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。
如果我告诉你，有人在后院的拖车实验室里打造了一个价值2.7亿美元的化学AI，而今天这个AI正在帮助重新设计数十亿人使用的各种产品，从泰诺到露得清。
What if I told you someone built a $270,000,000 chemistry AI that started in a backyard trailer lab, and today that AI is helping redesign products used by billions of people from Tylenol to Neutrogena.
让我们聊聊材料科学的未来。
Let's talk about the future of materials science.
欢迎各位来到Neuron播客。
Welcome humans to the Neuron Podcast.
我是科里·诺尔斯，和往常一样，由格兰特·哈维与我一同主持。
I'm Corey Knowles and joined as always by Grant Harvey.
今天过得怎么样，格兰特？
How you doing today, Grant?
还不错。
Doing good.
还不错。
Doing good.
今天非常兴奋，因为我们邀请到了阿尔伯特创新公司的首席执行官兼联合创始人尼克·托尔金，这家公司是一个AI平台，正在改变全球一些最大公司进行化学研究的方式。
Very excited today because we are talking to Nick Tolkien, CEO and cofounder of Albert Invent, an AI platform that is transforming how chemistry gets done at some of the world's biggest companies.
很好。
Excellent.
好的，尼克，欢迎来到Neuron播客。
Well, Nick, welcome to the Neuron.
很高兴有你。
Great to have you.
嘿，各位。
Hey, guys.
很高兴能来这里，谢谢你们邀请我。
Great to great to be here, thanks for having me.
很好。
Excellent.
我想，我们先来聊聊你的创业故事吧。
Well, guess I guess to start out, let's talk a little about your origin story.
我听说你和肯在后院的一个拖车实验室里创立了Molecule Molecule公司。
I understand you and Ken started molecule molecule corp in a literal trailer lab in the backyard.
是吗？
Is that right?
是的。
Yeah.
这真是个有趣的故事。
It's it's a fun story.
所以，是的，肯在那之前已经在化工行业工作了大约二十五年。
So, yeah, Ken has been in the the chemical industry for about twenty five years before that.
他实际上是在他父亲的油漆厂长大的。
He actually grew up in his dad's paint factory.
所以这个故事还要追溯得更早，也许你以后得请他上播客来讲讲这部分故事。
So the story goes even farther back, and maybe you'll have to get him on the podcast at some point to tell that part of the story.
但没错，早在2014年，他把一辆拖车开进了自家后院，启动了他的第三家材料科学公司。
But, yeah, back in, in 2014, he pulled a, a trailer into the his backyard, to start his third material science business.
这家公司，正如他所说，叫做Molekule。
That company, as he said, was called Molekule.
这主要是关于改变化学的发明方式。
And it was really about, changing the way that chemistry is invented.
因为他对整个职业生涯中化学领域几乎没有变化感到沮丧。
Because he was frustrated that it hadn't really changed in his entire career in the industry.
因此，这对我来说是一个绝佳的机会，可以加入他并帮助他实现这一使命。
And so it was a a great opportunity for me to join him and help him on that mission.
这太棒了。
That's awesome.
所以我很期待和你交谈，因为我觉得现在我们已经到达了一个阶段，我们今天所知的AI代理，有些人能很好地使用它们。
So I I'm really excited to talk to you because I feel like right now we've gotten to a point where the AI agents that we, know of today, you know, some people can use them well.
有些人则开始发现它们的局限性，尤其是在工作场所、B2B、SaaS领域。
Some people, are are finding the limits of how useful they are, especially in, like, workplace, b to b, SaaS, that area.
但我认为AI与科学是尚未被充分开发的领域，最具潜力，也最有可能带来积极影响。
But I feel like AI and science is the untapped kind of, like, area that is the most exciting where there's the most potential benefit for good.
我真的很期待从这个角度来探讨它，尤其是关于化学方面。
And I'm just, like, really excited to talk about it from that standpoint, especially for chemistry.
对于非化学家的人来说，当你说Albert接受了1500万个分子结构的训练时，这到底意味着什么？
So for folks who aren't chemists, when you say Albert is trained on 15,000,000, molecular structures, what does that actually mean?
你能给我们详细解释一下吗？
You wanna just walk us through that?
好的。
Yeah.
嗯。
Yeah.
