot of the medicines that we have today were kind of discovered quite randomly.
这有点偶然，但我们想做的，是让这个过程更工业化一点。我们希望能开发出必要的工具，建立抽象层——就像现代工程里那样——这样我们就能快速迭代，把这些方法引入生物学和药物研发中。
It was a bit serendipitous, and what we're trying to do is allow us to industrialize that process a little bit more and try to come up with the tools that we'd need in order to have abstraction layers, the same way we have that encoded in modern engineering, allows us to iterate really quickly and try to bring that, into biology and into drug discovery.
那我顺着这个问题问下去。
So let me ask you a question on that then.
你听听我这么理解合不合理。
So there's and tell me if this is a reasonable way to frame it.
现在几乎存在一个界限：一边是可以靠工程手段解决的，另一边则必须在现实世界里做实验验证。
There's almost this boundary between that which can be engineered and that which needs to be tested in the real world.
而且感觉这个边界一直在移动。
And it feels like that boundary has been moving over time.
比如药物开发过程中可以工程化的部分，而不是靠试错的部分，似乎在不断增加。
Like the proportion of the drug development process that can be engineered as opposed to trial and errored seems to be increasing.
这么看合理吗？
Is that a reasonable way to think about it?
如果合理的话，你能说说具体是哪些发展让这条边界不断移动的吗？
If so, can you talk about like what specific developments have pushed that boundary over time?
我觉得这也是一种理解方式。
I think that's one way to think about it.
我们通常这么看：实验室是验证你做的事情是否正确的重要环节。
The way that we often, look at this is like the lab is an important part to verify that what you're doing is correct.
其实在AI领域，验证也是个很大的主题。
And actually verification is a very big theme in AI as well.
如果你能评估模型有效，能验证它，那你就可以开始逐步优化，并取得进展。
If you can evaluate that your model works, you can verify it, then you can start to hill climb that and you can make progress on it.
所以我觉得实验室是其中非常重要的一环。
So I think the lab is a very important part of this.
问题在于，我们怎么把药物发现变成更像药物设计的过程。
And the question is how do we take drug discovery and make it look a lot more like drug design?
我们之所以把这个领域叫做药物发现，是因为我们经常是在大海里捞针。
So one of the reasons why we call this field drug discovery is we're often looking for a needle in a haystack.
我们会筛选上百万、上十亿的分子，试图找到一个有效的。
We'll screen millions, billions of molecules, try to find one that works.
但如果我们能反过来，输入你想要的理想分子状态，然后让模型把它生成出来，那会非常强大。
And if we can instead put in what is like the dream state of the molecule you want, and then the model can materialize that, that's going be really powerful.
所以这甚至不只是减少实验室测试的问题。
So So it's not even a question of like reducing lab testing.
我是说，也许减少测试是结果之一。
I mean, maybe that happens as a result.
实际上我可能会站在相反的角度来谈，也许我们反而会做更多的实验室测试，因为投资回报率会提高，就像现在软件工程师因为生产力提高而需求更大一样。
I actually might even take the opposite side of the coin and we can talk about that where maybe we'd actually do even more lab testing because the ROI will increase the same way there's more software engineers, there's or more demand for software engineers now that they become more productive.
实验室的需求可能会更大。
There might be more demand for the lab.
但我觉得关键是如何改变这里的范式，让它更偏向设计导向。
But I think the key part is how do we just change the paradigm here and how do we make it more design oriented?
我们能聊聊现在最先进的技术是什么样的吗？
Can we talk about what is state of the art today?
然后也许我们可以回顾一下过去。
And then maybe let's take a little trip down memory lane.
五年前、三年前、一年前，有哪些重大突破？近年来最先进的技术是如何变化的？
Five years ago, three years ago, one year ago, like what have been some of the major breakthroughs and like how has state of the art changed over the recent years?
是的，我觉得这个领域真的发展了很多，尤其是在过去十年里。
Yeah, I think the field has like, it's really evolved, especially like over the last decade.
其实你可以回顾一下，有一个每两年举办一次的蛋白质折叠竞赛。
So it really wasn't until like, you can actually look back, there's like this biannual protein folding competition.
通常每两年，很多学术团队会参加这个蛋白质折叠竞赛。
So every two years, usually, a bunch of academic groups who compete in this protein folding competition.
情况是这样的，就是会拿出一些大家以前从没见过的蛋白质靶点。
What would happen is, like, you kind of hold out these protein targets that no one's ever seen before.
这些靶点都不会公开上传到数据库里。
They never get deposited publicly.
然后各个团队就展开竞争，看谁能预测得最准。
And then all these groups compete so you can predict the proteins the best.
说实话，直到2018年，我们才开始看到性能有了质的飞跃，然后在2020年AlphaFold 2出现时，更是达到了一个新的高度。
And, like, really, it wasn't until 2018 where you started to see this, like, big step change in in performance and then finally, again, in, like, 2020 with AlphaFold two.
不过，说真的，深度学习出现之后，才真正推动了整个领域的发展。
But, really, like, it was kind of like the advent of deep learning that really, like, kicked off this whole field.
刚开始的时候，我们做的事情，让我想起了在导师实验室里的日子。
So at first, what you do, these are kind of like, bring me back to my original days in my adviser's lab.
我们当时参与开发的是最早一批利用深度学习实现这一目标的技术之一。
So we worked on, like, one of the first systems to do this with deep learning.
我们的思路是预测蛋白质中氨基酸之间的距离，然后把这些带有噪声、看起来不太完整的距离数据输入到一个生成模型里，最后由它来构建出蛋白质结构。
What we would try to do is predict, like, the distances between amino acids and a protein, then and we'd have some render, which then took in, like, these kind of noisy and complete looking distances and then emitted some protein from that.
非常了不起的是，即使信息相对有限，你也能构建一个机器学习系统，真正生成看起来像蛋白质的东西，而且大部分情况下是正确的。
And what's pretty remarkable is that with relatively minimal information, like, you could build a machine learning system that could actually produce something that really looks like a protein and, for the most part, was was correct.
不过，仍然存在很大的差距。
Like, there are still pretty big gaps.
它还没有达到我们今天所拥有的那种分辨率。
It didn't quite have the resolution of, like, what what we have today.
就像早期的图像生成模型一样，有点颗粒感，有点模糊，而现在呢，简直让人惊叹。
Almost like with the early image generation models where it was, like, a little bit grainy, a little bit blurry, and now it's like, oh my gosh.
就像，几秒钟内就能生成超高清晰度的图像。
Like, this is this is crazy high definition images, like, in seconds.
所以我们差不多看到了同样的演变过程。
So we kinda, like, saw that same evolution happen over time.
这一切始于蛋白质折叠。
So it started with protein folding.
然后，我想，一旦这个变得可行，人们就想解决其他一些子问题。
And then, I guess, once that became more realizable, there were kind of these other subproblems that people wanted to solve.
所以现在，给定一个蛋白质结构，我能不能设计出一段序列，让它折叠成那个样子？
So now given a protein structure, can I design, say, like a sequence which might fold to that?
这越来越接近药物设计的问题了。
And that's getting closer and closer to, like, the drug design problem.
所以你现在得想想，怎么开始设计那些能匹配特定形状或具备特定功能的蛋白质？
So you're you're kind of, like, now thinking, okay, how do I start engineering proteins that match a certain shape or might perform a specific function?
大家差不多是在2021到2022年左右独立研究这个问题的。
So people kind of studied that problem independently, and that was kind of like 2021, 2022 time.
然后这些想法就开始慢慢融合在一起了。
And then kind of these ideas began to merge together.
这真的就像是扩散模型出现的时候，我们开始能够同时生成蛋白质的结构和序列了。
It was really like the advent of diffusion models where we started to be able to like, okay, I can now generate a protein structure in a sequence kind of simultaneously.
而且我们可以让所谓的提示词变得越来越真实。
And I can start making these, like, what we would call prompts more and more realistic.
所以我现在可以给模型设定一个具有特定形状的目标，然后说嘿，我想在这里结合。
So I can now prompt these models on a target that might have a certain shape and I can say, Hey, I wanna bind over here.
就像是在问题上加上更多现实世界的约束条件。
Kind of like add more real world constraints on the problem.
你们在2024年看到了什么，让你们觉得那是创办公司的最佳时机？
What did you guys see in 2024 that made you think that was the right time to start the company?
我们进行了很多讨论。
There were a lot of discussions that went into it.
我记得其中一次早期的讨论，实际上是在马特家里，当时马特给我看了一些关于抗体抗原结构预测的结果。
I remember one of these, early discussions actually back in Matt's house when we were, you know, Matt was showing me some of these results on like antibody antigen, like structure prediction.
马特当时在谈论蛋白质折叠，但长期以来人们认为抗体的蛋白质折叠问题太难了。
Matt was just talking about protein folding, but for a long time people thought that protein folding for antibodies was just like too hard of a problem.
人们觉得，比如在蛋白质数据库中，抗体的数据不够多，不足以解决这个问题，因此大家认为抗体设计遥不可及。
People were like there was not enough data in antibodies like in the protein databank for instance to like solve this problem, and people as a result thought that like antibody design was going be out of reach.
早期人们用深度学习做蛋白质设计的工作，很多都不是针对抗体的，而是针对一类叫做迷你蛋白的蛋白质，这些迷你蛋白本身确实很有趣，但并不是大多数制药行业关注的重点。
A lot of the work that people even did on protein design in the early days with deep learning, it wasn't antibodies, it was these different class of proteins called mini proteins, which are actually really interesting in their own right, but they're not what most of the drug industry is looking at.
所以终极目标就是能否在计算机上设计抗体蛋白，尤其是那些具有治疗功能的抗体。
So the holy grail was whether we could design antibody proteins on the computer, especially ones that had all that therapeutic function.
我们的想法是，要是你连抗体长什么样都预测不出来，那还怎么去设计呢？
And our thinking was that if you couldn't predict what an antibody looks like, how are you ever going to design one?
一直以来，大家都觉得生物学挺吓人的。
Traditionally, people have thought like, biology is scary.
就是，要学的东西太多了。
Like, there's so much to know.
我对生物学特别发怵。
I'm terrified of biology.
说实话，我也是。
I am as well, honestly.
我本来根本不是学生物的。
My background was never biology.
我学的是纯数学，博士读的是理论计算机科学。
I studied pure math and started my PhD in theoretical computer science.
到了第三年之后，我才转去做深度学习、蛋白质结构预测这些方向。
And it was only after my third year that I ended up switching into deep learning, protein, structure prediction, all of this stuff.
所以这对我来说完全是个全新的领域。
So it was totally new field to me.
感觉简直不可思议。
Seemed insane.
但归根结底，它其实比大家想象的要简单得多，而且各个问题之间的联系也更紧密。
But at the end of the day, it's much simpler and the problems are much more interconnected than one might think.
所以我觉得，大家刚进入这个领域时，心里都会犯嘀咕：抗体到底是啥玩意儿？
So I think people, when they start going into the field, they're like, oh man, what's an antibody?
迷你蛋白是什么？
What's a mini protein?
这些其实都是氨基酸序列。
These are all just sequences of amino acids.
说到底，它们只是给模型的不同类型的提示词。
At the end of the day, these are just different types of prompts for the model.
就像你如果把一道数学题输入给 ChatGPT，它既能解答你的数学题，又能帮你搞定英语作业一样。
But, like, in the same way where you might, like, have a math problem that goes into ChadGPT, ChadGPT can both answer your math math problem and, like, help you with your English homework.
所以，真的，我们的模型也是同样的道理。
So, like, really, we we have the same type of thing going on with our models.
就像，我们只是有某种方式来表示这些氨基酸序列，然后我们有办法设计、预测它们。
Like, we just have some way of representing these sequences of amino acids, then we have a way of designing, predicting those as well.
而且我觉得从这个角度来看，事情就变得清晰多了。
And I think in that lens, things become a lot more clear.
Matt，你刚才提到了一点你的背景。
And Matt, you mentioned your background a little bit.
Josh，聊聊你的背景吧，然后更广泛地说，要完成这件事，需要很多不同的学科汇聚在一起。
Josh, talk a bit about your background and then more broadly, in order to pull this off, you have a bunch of different disciplines that kind of come together.
所以你能不能谈谈你的背景，以及团队其他核心成员的背景，还有这些是怎么结合起来的？
So can you just talk a bit about like your background and that of some of the other core members of the team and how these things all fit together?
是的，我从小就对AI和生物学很兴奋，所以我想我就像Matt一样，不是从理论计算机科学开始，然后进入那个领域，但我上了一个有干细胞实验室的高中，所以我小时候就一直对生物学很兴奋，而且我从小就是程序员。
Yeah, I've been excited about AI and biology, since I was kid, so I guess I'm like, Matt, didn't start with theoretical CS and get into that way, but I went to a high school with a stem cell lab, so I was just always excited about biology as a kid and I grew up as a programmer.
我真正开始我的职业生涯是在OpenAI，所以我是那里的早期团队成员。
I really started my career at OpenAI so it was on the early team there.
那时候它还是个非营利机构，所以在那儿工作挺不错的。
Was a nonprofit back then, so it was a pretty good time to be there.
我们做了 GPT-1、GPT-2，还有缩放定律。当时我就在想，既然模型能学会说英语、德语，为什么不能学会说 DNA 和蛋白质呢？
We did GPT-one, GPT-two, scaling laws and it was the question was like if the models can learn to speak English, German, why can't they learn to speak DNA and protein?
从那以后，这也就成了我的研究方向。
And that was kind of my research agenda since then.
我觉得这个时间点，大概也是 Matt 进入这个领域的时候。
I think that sort of intersects with around the time like Matt got into the field as well.
而且我觉得那段时间也挺关键的，对吧？
And it was a I think it was a pretty important time as well, right?
因为你看我们现在引入的那些方法，其实边缘上已经发生了不少变化，你懂的。
Because if you look at the kind of methods that we are bringing in, like there's been a lot of these, you know, changes on the edges, if you will, right?
Matt之前也简单提过这个领域的发展历程。
And, you know, Matt talked a bit about the history of what's happened in the field here.
但核心问题还是在于，我们怎么找到合适的深度学习架构、计算配置和模型结构来实现这一目标呢？
But, you know, it's all about like how do we find like the right deep learning architectures with the right compute configuration and the right model architectures to make this happen?
该把这项技术用在哪些合适的任务上呢？
What are the right tasks to apply it to?
我们也在讨论，当初是怎么判断2024年就是成立公司的最佳时机的。
We're talking about how we even knew that like, you know, 2024 is the right time to start the company.
就像Matt说的，这些都是不同类型的氨基酸序列，当时大家都觉得抗体这类问题太难了。
As Matt was saying, you know, these are all different like kinds of amino acid sequences and people thought that antibody class of problem was going to be too hard.
然后我们开始看到一些初步的迹象，表明这其实开始奏效了。
And we started to see the first signs of life that actually this was starting to work.
我觉得很大程度上得益于我们引入的一些新架构，当时主要就是扩散模型。
I think a lot of it fueled by some of the new architectures we're bringing to the problems, where I think it was diffusion models back then.
第一次有人能生成看起来合理的蛋白质，就是扩散模型出现的时候。
The first time anyone was able to generate reasonable looking proteins was the advent of diffusion models.
那会儿真的挺疯狂的。
So it was pretty crazy.
之前也有不少生成建模的方法，但只有在数据量足够大的时候才能勉强工作。
They were a bunch of generative modeling approaches that would kind of work if you had a bunch of data.
所以人们把这些方法用在了图像上。
So people got these working for images.
一路上有很多技巧可以让效果越来越好。
There are, like, a bunch of tricks to make this better and better along the way.
我们在自己的领域里也借鉴了很多这些想法。
Like, we've definitely borrowed a lot of those ideas in our domain as well.
但真正关键的是，当扩散模型出现之后。
But it was really, like, once diffusion models came around
那扩散模型为什么有效，有没有什么直觉上的解释？
And is there is there an intuition for why diffusion models work?
对。
Yeah.
扩散不是一个一步到位的过程。
So diffusion is is not a one step process.
在那之前，主要的生成设计范式是变分自编码器。
So, like, I think kind of up until this point, like, the the main generative design paradigm was, like, called variational autoencoders.
在这种情况下，你的意思就是把输入分布给压缩掉。
And in that case, you're, like, saying, I wanna just, like, compress my input distribution.
比如说，你手头有一些蛋白质数据。
So, like, you you have some, like, proteins.
你想把这些数据变成那种模糊的高斯向量。
You wanna make these look like kind of, like, fuzzy Gaussian vectors.
这个任务真的非常难。
That task is just, like, really hard.
也许现在，如果我们真的拼命去试，我觉得我们就能搞定它。
And maybe today, if we tried, like, super hard, I think we'd crack it.
但那时候，我觉得技术还没发展到能解决我们问题的程度。
But at the time, like, I I think, like, it wasn't really as developed enough to to work for our problems.
事实证明效果很好的方法，就是给模型更多思考时间，并展示更多例子给它看。
What turned out working really well was just kinda giving the model more time to think and showing it, like, more examples.
比如，这里有一个看起来有点残缺的蛋白质结构。
Like, here's, like, a slightly broken looking protein.
你怎么让它变得更好？
How do you make it better?
你可以把那个蛋白质不断破坏，让它看起来越来越嘈杂、越来越残破，然后教模型一些简单的捷径。
You can kinda break that protein more and more and more, and you can make it look more and more noisy, more and more broken, and teach the model just quick little shortcuts.
好的。
Alright.
这就是我让它稍微变好一点的方法。
Here's how I make it slightly better.
然后你就可以一遍又一遍地继续问。
And you can just keep asking over and over again.
让它稍微变好一点。
Make it slightly better.
让它稍微变好一点。
Make it slightly better.
把问题拆解到这种程度，对生物学来说效果真的很好。
And kind of, like, breaking the problem down to that scale worked really, really well for for biology.
对。
Yeah.
这个“让它变得更好”让我想起我和女儿们喜欢玩的一个游戏，我们让Chai-2.2给我们一只独角兽，然后我们让它变得更强。
The make it slightly better reminds me of a game that I like to play with my daughters where we have Chai-2.2 give us a unicorn, and then we make it stronger.
我们就不停地告诉它要变得更强。
And we just keep telling it to make it stronger.
等我们玩完了，我们就有了世界上最强的独角兽。
By the time we're done, we have the strongest unicorn in the world.
所以差不多，对吧？
So about the same, right?
对。
Yeah.
就是这么简单。
That's, it's that easy.
对。
Yeah.
好的。
Okay.
也许不完全一样。
Maybe not the same.
对。
Yeah.
所以我觉得在这个领域，你需要组建一个类似四语种的小组，就像复仇者联盟一样，有化学家、生物学家、AI专家。
So I feel like in this domain, you need to assemble kind of like a quadrilingual group of people, like an Avengers squad of chemistry people, biologists, AI folks.
这确实是个挑战。
And so that's a challenge.
你们是怎么找到这些人，说服他们加入团队的呢？你们的明星成员是谁？
How have you guys gone about finding people, convincing people to join the team, who are your superstars?
我们在这方面非常务实，Chai，如果你看看创始团队，大部分是AI研究员。
So we've been really pragmatic about this, Chai- if you look at the founding team, it was mostly AI researchers.
他们要么研究过模型扩展，要么在这个领域让模型工作，但真的，随着每一代模型，我们为下一个里程碑需要的人的类型也在变化，或者说可能扩大了，对吧？
So people who had worked on either scaling models or getting them to work in this domain, But really with each generation of model, the kind of people we've needed for the next milestone has changed, or I'd say probably has expanded, right?
你看，到了Chai-two的时候，我们就开始引进一些世界上最顶尖的抗体工程师和科学家了。
So if you look at Chai-two, right, that's the point when we started to bring in some of the most incredible like, you know, antibody engineers and scientists in the world.
比如Andy Young，他其实是Andy团队里的科学家。我们招他的时候，大家都问我们是不是转型去做全链条药物研发了，因为大家都觉得，如果Andy都加入你们团队了，你们不这么做才怪呢。
One of the scientists from Andy, Andy Young, actually when we hired him, people asked us if we had pivoted into building a full stack drug pipeline because they're like, you'd be crazy not to do that if like Andy joined your team.
不过我觉得Andy最近几个月做的抗体项目，可能比他整个职业生涯做的都多，这真的很酷。
But I think Andy has enjoyed running more antibody campaigns in the past couple of months than he's probably running his whole career, which is very cool to see.
团队里还有像Nathan Rollins这样的人。
You have folks like Nathan Rollins on the team.
Nathan其实是在家上学的，很早就开始上大学了。
Like Nathan was actually homeschooled and then started college very early on.
他14岁就加入了David Baker的实验室，就是那个拿了诺贝尔奖的蛋白质设计实验室，18岁就开始读博士了，而且他有很多特别有创意的想法。
So, you joined like David Baker's lab who won the Nobel Prize for Protein Design, like when he was 14, started his PhD when he was 18, and has so many creative ideas.
随着模型越来越好，我们也需要组建产品团队，因为虽然研究人员能把模型做得非常强大，但你还是需要设计合适的产品界面，让模型真正发挥作用。
As the model started to get better, we needed to build up a product team because while the researchers might get the models to a point that they're very powerful, you need to build the right product interfaces so that the models are actually useful.