所以，我们公司的一个核心信念是，你不能简单地拿现成的通用大型语言模型、生成式AI、机器学习等直接应用到科学领域。
So, you know, one of our core beliefs as a company is that you can't just take like off the shelf generic large language models, gen AI, machine learning, whatever, and start applying it to science.
我认为，如果这真的那么容易，早就有人做到了。
I think if that if it was that easy, it would have been done a long time ago.
因此，你必须利用世界上已存在的底层数据。
And so what you have to do instead is you have to take advantage of the underlying data that exists out there in the world.
而目前世界上主要存在两种形式的数据。
And there's basically two forms of data that exists out there.
有公开可用的数据。
There's the publicly available data.
比如专利格局、学术文献，以及来自学术界的各种资料。
So that's the patent landscape, the literature, you know, stuff coming out of academia.
还有企业数据。
And then there's the enterprise data.
Albert 的使命是帮助最大、最重要的企业利用这两种数据来源。
And our mission at Albert is to help the largest and and the biggest enterprises take advantage of both of those sources.
也就是公共数据和私有数据。
So the public and the the private.
当你提到1500万种分子时，这些实际上来自公共领域。
So when you mentioned 15,000,000 molecules, that's actually coming from the public space.
你知道，有很多由政府机构、学术界等发布的公开数据。
You know, there's there's a lot of publicly available data out there put out by government agencies and academia and and and the like.
我们利用这些公开数据构建化学领域的基础模型。
And we build foundational models of chemistry with that public data.
但坦率地说，问题在于，之所以并不容易，是因为公共领域的数据通常是成功的案例，对吧？
The problem, frankly, though, and and again, the reason it's not that easy is that generally the data in the public domain is, is the successes, right?
很少有论文或专利会专门发布所有的失败结果。
There aren't many papers and patents that are publishing just all the failures.
如果你还记得你做科学实验的时候，我猜你在实验室里做的，以及每个人做的实验，大多数都是失败的，对吧？
And if you've, you know, remember back to your your science experiments, I assume that most of the experiment, the experiments you did in the lab and that everyone does, they're failures, right?
这正是你学到东西的地方。因此，仅仅依靠这些数据，即使有如此庞大的公开实验数量，也不足以解决当今行业面临的某些问题。
That's actually where you learn the And so just taking that data, even with that large number of publicly available experiments, it's not enough to just crack some of the problems that the industry faces today.
哇。
Wow.
这真的很有趣。
That's really interesting.
你的平台被称为端到端的研发平台。
Your platform is called an end to end R and D platform.
这在实践中意味着什么？
What does that mean in practice?
假设我是你们的客户，比如汉高或凯视的科学家，使用Albert之前和之后，我的日常工作有什么不同？
Say I'm a scientist for one of your customers like Henkel or Kenview, what does my day to day look like before Albert versus after?
好问题。
Oh, good question.
在使用Albert之前，我认为这反映了整个行业的普遍情况，就像达·芬奇当年做化学实验时那样。
So before Albert, and and I think this is the industry at large, it's it it resembles, I think, what it was when probably like Da Vinci was doing chemistry back in the day.
所以，首先你得去弄清楚，你的原料有哪些？
And so, you know, first you have to, you gotta go figure out like, what are your ingredients?
你想测试哪些物质？
What are the things that you want to go, test?
通常这意味着你要走到仓库，看看货架，然后琢磨：好吧，我想研发一款新的洗发水，对吧？
And so generally, means you're gonna walk over to your stockroom, and you're gonna look at the shelf and you're gonna figure out, okay, you know, I I'm trying to make a new, let's take a new shampoo, right?
举个例子，我要开发一款新的洗发水。
As an example, I'll make a new shampoo.
这里面需要很多成分。
You need a lot of stuff in there.
这可不是只有一种成分。
That's not just one ingredient.
这就像烤蛋糕一样。
It's like baking a cake.
你要去收集所有这些原料，然后决定用哪些。
You know, you go get all those ingredients and figure out what you wanna use.
作为化学家，你通常要解决两个主要问题。
And there's two main problems as a chemist that you generally are trying to solve.
要么你是在做渐进式改进。
Either you're trying to make an incremental change.
比如你遇到了关税问题或供应链问题，需要将一种成分换成另一种，或者你试图创造一种前所未有的全新产品，具备客户一直期待的惊人特性。
So maybe you're dealing with a with a tariff issue or you're a supply chain problem, and I need to swap one ingredient for another, or you're trying to make like a net new novel integration, some product that the world's never seen before that has some amazing attributes that customers have been asking for.
无论你打算做什么，都必须弄清楚如何与供应商合作。
Regardless of what you're trying to do, you have to go figure out how do you work with your vendors, right?
这就需要去你的原料库，查看你有哪些材料，也许还要了解别人做过什么。
So that's going to your stockroom, figuring out what you've got there, maybe figuring out what somebody else has done.
很多时候，就像站在饮水机旁，去问别人：嘿，我正想做一个新的环保洗发水，在这个应用场景下，我在寻找一种来自绿色原料的聚合物。
A lot of sitting by the water cooler, so to speak, and asking, you know, people, hey, I'm trying to make a, you know, a new sustainable shampoo in this use case, and I'm looking for a polymer that comes from a green source.
所以它不是石油基聚合物，而是来自可持续的、可能是生物来源的材料。
So it's not an oil drive polymer, but it's coming from a sustainable, maybe bio source.
你以前在行业内有没有发现过类似的东西？
Have you ever found something like that before in the industry?
于是你搞清楚了这一点，这就是我们所说的‘发现’，对吧？
And so you figure that out and that's We call that discovery, right?
前人已经做过哪些工作，这样你才能站在他们的肩膀上。
What has been done before, so you can stand on the shoulders of those who come before you.
历史上，这主要靠口口相传，对吧？
Historically, that's been very word-of-mouth, right?
然后你确定了自己想做什么，就会拿出纸质笔记本，开始草拟一个看起来不错的实验方案。
Then you figure out what you kinda wanna do and you'll take out your paper notebook and you'll start to sketch, you know, what what looks like a good experiment that I wanna go run.
我要把这些东西混合在一起。
I'm gonna mix these things together.
我会以某种特定方式使用它。
I'm gonna apply it in certain way.
我想测试并表征它。
I might wanna test it and characterize it.
如果是洗发水，最终总得有人把它涂在头上，真正地用它洗澡，然后给出反馈。
If it's a shampoo, somebody somebody is eventually gonna have to put it on their head and actually, you know, bathe with it and, like, give feedback.
嘿。
Hey.
我的头发感觉滑滑的，或者感觉特别好。
It feels my hair feels slippery or it feels really, you know, nice.
我洗干净了吗？
It's Am I clean?
我的皮肤
Is my skin
还在吗？
still here?
没错。
Exactly.
所有这些好的
All that good
东西。
stuff.
对吧？
Right?
从一到五给你的皮肤质量打分，所有
Rank it, your skin quality from one to five, all
这些好的方面。
that good thing.
而且，
And,
然后你收集所有这些数据，但通常第一次就成功很难。
and then you you collect all that data and and you generally don't hit it on the first mark.
所以你会问，从中学到了什么？
And so you say, what what did you learn from that?
对吧？
Right?
这就是科学方法，非常经典，我们学校里学过的所有东西。
And this is scientific method, very classical, everything, you know, that we learned in school.
但你可以看到，这是一个非常手动且缓慢的过程。
But you can see that's a very manual slow process.
对吧？
Right?
仅仅进行一次实验就可能需要几天甚至几周的时间。
Just to run one experiment could take you days, weeks potentially to go do that.
因此这很昂贵。
And so it's expensive.
你不可能生成数百万个实验数据点。
You're not gonna be generating millions of experimental data points.
这很慢。
It's slow.
而且你发现新信息的方式仍然是非常手工的，就像长期以来一直做的那样。
And the way that you discover new information is still very analog and kind of, you know, how it's been done for a long time.
因此，通过Albert，我们改变了这一切。
And so with Albert, we change all that.
我们把整个工作流程——从你公司或行业之前已有的发现，到最终在人头上测试洗发水的应用——全部捕获到一个单一的数据源中。
We take that entire workflow from the early discovery of what's been done before in your company or in the industry to the final application testing that's being done with the shampoo in somebody's head, and we capture all of that data in a single source of truth.