所以我们就开始招那些打造过我们熟知的最令人兴奋的产品的人。
And that's where we started to bring on people who have, built some of the most exciting products we know about today.
比如我们的联合创始人杰克之前在Stripe工作，穆内兹是Stripe代码贡献排名前十的工程师，还有尼尔，在安全变得重要之前，他经营着自己的网络安全公司，后来我们把这个部署给了大客户。
Like our co founder Jack worked at Stripe, Munez, who was one of the top 10 code contributors at Stripe, Neil, who ran his own cybersecurity company before security started to become very important as we deployed this to our big partners.
我们开始扩大规模，引进了很多擅长GPU优化的人才，可以说是专门来扩展我们的系统，确保在大规模运行时不会出问题。
We started to scale up, we've brought in people who've really done a lot of like the GPU hacking, if you will, in order to scale up our systems so that don't break when we're running them at scale.
前几天，我们收到了一封来自超大规模云服务商的邮件，或者说是Slack消息，说我们有一个集群出了问题。
We had an email from or a Slack message from one of our hyperscalers the other day where we had a cluster.
我觉得这也是个问题。
I think that's an issue too.
他们说，哎呀，GPU太热了。
And they're like, Oh, the GPUs got too hot.
你们可能跑得太多了。
I think you guys are running too many.
我们心想，这不就是我们的目的吗？
And we're like, Isn't that the point?
对吧？
Right?
那大概就是，好，至少我们在做我们的工作。
That's probably was like, Good, we're doing our job at least.
对，就是，你怎么说服那些人加入呢？
Yeah, it's like, How do you convince those people to join?
我觉得，还是那句话，很大程度上取决于结果和明确的需求。
I think, again, a lot of it comes back to the results and like a clear need.
我们以前在抗体设计模型里没有雇过抗体工程师。
We didn't hire antibody engineers before in an antibody design model.
那些人能做什么呢？
Like, what are those folks going to do?
我们甚至，我觉得，在开始那个趋势的时候还挺担心的，因为对于某些下一代格式，比如复杂的抗体，像多特异性抗体，它们甚至跟Chai-two都不兼容。
We were even, I think, worried when we started that trend because for some of the next generation formats, the complicated antibodies like multi specifics, like they didn't even work with Chai-two.
所以，当那些人来了之后，实际上花了几周时间，模型才能达到可以处理这些有趣案例的程度，但幸运的是，进展够快，最终把它带上了线。
So it actually took a couple of weeks when some of those people showed up before the models could work at a point that they could work on some of these interesting case studies, but fortunately the progress was fast enough to kind of bring that online.
所以，我觉得我们一直在不断调整那个团队，朝着下一个里程碑前进。
So, I think we're always like evolving that team and going for that next milestone.
结果我们的团队也变得非常精简。
We've got the team very small as a result, too.
这样一来，每个人多少都有点超负荷，我觉得，所以我们得学会分清轻重缓急。
So this way, everyone is a little bit like slightly over capacity, I think, which means we have to prioritize.
这逼着我们去处理那些真正最重要的事情。
It forces us to work on the things that really matter most.
我觉得在研究方面也一样，我们的创始工程师之一凯文·吴，他做出了第一个蛋白质扩散模型，应该是全世界第一个。
I think on the research side as well, one of the founding engineers, Kevin Wu, he had the first, I think it was the first protein diffusion model, like, ever.
这也说明了凯文的执行力有多强。
And that speaks to Kevin's speed of execution.
他真的是个了不起的工程师。
Like, he is a heck of an engineer.
而且我觉得，从第一天起，工程能力就一直很重要。
And, like, I think engineering has just always been, like, important since day one.
所以实际上，连我们的研究员也都是非常优秀的工程师，我们真的很在意打造高质量的代码库。
So, really, even our researchers are all excellent engineers, and we really care about building a high quality code base.
说到底，我们本质上是一家软件公司。
At the end of the day, we are technically a software company.
我们是AI研究员。
We're AI researchers.
我们是蛋白质设计师。
We're protein designers.
我们做很多事，但最终交付的是一些软件。
We're a lot of things, but our deliverable is some piece of software.
所以我们一直保持很高的标准，同时也在努力平衡，引进优秀的研究人才和真正能推动前沿的人。
So we've kept that, like, kept that bar really high while also trying to, like, you know, level that with great research talent and people that can actually, like, push the frontiers of what's possible.
你在这个领域好像经历了不少关键时刻。
And you've had a number of, it seems like moments in the field.
比如AlphaFold，我们知道了如何预测蛋白质折叠。
Like, AlphaFold was an we can figure out how a protein folds.
然后扩散模型，我们可以生成蛋白质了。
And then the Diffusion Models, like, we can generate proteins.
看起来你们最新的模型又带来了另一个突破。
And it seems like your latest models are a kind of another moment.
我们不仅能生成看起来像蛋白质的分子，它们还具备治疗功效。
We're not only generating molecules that look like proteins, but they also have therapeutic properties.
比如，它们能结合得非常紧密。
Like, they can bind really tightly.
它们的命中率非常高。
They have really high hit rates.
你能聊聊你们模型现在生成的分子质量怎么样，以及为了让模型具备这种能力，背后都做了哪些工作吗？
Can you talk sort of about the quality of the molecules that your models are producing now and everything that sort of went in to those models to make them able to do that?
对。如果我们看看最终生成的分子质量，这其实回到了我们创业初期提出的一个核心观点：我们非常希望专注于从头设计（de novo）这些分子。
Yeah, if we look at the quality of the molecules that's coming out, if it goes back to, you know, one of the theses when we started the company, that we really wanted to focus in on this like de novo generation of the molecules.
如果你回顾当时很多药物发现领域的AI研究，重点通常是如何在现有分子的基础上做一点优化。就像我们刚才提到的那样，但那种方式通常是‘实验室闭环’式的，也就是拿一些数据，试着通过这种方式来改进它。
If you look at what a lot of the drug discovery AI work was at the time, it was about how do I take a molecule and just make it a little bit better, right, which we just talked about in a sense, but it was doing it with a lab in the loop style, right, where I take some data, try to make it better that way.
而我们一开始问的问题是：我们能不能直接在电脑上完成所有这些工作呢？
And the question we started with is, well, can we actually just like do all that on the computer, right?
有没有可能，也许数据足够多，或者我们能收集到足够多的数据，这样就可以直接零样本生成一个具备很多这些性质的分子？
Is there a way that the maybe there would be enough data or we could collect enough data so that we could just zero shot a molecule that has a lot of these properties?
所以，要达到那个目标，我们首先需要做到的是设计出成功率非常高的分子。
So the first thing we needed to do to get there was to design molecules with really high success rates.
我们刚创办公司的时候，抗体设计的最先进水平大概只有0.1%的结合率，也就是说你设计的一千个分子里，只有一个能在实验室里真正结合。
When we started the company, the state of the art for antibody design was about like a 0.1% binding rate, so one in a thousand of the molecules you design would actually bind in the lab.
所以，首先这意味着你得筛选很多分子才能找到几个好的。
So, first of all, that means you have to screen a lot of molecules to find some good ones.
这也意味着，你从过程中得到的梯度其实也很弱，所以对于很多靶点，你根本得不到任何阳性结果。
It also means that, like, the gradient you get on your process is actually quite weak as well, so for many targets you won't get any hits.
对于那些你确实能拿到一些阳性结果的，也不够用来真正判断这些分子是不是具有药物那样的性质。
For the ones that, you know, you do get some hits, you won't have enough to actually see whether you're having, like, the drug like properties.
所以，我们集中精力思考的就是：怎么让这个过程变得更加准确？
So, we really focused in on how do we just make this process more accurate?
我们用Chai-two模型把成功率提高到了大约15%。
We got to with our Chai-two model about like a fifteen percent success rate.
所以现在如果你筛选1000个分子，能得到150个回来，那你就可以开始对分子的性质做一些有趣的统计了，对吧？
So, now if you screen 1,000 molecules, you're getting 150 back, now you can start to get some like interesting statistics on the properties of the molecules, right?
这让我们能够在此基础上迭代，在实验室里建立正确的评估，并且真正尝试去优化它。
And that allowed us to like iterate on that, build the right evaluations around that in the lab and really try to hill climb that as well.
我们现在已经达到了一个点，可以在引擎一开始就融入很多这些不同的性质。
And we're getting to a point now where we can actually bake in a lot of these different properties from the start into this engine.
然后也许你会想，这到底是怎么发生的，或者我们是怎么处理这个问题的？
And then maybe if you think about, you know, how does this happen or like, how are we approaching this as well?
我们为什么认为它会继续改进？
Like, why do we think it's gonna continue improving?
对。
Yeah.
说实话，我觉得乔希和我都对这个方法见效之快感到惊讶。
So like, honestly, I think Josh and I were both surprised at how quickly this worked.
当我们最初做预算的时候，我们想，也许20%的命中率，大概要三四年吧。
Like, when we were originally budgeting this, we're like, ah, maybe, like, 20% hit rate, like, three or four years.
我们当时真的就是冲着1%去的。
We were, like, really shooting for, like, a 1%.
我们觉得百分之一就已经很了不起了。
We thought one in a 100 would be amazing.
我们觉得这会是开创性的东西。
We're like, this is gonna be, you know, a groundbreaking thing.
我们有一套理念和方法，结果就是效果特别好。
We had a philosophy on, like, an an approach that we wanted to take, and it just, like, ended up working really well.
而且这个方法跟机器学习其他领域成功的做法非常相似。
And that approach is, like, very similar to what's worked in the rest of machine learning.
大家把生物学当成一个特殊问题或者特殊领域，但实际上，它跟自动驾驶语言模型的原理是一样的。
So people kinda treat biology as this, like, bespoke problem or, like, bespoke field, but, really, it's, like, the same principles as, like, self driving LMs.
原理都一样，但我们之前聊过怎么找到缩放定律。
Same all principles, but one of the things we've talked about before is how you manage to find scaling laws.
把一串文本变成token是一回事。
And it's one thing to tokenize a string of text.
但把生物学变成可处理的单元，又是另一回事了。
It's another thing to tokenize biology.
你能简单说说，在不透露核心机密的前提下，讲讲这个挑战以及你们是怎么解决的吗？
Can you say a couple words about, without giving away any of the magic, know, can you just say a couple words about that challenge and how you guys solve that?
好的。
Yeah.
我很喜欢Chai的一点是，我们是一家非常信奉‘苦涩教训’的公司。
So one thing that I really like about Chai is we're a very bitter lesson built company.
我们坚信要扩大数据、模型和算力的规模。
So we really believe in scaling data, scaling models, scaling compute.
要做到这一点，显然需要找到缩放定律。
In order to do that, obviously, need to identify scaling laws.
否则，你只是在浪费时间和资源。
Otherwise, you're just kind of wasting time and resources.
我觉得，在不透露太多的情况下，我们的一大指导原则就是简单。
And I think without without giving away too much, like, one of our big guiding principles is just simplicity.
所以，比如说你看一个模型，比如Chai-one。
So, like, when you look at a model, like, let's say, Chai-one.
然后，我觉得Chai-one里面有23个不同的子模块。
So, like, there are, like I think there are 23 distinct submodules in Chai-one.
当你想要对这个东西进行迭代的时候，就会变得很难，因为你得独立理解每一个子模块。
And, like, when you're trying to iterate on something like that, it gets really hard because you're like, I kinda I need to understand each of these sub modules independently.
你得了解它们所有的行为和动态，但这样做其实不太能扩展。
I need to understand all of their behaviors, their dynamics, and, like, that doesn't actually scale that well.
于是你可以开始想，好吧。
So, like, you can start to think, okay.
我怎么简化这个？
How do I simplify this?
我怎么识别什么才是真正重要的？
How do I identify, like, what's what's really important?
一旦你有了这些，整个研究过程和识别这些扩展方向就会变得简单很多。
And once you have that, like kind of the whole research process and like identifying these types of scaling directions becomes a lot simpler.
还有一点很有意思，我们借鉴了深度学习其他领域那些行之有效的经验。
One of the other things too, think it's interesting about this is we take a lot of these lessons from what's worked in the rest of the deep learning space.
但正如你指出的，数据本身是不一样的。
But as you're pointing out, like the data itself is different.
对。
Yeah.
Chai 的模型完全是从零开始构建的。
The models at Chai are completely built from scratch.
我们并不是像微调 GLM 那样，在某个蛋白质上做些调整，而是从头开始构建一切。
We're not like fine tuning GLM or something like that on some We protein build everything from the ground up.
我觉得很多公司在建设过程中，其实是坚持一种理念，不断围绕它迭代，并确立一些指导原则。
I think a lot of the company building process though is taking a philosophy and actually sticking with it and iterating on that and just having some guiding principles.
当你创办一家类似前沿研究型公司时，比如 Chai 几乎就像这种新型实验室一样，对吧？
When you're building like a blue sky research company, if you will, you know, Chai is almost like one of these neo labs, right?
我们正致力于解决一个宏大的 AI 难题，随着取得进展，这也会为客户带来新的机会。
Like we have this big AI problem we're going after where as we make progress on it, know, that opens up opportunity for our customers.
但如果你要从事如此开放性的工作，你需要一些原则来指导你。
But if you're going to work on something so open ended that way, you need some principles to guide you.
我认为我们在公司里贯彻这些原则方面做得非常好，一直在努力。
And I think we've done a very good job on like tracking those principles in the company, working on it.
所以很多马特提到的
So a lot of the things that Matt
你能分享一下吗？
Can you share them?
指导原则是什么？
What are the guiding principles?
我认为简单性是马特提到的原则之一。
So I think simplicity was one of the ones that Matt mentioned.
这就像那个苦涩的教训，比如扩展计算、数据和模型。
It's this like bitter lesson, pileness of like, you know, scaling compute and data and models.
要非常严谨。
It's being really rigorous.
在这个领域，这一点非常重要。
That's something that's so important in this space.
在生物学中，你很容易自欺欺人。
You can fool yourself so easily in biology.
比如，湿实验室的误差范围其实也很大，所以这跟你看代码生成之类的情况有点不同。
Like the error bars in the wet lab are actually quite large as well, so it's actually a little bit different than if you look at like cogeneration, for instance.
你看SWE bench，大概一年前，人们会说，哦，我得了16%，然后17%，然后18%。
If you look at SWE bench, people were like, Oh, this is maybe like a year ago, people were like, Oh, I got 16%, then 17%, then 18%.
我的意思是，在生物学中，如果你的实验误差是正负5%，那可能都差不多，所以这实际上意味着，对于模型带来的阶段性变化，门槛很高，但你也需要对自己诚实，判断是否真的在进步。所以你可以提出一个花哨的模型，在一两个新任务上看起来表现不错，但如果你真的想构建一个能推动领域发展的产品，展示它在更广泛的任务上有效是非常重要的。
I mean, in biology, if you're like plus or -five percent in your lab, like that might all be the same, so it actually just means the bar is really high in terms of the step changes that you want to see with the models but you also need to be really honest with yourself about whether you're making progress or So you could come up with some fancy model that looks like it works well in, like, one or two new tasks, but it's very important to show that that works more generally if you're actually trying to build a product that can, like, bring the field forward.
是的，我觉得这很有趣，因为生物学是那种本质上不太容易验证的领域之一。
Yeah, and that's, I think, pretty interesting because biology is one of those inherently not so verifiable domains.
而你们在向客户和我们这样的人展示进展方面做得很好，我们对生物学了解很少。
And you guys have been really good at sort of showing your progress to customers and to people like us, you know, very little about biology.
那么，我们能谈谈评估和模型进展中可验证的部分吗？
And so can we talk a little bit about the evals and the verifiable part of the model progress?
那你们怎么知道自己的模型在进步呢？
Like how do you guys know that your models are getting better?
嗯，首先我觉得这其实是一个比较容易验证的领域。
Well I would say first of all that I actually think this is one of the more verifiable domains.
它的结果其实非常客观。
It's actually a very objective readout.
比如说你看像代码生成这种任务，对吧？
If you look at something even like cogen, right?
像验证任务，比如我的代码能不能编译通过？
Like maybe as a verifiable task like did my code compile?
能不能通过这些单元测试？
Did it solve these unit tests?
但你怎么衡量代码的质量呢？
But how do think about the taste, right?
比如我是不是写了一些很烂、很难维护的代码？
Like did I write some really sloppy code that can't be maintained?
那会是什么样子呢？
Like what does that look like?
当我们在实验室设计一个分子时，我们其实可以非常具体地了解这些性质中的许多，对吧？
When we think about designing a molecule in the lab, we can actually be quite specific about many of these properties, right?
所以也许我们得到一个能结合靶点的分子，但我能制造它吗？
So maybe we get a molecule that binds the target but can I manufacture it, right?
那可能就是你所说的某种技术债务。
That might be your, you know, version of like some tech debt.
这是可以测量的。
You can measure that.
而且我认为这些评估实际上让这个领域更可验证。
And I think those evals again actually make this domain more verifiable.
也许验证它需要更长一点的时间，对吧？
Maybe it takes a little bit longer to validate it, right?
不像五秒钟就能得到结果并运行单元测试那样。
It's not like five seconds to like get a readout and run a unit test.
你可能得在实验室里花上几天、几周才能得到那个结果，但至少你能对自己诚实。
You might have to spend a couple days, a couple weeks in the lab to get that readout, but at least you can be honest with yourself.
对。
Yeah.
当我们谈论进展和模型有多好时，生物学中有一类靶点，正如你提到的，可以用传统的筛选方法来解决。
When we talk about progress and how good the models are, there's like a domain of targets in biology that you can, as you mentioned, sort of address with traditional screening methods.
这些方法耗时很长，非常缓慢且原始。
They take a long time, they're very slow and rudimentary.
还有一些靶点用现有方法根本无法解决。
And then there are targets that just aren't addressable with existing methods.
它们因为各种原因不可成药。
They're not druggable for whatever reason.
那么，在模型进展方面，在处理现有靶点和更快生成分子方面，我们处于什么位置？
And so where are we in terms of model progress, in terms of working on existing targets and generating molecules faster?
光谱的一端。
One end of the spectrum.
而另一头则是解锁全新的生物学机制、新靶点，以及以前没法成药的那些东西。
And then the other end of the spectrum is unlocking novel biology, new targets and things that we couldn't drug before.
这其实又回到了我们核心的建模理念。
This kind of goes back to kind of like our core modeling philosophy.
所以很多时候我们都会想，要是能有个第二十四模块就好了，那样我们就能真正解锁那个新靶点。
So there have actually been plenty of times where like, Man, if we had a twenty fourth module, like, we can actually, like, unlock that new target.
然后我们会问自己，这真的是我们要长期维持的方向吗？
And we're like, is that really something that we wanna maintain long term?
这是渐进式的改进，还是说实际上是一种累积性的提升？
Is this incremental, or is this, like, actually a compounding improvement?
这真的能帮我们更好地泛化到更广泛的那类靶点吗？就是那些我们目前很难攻克的靶点。
Will this actually, like, help us generalize to the broader class of, like, these this whole class of targets that we we really can't hit?
所以我们的理念是这样的，好吧。
And so, like, our our philosophy has been, like, alright.
我们就继续专注于确定你的扩展规律。
Let's just, like, continue to focus, like, identify your scaling laws.
就是说，总有一个点，如果模型能把损失降得更低，它就必须理解它正在处理的目标的一些非常本质的东西。
Like, there there comes a point where, like, if the model is able to push loss down even further, it has to understand, like, something very intrinsic about the target that it's operating on.
举个例子，前几天我和保罗在聊，也许我们看待蛋白质上某些糖基化位点的方式，我们可能会想，哦，也许我们应该用不同的方式来表示它们，或者类似的想法。
So one example we were talking about the other day, Paul and I, maybe the way that we're looking at certain glycosylation sites on proteins, we're like, Oh, we might want to represent them differently or something like this.
然后我们就想，即使我们不表示它们，也有一些序列基序会告诉模型，这里应该有一个糖基化位点。
And we're like, Well, even if we didn't represent them, like, there are certain sequence motifs that will tell the model, like, there should be a glycosylation site here.
为了进一步降低损失，模型就应该学会这一点。
In order to drive, like, loss down further, the model should just, like, have to learn that.
所以，目标上有所有这些隐藏的特征，如果你真的相信缩放定律，你就相信模型会达到那个水平。
So there are, like, all these hidden features of targets where, like, if you really believe in scaling laws, you believe the models will get there.
这类目标，就应该随着更好的模型而解锁。
Like, these types of targets, like, should just unlock with better models.
当然，你仍然需要非常认真地对待这件事，仍然需要所有适当的验证。
And, of course, like, you still have to take this very seriously and, like, you still need all the proper validation.
你需要真正挑战自己，确保这真的有效。
You need to, like, really challenge yourself and like make sure that this is truly working.
不过我觉得我们一直以来的思路是，随着模型越来越强，我们应该能攻克更多目标。
But I think our approach has always been like, you know, with better models, like we should be able to unlock a
攻克这些目标。
lot of these targets.
另一个有意思的点在于，我们之前聊过这个领域的跨学科性质。
One of the other interesting things is if we look at we talked about like the interdisciplinary nature of this.
如果我们看看 Chai 内部不同的团队，你会发现每个团队对‘难解目标’的定义其实完全不一样。
If we look at the different teams at Chai, what people will call a hard target is actually different in literally every team.
所以在科学团队那边，可能会遇到像那些无法成药的GPCR靶点这样的问题。
So on the science team it might be, you know, like an undruggable GPCR target.
目前还没有人能够以功能性的方式调控这类靶点。
No one's gotten something that has like, you know, modulated that in a functional way.
而在机器学习研究团队看来，问题可能是模型根本没法把这个结构折叠起来。
On the ML research team it'll be something like, Oh, there's like the model just can't fold this thing up.
它完全不知道这东西长什么样。
It doesn't know what it looks like.
展开剩余字幕（还有 298 条）
然后在产品团队，可能看起来像是，哦，我有很多这样的修饰，比如糖基化，而且它是一个膜蛋白。
And then on the product team, it might look like, Oh, I've got all these like, you know, modifications like my glycosylations and it's a membrane protein.
我怎么把它呈现给用户呢？
How do I represent that to the user?
实际上，每个团队对难点的定义都不同，我觉得这反而是个优点，不是缺点。
And actually the fact that it's different for each of these groups, I think is a feature rather than a bug.
这意味着，如果我们想在这里取得广泛的进展，每个人都在以不同的方式并行推进，这样我们就能一直平稳地取得进步。
And it means that if we want to make like broad progress over here, everyone is kind of pushing in parallel in these different ways and that means that there is very smooth progress that we can make all the time.
而且在Chai，通常不会只有一个瓶颈。
And there's usually not one bottleneck at Chai.
不是那种，哦，如果我们多一个模块就好了，或者如果我们把它推到产品里就能解锁什么。
It's not like, Oh, if we only had that one extra module, things would work, or if we only, you know, like try to push this into the product in some way we could unlock it.
我们正在努力构建一个统一的解决方案，因为归根结底，公司的目标是打造一个分子计算机辅助设计套件。
We're trying to build this unified solution because at the end the day, the goal of the company is to build a computer aided design suite for molecules.
不是要做一两个分子，也不是要推出五条有趣的疗法管线。
It's not to make one or two molecules, it's not to get a pipeline of like five interesting therapies that we bring to market.
而是要改变药物发现的方式。
It's to change the way that medicines are discovered.
如果我们想做到这一点，就必须攻克所有难题，不管你怎么定义难题。
And if we're going to do that, we need to work on all the hard targets regardless of how you define hard.
对。
Yeah.
再详细说说这个分子计算机辅助设计套件的想法吧。
Say more about this idea of computer aided design suite for molecules.
那是什么意思？
What does that mean?
核心就是回到那个观点，让生物学更像一门工程学科。
So at its core, it goes back to this point about making biology look more like an engineering discipline.
所以我们不是去大库里捞东西，也不是做大量的试错。
So we're not going and fishing something out of a large library or doing a ton of trial and error.
你希望一开始就能明确设计分子的原则，然后有一个引擎能把它实现成我们将在实验室里制造的具体分子。
You want to be able to specify upfront the principles that go into designing your molecule then and have an engine that can actually realize that into some molecule that we're going go and create in the lab.
而且你看，你可能还是需要在实验室和模型上做一些迭代，因为也许你的假设是错的。
And look, you still might do some iteration on the lab and on your model because maybe your hypothesis was wrong.
但我们真正想加速的，还是那个计算机辅助设计套件，这样你就能很快从想法变成可测试的假设。
But what we want to speed up is actually, again, have that computer aided design suite so that you can go from idea to testable hypothesis very quickly.
如果现在这个循环大概需要九个月，我不知道，才能发现你的分子，而如果只需要九周或者九天，你知道，每提升一个数量级，你就能处理的想法数量就会大幅增加。
And if that loop right now takes something like nine months, I don't know, to go and like discover your molecule versus if it takes nine weeks or it takes nine days, you know, each order of magnitude just scales in a very big way the number of ideas you can really sort through.
我觉得这最终就是整个领域如何走向更好的药物的方式。
I think that's ultimately how the field is going to converge on better medicines.
这真的回到了人们有时会讨论的，我们关心的是速度还是目标的难度，Sonali，你也提到了这一点。
It really comes back to like people sometimes talk about do we care and you kind of noted on it, Sonali, do we care about like speed or do we care about like the difficulty of the targets?
在某种程度上，这两者也会汇聚在一起，因为一个困难的目标，如果我们能更快地迭代假设，那么也许就更容易攻克它。
At some point they converge in this way as well, because a hard target, if we can like iterate through hypotheses a lot faster, then maybe it'll be easier to crack it.
所以如果我们能暂时抛开现在的现实，展望足够远的未来，不是从现在往前想，而是从未来往回看，2035年、2040年、2100年，随你选。
So if we can maybe detach ourselves from the present reality and go far enough into the future that we're not thinking present forward, we're actually thinking future back, 2035, 2040, 2100, whatever you want.
这个行业会是什么样子？
What does the industry look like?
比如说，假设分子计算机辅助设计套件已经成为标准。
Like, let's imagine that computer aided design suite for molecules has become a standard.
假设很多创新已经流向下游，进入湿实验室环节。
Let's imagine a lot of innovation has flowed downstream into some of the wet lab parts of the process.
你能描绘一下十年后这个行业可能是什么样子吗？
Like, can you paint the picture of what the industry might look like ten years from now?
医生。
Doctor.
是的，我觉得那会是一个非常激动人心的时代，我们可以从几个角度来看。
Yeah, think it's going to be a really exciting time, and we can look at this on a couple of angles.
首先，我们开发的药物质量有望提高。
So first of all, the quality of the medicines that we develop, will hopefully go up.
今天有很多分子进入临床，其实很难发现，你找到一个差不多80%的分子，可能就想着推进它，赶上时间表，它可能会对某些患者有益，但一年后可能就被别人超越了，这真的不是行业资源最有效的利用方式。
There's a lot of molecules we put into the clinic today that there is really hard to discover molecule, you find something that's like 80% of the way there, maybe I'm just going advance it anyways, like hit my timelines, it's probably going to benefit some patients, but then you get beat, you know, a year later by someone else and it's really just not the most efficient spend of resources in the industry.
还有一些疾病，今天实在太难攻克了。
You've got the kind of diseases that are just too hard to go after today.
人们一直在尝试治疗阿尔茨海默症，但遗憾的是，进展远不如我们所愿。
People have been trying to drug Alzheimer's forever and unfortunately, you know, haven't made as much progress as we'd like.
还有一些疾病，目前去研究它们根本不划算。
You have things that just aren't economical to go after today.
想想个性化药物、罕见病，这些患者群体可能更小，但如果我们能更快地迭代这些想法，更快地推出产品，并且成本更低，那么这些领域也可能变得触手可及。
Think about like personalized medicines, rare diseases, things where maybe the patient population is going to be smaller, but again, if we can iterate through those ideas faster, if we can launch something faster, if we can do it in a less expensive way, then those probably come into reach as well.
所以我觉得我们可以发力的方向太多了。
So I think there are just so many different axes that that we're able to push on.
这意味着未来真的非常光明。
And I think that means that the future is really bright.
以前总是说，你要么想做
It used to be like, you either want to
同类首创，要么做同类最优。
be first in class or best in class.
对。
Yeah.
现在呢，你反而想成为最后一个，因为你就是想成为最终的答案。
And now it's like, you want to be last in class because like, you actually just want to like be the final answer.
所以我觉得你会看到很多，就是在设计药物的时候，会更有针对性。
So I think a lot of what you'll see is just way more intentionality in the types of drugs that you're designing.
这种药会非常针对特定的疾病。
This drug will be super specific to the disease of interest.
它不会有现在很多药物都有的某些相互作用，某些负面相互作用。
It won't have certain interactions, certain negative interactions that a lot of drugs today do.
其实这些东西，大部分都可以用计算来建模，也许现在还不能完全做到，但肯定有办法达到那个目标。
A lot of this stuff is actually, you're able to model most of this, computationally, like maybe not today, but there's definitely a path towards getting there.
我觉得这对我来说是最令人兴奋的事情之一。
I think that's like one of the most exciting things for me.
太酷了。
Very cool.
我们能聊聊你们做的商业模式决策吗？
Can we talk about a business model decision that you guys made?
因为我觉得很多时候人们会把Chai和Isomorphic归为一类。
Cause I think a lot of times folks think about Chai and isomorphic in the same neighborhood.
Isomorphic是在开发药物。
Isomorphic is developing drugs.
而你们是在帮助现有的制药行业更高效、更好、更快、更便宜地开发药物。
You guys are enabling the existing industry to develop drugs more efficiently, better, faster, cheaper than they had before.
你们为什么决定走这条路，而不是像Isomorphic那样呢？
Why did you decide to go down that path versus, you know, the isomorphic path?
是的，首先我觉得这两条路都很好，都能创造巨大的价值。
Yeah, first of I think both of these paths are great and they can create tremendous value.
我们一直对为行业建设基础设施感到非常兴奋。
We've always been really excited about building infrastructure for the industry.
当我们几年前在Matt家看到那些结果时，我们就押注未来大多数药物都会以这种方式被发现。
Our bet when we started to go back to those results in Matt's house, you know, a few years ago was that this is how most future drugs are going to be discovered.
如果真是这样，那就需要有人去建设这样的基础设施，并让它成为现实。
And if that's the case, somebody needs to go and like build that infrastructure and make it happen.
我觉得这也有部分原因是我们的创始团队，包括我和Matt，以前都在一些公司工作过，在那里我们搭建了完整的药物研发管线，对吧，我们当时也在建AI模型，也在把那些药物推进到临床。
I think part of this too is a lot of our founding team, like including Matt and I, had worked in companies before where we had built these full stack drug pipelines, right, and we were building AI models, we were putting those drugs in the clinic.
我们同样也深受'惨痛教训'的影响，我们的想法是，随着模型越来越好，我们宁愿把更多的钱花在改进模型上，而不是把资源分散到临床试验那些事情上。
Again, we're pretty bitter lesson pilled as well, and our thinking was as the models get better, we want to be spending more of the money making better models as opposed to diverting those resources into clinical trials and things like that.
我们商业模式的一个有趣之处在于，模型变得越好，实际上就让我们更有资格继续加大投入，对吧？
And one of the interesting things about our business models, as the models get better, it actually wins us the right to continue investing more in them, right?
而且，你们看，我们现在跟这些制药公司的合作已经能带来回报了，我们很多人都可以继续投入进去。
And you have, you know, partnerships with these pharma companies that are paying off today and a lot of us can invest in it.
所以这样来看，这个业务的可扩展性要强得多。
So it's a much more, you know, scalable business in that way.
而且我觉得我们最终能给世界带来的影响力也会大得多。
And I just think about the ultimate impact that we can create for the world is a lot larger.
我们一直都想和整个生态系统广泛合作。
We've always wanted to just partner broadly with the ecosystem.
这又回到了刚才说的那个观点，就是在生物学领域不能自欺欺人。
It goes back to this point about fooling yourself in biology.
如果你只研究少数几个药物靶点，你可能会做出最精妙的分子，为世界创造大量价值，但你可能会只见树木不见森林，因为你的模型可能无法推广到那一年人们会制造的其他500个分子。
If you work on a small number of drug targets, you might come up with the most exquisite molecules, creating a ton of value for the world by doing that, but you might miss the forest for the trees because maybe your model doesn't generalize to the other 500 molecules that people are going to make that year.
如果你去和别人合作，你就不能自欺欺人。
And if you go and partner with people, you just can't fool yourself.
比如，看看我们的合作伙伴，礼来、诺华、Argenx、辉瑞，这些公司可不是那种会把这些事情当儿戏的。
Like, if you look at our partners, Eli Lilly, Novartis, Argenx, Pfizer, like these are not companies that are, you know, they take this stuff for granted.
你必须真正在这些合作中拿出成果，他们才会认真对待你。
Like, you have to really deliver on these partnerships for them to take you seriously.
这意味着不能只是有时候有效。
And that means it can't just work like some of the time.
比如我们在Chai发布模型时，它们必须真正有效。
Like when we ship models at Chai, they really have to work.
它们必须为我们的合作伙伴带来价值。
They have to deliver value to our partners.
而不是像，哦，我们做了个分子，它不完全有效，我们让化学家稍微清理一下就行了。
And it's not like, Oh, we made some molecule, it doesn't fully work, we're gonna have our chemists like clean it up a little bit.
所以这几乎是一个更难做成的生意。
So it's almost a harder business to pull off.
我觉得这也是为什么很少有人尝试过这个方向的原因之一。
I think that's one of the reasons why you haven't seen it pursued many times.
如果你做的是药物管线，那可能后面还有机会去修正一些问题。
If you work on a drug pipeline, again, like, there might be ways to fix things up later.
当然，临床结果最终会证明一切，但如果你是基于合作模式的，那你就必须对自己的模型非常严格。
There will be the proof of like what happens in the clinic, of course, but when you have this partnering based model, you have to be really rigorous about your models.
你的模型必须非常可靠，否则那些合作伙伴不会轻易找上门来。
Your models have to work really well because otherwise those partners are not going to come easily.
所以，这确实在很多方面让我们的工作更难了，但我觉得如果能做成，这也是更有回报的一条路。
So, it's made our life harder, I think, in many ways, but I think it's also the more rewarding path if we can get it to work.
那你学到了什么？
Well, and what have you learned?
我是说，你算是走出了实验室，进入了现实世界，为真正的制药公司创造实际价值。
I mean, you know, you're out of the lab, so to speak, and in the real world, you know, delivering real value for actual pharma companies.
当你开始和这些合作伙伴合作时，你学到了什么？有没有什么意外的地方，比如他们的需求和你的预期有什么不同，或者他们在这方面的成熟度和你的预期不太一样？
What have you learned as you started to work with these partners in terms of any surprises in terms of how their needs might have differed from what you expected or how their level of sophistication around this might be different than what you expected?
和他们合作的过程中，有没有什么意外或者收获？
Any surprises or any learnings from working with these partners?
对，当我们开始这些合作的时候，很多人跟我们说，制药行业不太会用AI，不是那种技术前沿的行业之类的。
Yeah, so when we went into these partnerships, I think a lot of people told us that like, pharma doesn't know how to use AI, these are not, it's not like a tech forward industry and things like that.
说实话，我们的经历并不是这样的。
And to be honest, that hasn't really been our experience.
我觉得这些合作伙伴都是非常严谨的，他们的客户也很严谨，对吧？
I think that these are, again, they're very rigorous partners, very rigorous customers, right?
在部署这些技术之前，他们会逐一验证我们的每一个说法，对吧？
And they're going to test every one of our claims, right, before they start to deploy these things.
但一旦他们看到数据，就会全力投入，对吧？
But when they see the data, they go all in, right?
因为制药行业本身就是一个创新的行业。
Because pharma is an innovation industry.
我觉得光是想想制药行业的经济模式就很有意思。
I think it's interesting just to think about even the whole economics of the pharma industry.
如果你在制药行业开发一个产品，比如一种药，你只能在特定时间内拥有这种药的独家权，对吧？
If you build a product in pharma, right, like a drug, right, you only have exclusivity on that drug for a certain period of time, right?
而且你必须不断再投资。
And you have to continue to reinvest.
礼来现在是一家市值万亿美元的制药公司。
Eli Lilly is a trillion dollar pharma company right now.
如果他们不推出更多重磅药物，他们不会永远保持万亿美元市值。
If they don't get more blockbuster drugs, they will not be a trillion dollar pharma company forever.
我觉得这迫使这些公司必须真正专注于采用新技术、部署这些技术，并努力保持领先。
And I think that forces these companies to really be on their game of adopting new technologies and deploying them and trying to stay ahead.
制药行业竞争非常激烈。
Pharma is a very competitive arena.
有很多人都在努力把这些药带给患者，这对患者来说当然是好事，但这也意味着你必须时刻保持最佳状态。
There's a lot of people trying to bring these drugs to patients, by the way is great for patients, but it means that you have to be on top of your game here.
我觉得这意味着，一旦你跨过那道门槛，模型开始发挥作用，我们看到这种采用速度会非常快，人们会开始以极具创意的方式思考如何使用这些模型。
And I think that means that, you know, once you're through that door and your models are working, we've seen like an upscale this adoption very quickly and people thinking about how to use the models in incredibly creative ways.
我也很喜欢它带来的激励机制，比如更倾向于合作模式。
I really like the incentive structure that it creates as well, like taking more of the partnership model.
随着我们的模型越来越好，我们给客户带来的结果也越来越好，等等。
As our models get better, we get better results to our customers and so on.
所以我觉得这对我们来说也是一个很好的附带好处。
So I think that's just a nice side effect for us as well.
Chai 一直非常专注。
Chai has been incredibly focused.
而且，真的，随着模型越来越好，你可以用更好的数据和更好的模型不断迭代，等等。
And, like, really, as the models get better, you can kind of, like, iterate on a better model with better data and so on.
所以，我们看到的很多情况，以及我们的合作伙伴使用模型的方式，比如生成 binders、antibodies 等等。
So, like, a lot of what we see and, like, a lot of what our partners are using the model for, like, generating, you know, binders, antibodies, so on.
我们也在内部大量使用自己的模型，进行 dogfooding。
Like, we're also, like, doing a lot of dogfooding in house.
我们有一个完整的科学团队，他们正在使用这些模型，试图了解它们能做什么。
We have a whole science team that's using the models and just trying to understand what they can do.
很多反馈都归结为：既然我们已经解锁了某个用例，我们能生成哪些新类型的数据？
And a lot of that comes back to, okay, now that we've unlocked use case X, what type of new data can we generate?
这样怎么能让模型变得更好？
How can we make the model better that way?
实际上，这正是Chai的长期愿景。
And really, that's been the long term vision of Chai.
我们从未想过要构建一个能解决所有问题的单一模型。
We never thought of, we're gonna build this one model that's gonna solve all of this.
我们知道，就像在其他领域一样，这个模型会有多次迭代。
We know that they're just like in other fields, they're gonna be multiple iterations of this model.
随着模型越来越好，我想你会得到这种飞轮效应。
And as you get better and better models, you get this flywheel effect, I guess.
你提到了我们能生成哪些数据。
And you mentioned what data can we generate.
我们能聊聊数据这块吗？
Can we touch on data for a minute?
你们总不能直接去网上抓数据，然后就有构建模型所需的全部数据了吧。
You guys can't exactly just go scrape the internet and have all the data you need to build your models.
那数据是从哪儿来的呢？
Where does the data come from?
它是怎么随着时间不断积累的？
How does it kind of compound over time?
你能简单说说数据作为模型输入的情况吗？
Can you just say a word about data as an input to your models?
对，我觉得数据的主要来源，或者说最标准的数据来源，就是那个蛋白质数据库。
Yeah, so I think the primary source of data or the gold standard source of data is this protein database.
这个数据库其实是真正的实验室科学家，从1970年代开始，就一直在提交蛋白质和其他分子的晶体结构。
So this is like it's actually just legit lab scientists, like, who since, like, 1970 have just been depositing crystal structures of, like, proteins and other molecules over time.
说实话，如果没有这个，结构预测和设计根本就不会存在。
And, like, really without that, like, structure prediction and design wouldn't have been a thing.
所以这些就是结构数据的一个来源。
So there are these so that's, like, one source of structural data.
我刚入行的时候，真的是走了一条非常专注于结构预测的路线。
When I started in the field, I really took, like, a structure pilled approach.
我那时候对结构预测特别着迷。
I was, like, super interested in predicting structure.
比如，怎么设计一个机器学习模型，让它输出三维坐标？
Like, how do you think about a machine learning model that can output three d coordinates?
这不像语言模型那样。
Like, that's that's not look like an LLM.
也不像图像模型那样。
It doesn't look like an image model.
它真的自成一体。
This is really in its own class.
有意思的是，Josh当时走的完全是相反的路线。
Josh, interestingly, was taking, like, the exact opposite approach.
他是ESM的原始作者之一，那项工作可以说是理解如何将语言模型应用于蛋白质序列的开创性成果。
So he he was, like, one of the original authors on ESM, and that was, like, a really, like, a seminal work in understanding how to apply language models to protein sequences.
那项工作最酷的地方在于，如果你能训练一个语言模型来理解蛋白质序列，最终它会在内部隐式地表示出三维结构。
What's really cool about that work is if you can train a language model to understand protein sequences, what ends up happening is it ends up kind of, like, representing the three d structure internally.
而这种情况发生的原因非常有趣。
And there's, like, a really interesting reason why this happens.
所以，为了预测缺失的氨基酸，就像预测句子中的下一个词一样，我想预测蛋白质中的下一个氨基酸。
So, like, in order to predict, like, missing amino acids so, like, same way that you'd predict, like, next word in a sentence, I wanna predict next amino acid in a protein.
要有效地做到这一点，你真的需要理解，对吧？
In order to do that effectively, like, you really need to understand, like, okay.
氨基酸周围的微环境是什么样的？
What does amino acids, like, immediate microenvironment look like?
因为那能告诉你，哪些氨基酸与它周围的环境相容。
Because that kinda tells you, okay, what are, like, the compatible amino acids with everything surrounding it?
而要做好这一点，你需要理解蛋白质的三维结构。
And in order to do that well, you need to understand the protein's three d shape.
所以我当时采用的是那种非常基于结构的方法，而Josh比我更信奉“苦涩的教训”。
So I was taking, like, this this really structure based approach, and Josh was even more bitter lesson pill than me.
他说，我们就直接看序列，结构自然就会浮现出来。
He's like, we're just gonna, like, look at the sequences, and is just gonna emerge.
所以，是的，这两个主要数据来源，一个是蛋白质数据库用于结构，另一个是那些极其庞大的序列数据库，可能达到数万亿个token的规模。
So, yeah, the two main sources of this data, again, like protein database for structures, and then, like, these massive, massive, maybe even, like, order of, like, trillions of token sequence databases.
一旦你有了很好的模型，你就可以在序列数据库上运行它们，从而得到新的结构。
And what you can do once you have really good models is just run them on the sequence databases to get new structures out.
所以，你会有这种叠加效应，模型越好，预测这些结构就越准确。
So again, like you have this compounding effect as your models get better, they get more and more accurate at predicting these structures.
然后你就有更多的训练数据来训练下一代模型。
And then you have more and more training data for the next series of models.
有时候人们会在Chai问我们，现在在Chai，你们到底在追求哪种范式？
Sometimes people ask us at Chai about, these days at Chai, like which paradigm are you actually going after?
我喜欢我们团队的一点是，其实我们两者都不是，我们非常务实。
I think one of the things I love about our team is that it's actually just like neither, we're very pragmatic.
我们就是想解决问题，不关心是序列方法还是结构方法，实际上最后肯定两种都会用到。
Like we want to solve the problem, we don't really care is it a sequence approach, it a structure approach, and practice it's going to end up being like both of course.
你可能会惊讶，互联网上生物序列的数据可能比英文文本还要多。
I think you'd be surprised but there might even be more like biological sequence tokens on the internet than English language tokens.
而且现在很多数据其实没什么用，可能没整理过，噪音也很大，但数据量是很大的，怎么把它们整合起来也是一门艺术。我觉得更让人兴奋的是，模型已经发展到可以在实验室里设计东西了，比如针对新的靶点，我们可以用模型来生成数据。
And now a lot of that data is not that useful, it might not be redone, it might be very noisy, but there's a lot of data out there, there's a lot of art in like bringing it together, and I think also the exciting thing is as the models have gotten to a point now where we can design things in the lab, right, to new targets for instance, we can actually use the models to generate data as well.
所以我们在Chai做的那些实验会产生很多副产品，这些也能帮助模型变得更好。
So there's a lot of exhaust from like all the experiments that we're doing at Chai, which also helped to make the models even better.
这和之前大语言模型的发展轨迹很相似。
So it's a similar kind of takeoff that we saw with LLMs.
比如我在OpenAI的时候，我们试着用强化学习来训练GPT-1架构。
Like when I was at OpenAI, we worked on reinforcement learning of like a GPT-one architecture.
但没成功，因为模型还不够好。
Like did not work because the models weren't like good enough.
后来基础模型足够强了，才开始能做这类实验。
But once the base model got good enough, then you could start to do those kinds of experiments.
我觉得在我们这个领域也开始出现类似的情况了，模型已经达到了一个水平，大家重新开始关注数据，怎么真正把模型纳入到这个过程中来？
And I think there's a similar analogy that's starting to happen in our world now as well, where the models have reached a point where there's actually like a renewed interest in data how do we actually bring the models into the loop on making that happen?
然后这就形成了一个非常有趣的循环，让模型不断改进。
And then that creates another really interesting cycle on compounding improvement of the models.
对。
Yeah.
我们刚才聊到了模型的持续改进，你提到制药行业是一个竞争极其激烈的领域。
We're talking a bit about compounding improvement in the models, and you said something about how pharma as an industry is an incredibly competitive landscape.
有意思的是，我觉得现在大家对用AI和机器学习来生成蛋白质和分子又重新燃起了兴趣。
And it's interesting because I think there's been a renowned interest in using AI and using ML to generate proteins and molecules.
所以你们这个领域实际上已经变得竞争非常激烈了。
And so your arena has actually become quite competitive.
而你们显然在保持前沿方面做得非常出色。
And you guys have obviously done an amazing job at staying at the frontier.
你们已经部署了一年，也锁定了一些制药合作伙伴，这些合作让你们的模型变得更好。
It's been a year of deployment for you, you've locked up a number of pharma partnerships that are making your models better.
那你们是怎么看待竞争格局和保持在最前沿的呢？
But how do you guys think about the competitive landscape and staying at the front of the frontier?
首先，我觉得还是要看结果，别自欺欺人，要严谨。
Well, first of all, I think it goes back to like looking at the results and not fooling yourself and being rigorous.
所以我们做了很多模型评估，主要是为了让模型本身不断进步。
So, one of the reasons why we do a lot of evaluation of the models, we mainly do it just a hill climb in the models itself.
我觉得我们带来的很多能力，都是这个领域里首创的。
I think if you look at many of the capabilities that we've brought in have kind of been like first in the field.
看看我们的Chai-two模型，成功率高到不需要再做大规模文库筛选就能看到结果。
You look at our Chai-two model, right, like getting to success rate of the models that you didn't have to do large library screening anymore to see results.
几个月后，我们又展示了如何把这些可开发性属性，比如分子的可制造性，都整合进来。
A couple months later showing how we could bring in a lot of these developability properties you've talked about before, like the manufacturability of the molecules.
所以很多时候，我们是在推动模型前进，看这些能力能不能出现，然后尽快锁定它们。
So, in many cases, we are pushing the model forward and trying to see these capabilities emerge, and then we try to quickly like lock those in.
怎么让这些能力更加突出，达到生产就绪的状态，然后交付给客户。
Like how do we make those capabilities like even more pronounced that they become production ready and we can like ship them to our customers?
我想跟团队说的一件事是，我们并不是和其他模型提供商正面竞争。
I think one of the things I like to tell the team is that, you know, it's not like we are head to head like with other model providers or something like that.
我们所有人，我觉得，都是在跟大自然较量。
We're all of us, I think, are working against nature.
大自然其实一直是个不错的基准。
Like nature has actually been a pretty good baseline.
人们长期以来一直用某种方式做药物发现，马特提到过，也许你不想给Chai模型加上模块24，但人们已经给现有的湿实验方案加上了模块240，并且这些方案已经被优化得相当不错了。
People have done drug discovery a certain way for a long time and, you know, Matt talked about how like maybe you don't want to add like module 24 to the Chai model, but people have added like module two forty to like the existing wet lab protocols and they have been like tuned quite considerably.
我们团队有位科学家叫安迪·杨，他是最早在MIT做酵母展示的人之一，大概二十年前，他在辉瑞和基因泰克有二十年的经验，真正磨练了这些方法，还拥有一个抗体药物的批准。
One of the scientists on our team, Andy Young, was one of the first people working on yeast display at MIT actually like two decades ago and he's got twenty years experience in like Pfizer and Genentech, like really honing in these methods, has a drug approval to his name in antibodies.
我觉得你看看像他这样的人，安迪知道如何用现有工具做出好抗体，而这实际上是我们需要达到的标准。
I think you look at someone like that and like, you know, Annie knows how to make a good antibody with existing tools and that is actually the bar that we need to clear.
当然，我认为AI的上限会比我们以前能达到的高得多。
Now, of course, I think the ceiling in AI is going be a lot higher than what we've managed to do before.
否则，我们做这个还有什么意义呢？
Otherwise, what would be the point of doing this?
我们创办公司不是为了做一只快十倍的老鼠，对吧？
We didn't start the company just to make, you know, a 10 times faster mouse, right?
我们是为了做出以前不可能实现的突破性药物。
We started this to make breakthrough medicines that weren't possible before.
但最终，这是我们必须达到的标准，才能获得认可。
But ultimately that is the bar that we needed to clear in order to get adoption.
我觉得几个月前我们就达到了那个转折点。
I think we hit that inflection point a couple months ago.
这就是为什么你看到很多大型制药公司的公告。
That's why you've seen a lot of these big pharma announcements.
但现在我们只需要继续努力，让这些变得更好。
But now we just need to continue to hone in on like making these things even better.
如果有听众觉得，哇，这听起来很酷。
For somebody who's listening who thinks, wow, this sounds pretty cool.
想知道在Chai工作是什么感觉。
I wonder what it would be like to work at Chai.
在Chai工作最好的地方是什么，最不好的地方又是什么？
What is the best thing about working at Chai and what is the worst thing about working at Chai?
对，这个我可以说说。
Yeah, I can speak to some of this.
最不好的地方，我临时想想。
I'll think of this on the fly, the worst thing.
我觉得最好的地方就是，大家真的都特别有使命感。
So I think like the best thing is just like how actually like mission driven everybody is.
每个人都非常投入自己做的事情。
Like everyone is so dedicated to what they're doing.
我在其他公司工作过。
I've worked at other companies.
最接近这种感觉的，可能是我读博时实验室里的那些人。
The closest I've ever seen to this is maybe some of the guys in my PhD lab.
但这里的每个人都特别特别有干劲。
But everyone is just incredibly, incredibly motivated.
我们大家都非常努力。
We all work really hard.
我们有一个明确共同的目标。
There's an obvious shared goal.
我觉得这真的很难得。
And I think that's really rare to see.
我觉得这要归功于我们从一开始就有的专注。
And I think this goes back to just the focus that we've had since the beginning.
我们一直有清晰的理念和计划，知道怎么去实现目标，怎么让事情变得更好。
We've always had kind of a clear philosophy, a clear plan on how things are going to get there, how things are going to get better.
而且，Chai的每个人都真心认同这一点。
And really, everyone at Chai is very bought into this.
看到大家投入这么多，真的非常令人惊叹。
It's pretty amazing just to see, like, the amount of dedication that that everyone's putting in.
在Chai最不喜欢的一点，可能不是蒲公英巧克力本身。
Least favorite thing about Chai, not directly on top of dandelion chocolate, maybe.
不是。
No.
可能是下一个办公室吧。
Maybe the next office.
对。
Yeah.
其实，现在我最不喜欢的部分大概是……我想是想办法让系统在大规模下正常运行。
Actually, so probably the least favorite part now is just like, I guess it's getting things to work at scale.
说实话，当时我根本不知道那是什么意思。
And really, didn't even know what that meant.
其实我们刚开始做 Chai 的时候，手里只有 128 张 GPU。
Actually, when we started Chai, we had 128 GPUs.
我当时就觉得，这已经是实验室能达到的最大规模了。
And I was like, this is like the most scale possible for a lab.
真的觉得太疯狂了。
Like, this is crazy.
我博士课题组那会儿，四个人共用八块GPU。
I came from like, you know, my PhD group where there were four of us sharing eight GPUs.
当时我就觉得，自己算得上GPU大户了。
And I was like, I just felt GPU rich.
那简直太疯狂了。
It was crazy.
所以现在在Chai，我们需要维护的基础设施多得多。
So now at Chai, we have a lot more infrastructure to maintain.
计算资源也多了不少。
We have a lot more compute resources.
幸好GPU可以并行，但这也带来了它各自的问题。
Luckily GPUs are parallelizable, but that also kind of brings up its own set of problems.
比如，怎么让集群长期保持健康运行？
So just like, how do you keep a cluster healthy over time?
怎么才能完成那么大规模的训练？
How do you get that large training run?
那你怎么让它连续运行好几个月呢？
So how do keep you that running for months on end?
即使它中途出问题了，你怎么自动恢复这些任务？
And even when it does die, how do you automatically resume these things?
你怎么降低所有通信开销？
How do you keep all of your communication down?
怎么优化模型，充分利用现有的资源？
How do you optimize the models and make the best use of the resources that you have?
我觉得这些问题会不断冒出来。
So I think these are a lot of problems that they continuously pop up.
这些是好问题，但我觉得它们也真的很难解决。
They're good problems to have, but I think they're also really difficult to solve.
是的，我很期待能解决这些问题。
Yeah, and I'm just excited to work on this.
对，我记得我们合作过几次的一位CEO，弗兰克·斯洛特曼，说过一句话：你要么承受失败的痛苦，要么承受成长的痛苦。
Yeah, I remember one of the CEOs that we've worked with a couple of times, Frank Slootman, had this line about, You either have the pain of failure or the pain of growth.
你宁愿选择成长的痛苦。
You'd much rather have the pain of growth.
对，完全正确。
Yeah, that's exactly right.
我觉得我最喜欢的部分大概是结果。
I think my favorite part is probably the results.
听起来有点老套，但确实有效。
And that sounds a bit cliche, there's nothing That's working.
它真的有效，对吧？
It's working, right?
而且知道你在做这件事，你知道，我觉得公司里很多人，对吧，我们已经在AI领域工作了很久了，对吧？
And just like knowing that you're, you know, I think many of us in the company, right, like we've been working in AI for a long time, right?
积累了这么多经验。
There's all this experience you've built up.
知道你在把它应用到真正重要的事情上，就拿我和Matt来说，我们在这个问题上已经工作了大概十年了，对吧？
To know that you're applying it to something that really matters, just even, I think just take Matt and I, like we've been working on this problem for like ten years, right?
几年前，我看着一些东西，比如我们在写一些很酷的论文，大家都在庆祝。
And a couple of years ago, I'm looking at something like, yeah, we're writing some cool papers, like everyone is like celebrating this.
我们真的在让世界变得更好吗？
Are we actually making the world better?
这真的会影响一些病人吗？
Like, is this actually going to impact some patients?
我觉得现在答案是肯定的，我们确实达到了一个能对世界产生重大影响的阶段。
And I think now the answer is like, actually yes, like we have reached the point where this is going to make a big difference in the world.
每次取得突破性成果，每当产品有新功能让客户的生活更轻松，每当科学团队带回新的实验室结果，说实话，总是让人兴奋地意识到，这真的会以相当深远的方式改变世界。
And every time you get one of these breakthrough results, anytime there's a new feature on the product that makes our lives of our customers easier, whenever there's like new lab results coming back from the science team, it's just always so honestly it's exhilarating to realize like this is actually going to change the world in a pretty profound way.
我觉得这也许也涉及到最不喜欢的一面，对吧？
I think that's also then maybe it comes to the least favorite side, right?
你知道，我们在经营公司，有真正的合作伙伴依赖我们，事情必须得成功，对吧？
Like, you know, we're running the company and it's like we have real partners that are relying on this and like things have to work, right?
你发布了一个新模型版本。
And you ship a new model generation.
那你怎么确保里面没有bug呢？
Do How you make sure that there's no bugs in that?
你怎么确保不会出现回归问题，对吧？
How do you make sure you don't have regressions, right?
这不再只是那种天马行空的研究问题了，比如‘哦，我们有了些不错的结果，然后继续前进’那种。
This is no longer just like a, again, the blue sky research problem of like, oh, we got some cool results and move on.
随着团队壮大，我们不得不把生产级代码库作为高度优先的事项。
We've had to have really high priorities on like, you know, having production level code bases as the team grows.
我们怎么确保代码库的状态能让更多人参与贡献呢？
How do we make sure that code base is in a state that more people can contribute to this?
所以，我们另一位联合创始人杰克喜欢说，如果你想长期快速前进，有时短期就得稍微慢一点，确保你构建的东西——这又回到了那个复利的概念。
So, something that one of our other co founders, Jack, likes to say is that if you want to move fast in the long term, you sometimes have to just move a little bit slower in the short term, right, and make sure that you are building something, again, that goes back to that compounding idea.
这又回到了我们不会为了下一个功能而添加第25个模块。
Goes back to we don't add module 25 to make the next thing work.
所以，有时候你特别兴奋想赶紧得到下一个结果，就想直接冲进去，但我们有真实的合作伙伴，世界上一些最大的公司现在都在依赖我们，重要的是我们要意识到这一点，认真对待这份责任，确保我们构建的系统能持续运行。
So, sometimes, you know, you're like so excited to get to the next result and you just want to jump into it, but we have real partners, some of the biggest companies in the world that are now relying on us, and it's important that we realize that, we take that responsibility to heart, and we make sure we're building systems that, you know, continue to work.
我有一个特别想问的问题。
I have a burning question.
你们为什么叫Chai Discovery？
Why are you guys called Chai Discovery?
化学和AI。
Chemistry and AI.
但我们也喜欢印度奶茶。
But we love Chai tea as well.
办公室里有很多和奶茶相关的东西。
It's a lot of Chai themed stuff in the offices.
没错。
That's right.
问得好。
That's a good question.
我之前也不知道。
I didn't know that either.
乔希是个有远见的人，那都是他的主意。
Josh is the visionary, That was all him.
这个名字确实很好记。
It is a very user friendly name.
对。
Yeah.
我们也想要一个名字，因为生物技术公司的名字都那么复杂。
We also wanted a name that like biotech companies have such complicated names.
是的。
Yeah.
我们想要一个更简单的名字。
We wanted something that's gonna be much simpler.
对。
Yes.
我们想让这一切变得更简单，对吧？
We're trying to make this whole thing simpler, right?
对。
Yeah.
所以我们得有个简单的名字来搭配。
So we need a simple name to go along with it.
太棒了。
Awesome.
接下来六到十二个月，你们最期待什么？
What are you guys most excited about in the next six to twelve months?
对我来说，就是正在进行的那些部署。
I think for me, it's just the deployments that are happening.
我们已经宣布了几个合作伙伴关系，我真的很期待听到合作伙伴带来的成果。
So we've announced a couple of these partnerships and I'm really excited just to hear about the results that our partners are bringing online.
这又回到了真正产生影响这一点，而且这也是为什么我很高兴看到这些合作进展顺利，你知道，即使是在签署协议的时候，以及当我们和这些人合作时，模型并没有被束之高阁，而是真正用在了实际项目上。
It goes back to this point of like making a real difference and also why I'm so happy to see how these partnerships are going even, you know, posting the agreement and as we're working with these folks that the models are not sitting on a shelf somewhere, like they're actually being used on real programs.
人们正在尝试攻克毁灭性的疾病，如果Chai能给他们一个有效的分子，那真的能改变患者的生活。
People trying to approach devastating diseases where, you know, if Chai could give them a molecule that works, it could really change the lives of patients.
所以我真的很期待看到结果如何。
So I'm really excited to see how that goes.
这里的进展速度真是惊人，模型的发展速度也一样。
Just the pace of progress here is incredible, but also just like the pace of the models.
比如一年前，你不可能凭空生成一个分子，还能有把握觉得你的项目会成功。
Like a year ago, you could not zero shot a molecule and like, you know, have a good sense that your program was going to work.
但现在情况已经变了。
Like now that's changed.
现在有人可能凭空生成一个分子，然后说，我觉得我们能把这个项目推进到临床，一年后，他们甚至可能已经有第一批分子进入患者体内了。
Someone might zero shot a molecule and be like, I think we're going to bring this program to the clinic And then a year later, you know, they might even have some of those first molecules going into patients.
这种速度简直不可思议。
And just the speed of that is just incredible.
而且，有时候你会想到这些就起鸡皮疙瘩，比如，我的模型会……马特在我们上一家公司有一些专利，关于在电脑上生成分子，而现在这些东西已经用在患者身上了，想想这个规模，我不知道，几年后，我们会有几十个还是几百个Chai分子进入人体？
And it's, you know, sometimes you get some shivers thinking about this that like, okay, my model is going to, like Matt has some patents from our last company about, you know, like just generating the molecule on the computer, and these things are now in patients, just to think about the scale, like, I don't know, few years from now, do we have dozens, do we have hundreds of like Chai molecules going into people?
想到这对患者可能意味着什么，真有点让人震惊。
It's a bit mind blowing to think about what that might look like for patients.
我觉得，让Chai与众不同的一个点就是大家的投入。别人总说，你工作真多。
I think one of the things that, again, what makes Chai unique, the dedication, people are like, Man, you work a lot.
你不会累垮吗之类的？
Aren't you burned out or whatever?
其实这很轻松，也很有动力，当你看到他们取得的进展，就会想，下一个模型肯定会更好。
It's actually really easy and it's very motivating when you're making Seeing the progress they're making and just thinking, Man, the next model is going be even better than the last.
我们发现了这个新东西，等等。
We identified this new thing, so on.
这真的太让人有动力了。
That's so incredibly motivating.
对我来说，更像是，我们接下来能解锁什么？
For me, it's more of just like, What can we unlock next?
怎么让这些东西更可控？
How do we make these things more controllable?
可控？
Controllable?
而且，当有人带着特定目标来找我们时，我们不只是希望得到好的亲和力之类的，你知道，我们真的能控制这个吗？
And like, when someone comes to us with a certain target, instead of just like hoping we get good affinity or something like that, you know, can we actually control this?
我们能说，我们想要一个正好是10纳摩尔亲和力的结合剂吗？
Can we say like, we want exactly a 10 nanomolar binder?
诸如此类的事情。
Things like that.
我觉得有很多技术问题，我们差不多快要解决了。
Like, there are a lot of technical things that I think are, we're like kind of right on the brink of solving.
对我来说，把这些东西搞定，把这一切都完成，然后看看接下来会走向哪里，这真的很让人振奋。
And for me, it's really motivating just to pin those things down and just get all of this over the line and see kind of where that leads to next.
太棒了。
Awesome.
马特，乔什，感谢你们做工程生物学，也感谢你们今天和我们分享你们的故事。
Matt, Josh, thank you for engineering biology and thank you for sharing your story with us today.
谢谢你们。
Thank you guys.
谢谢，谢谢
Thank Thanks
感谢你们的邀请。
for having us.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。