因此，为组织内所有信息的本体建立一个记录系统。
So a system of record for the entire ontology of information in the organization.
然后，这成为所有后续实验的基础。
And then that serves as the basis for all of what the next experiment does.
所以现在你有一位在德国的同事正在运行类似的实验。
So you've got a colleague in Germany now who's running a similar experiment.
如果这种聚合物对你效果不好，他们就不会再使用它，因为他们知道它效果不好，或者他们知道它哪里有效，从而可以更好地利用它。
They're not gonna use that same polymer if it didn't work well for you because they know that it didn't work well, or maybe they know how it did work well and you can use it better.
因此，我们某种程度上是在实现知识的时间转移。
And so what we're doing is we're time shifting knowledge to some extent.
这听起来可能有点疯狂，但我们的做法是将原本只存在于某个人大脑或纸质笔记本中的知识，分发到整个组织中。
And that might seem kind of crazy, but it's we're taking knowledge that used to be encapsulated in one person's brain or in a paper notebook, and we're distributing it throughout the organization.
因此，每个人都能自由访问，并做出更明智、更快的决策。
So everybody has free access and can make better and faster decisions.
所以，如果你现在正在开发任何重要的项目，无论是应用、API还是AI功能，这个都值得你特别关注。
So if you're building anything serious right now, whether it's apps, APIs, AI features, this is one you really want to pay attention to.
Google开发者计划高级版专为那些真正希望加快交付速度、而非整天只阅读文档的开发者设计。
The Google Developer Program Premium plan is designed for developers who actually want to ship faster, not just sit and read docs all day.
因此，当你升级为高级版时，即可解锁Google全套开发者工具，包括Gemini代码助手和Gemini命令行工具。
So when you go premium, you unlock the full breadth of Google's developer tools that includes Gemini Code Assist and the Gemini CLI.
你可以直接在工作流中编写、调试和迭代，AI是内置的，而不是事后附加的。
So you can write, debug and iterate with AI built directly into your workflow, not bolted on after the fact.
而真正重要的部分在这里。
And here's the part that really matters.
您还将获得高达540美元的生成式AI和Google Cloud积分。
You also get up to $540 in Gen AI and Google Cloud credits.
这为您提供了真正的实验空间，可以进行原型设计，甚至扩展项目，而无需担心第一天就把预算花光。
That's real runway to experiment, to prototype, or even scale without worrying about burning all of your budget on day one.
无论您只是在测试新的AI功能、部署生产工作负载，还是希望以更少的麻烦加速进展，Premium 都能立即为您提供真正契合现代开发者工作方式的工具和积分。
So whether you're just testing new AI features, deploying production workloads, or you're trying to move faster with fewer headaches, Premium gives you tools and credits right away that actually support how developers work today.
如果您准备更智能地构建，并从Google生态系统中获得更多信息，请加入Google开发者计划的高级版。
If you're ready to build smarter and get more out of the Google ecosystem, go premium with the Google Developer Program.
今天就前往 developers.google.com/program 了解详情。
Check it out today at developers.google.com/program.
网址是 developers.google.com/program。
That's developers.google.com/program.
我们将在描述中为您提供完整链接。
We'll have a full link for you down in the description.
现在回到我们的节目。
And now back to our show.
你们在向企业客户推广时，有没有遇到他们不愿意共享数据的阻力？
Do you get any pushback from enterprise clients on that who don't wanna share their data?
因为我觉得，试图在科学领域创新时，最大的问题之一就是，这些独立的参与者往往在一定程度上垄断知识，以便从中获利。
Because I feel like that is one of the biggest problems with, you know, trying to, you know, innovate in science is you have all these independent players who are trying to, like, gatekeep knowledge to a certain extent, you know, so that they can benefit from it.
是的。
Yeah.
我认为在一家公司内部，如果你看一家企业，比如像肯维这样的公司，你会听到这样的说法。
I think inside of the a company, I think there you know, if you're looking at one enterprise, you'll hear, like, a Kenview, for instance.
我认为现在有巨大的推动力去实现知识或数据的民主化，因为他们意识到市场环境变化如此之快，如果不这么做，他们怎么跟上消费者趋势和消费者行为的变化呢？
I think that there's huge push to try to democratize that knowledge or democratize that data because they've realized that the market conditions are changing so quickly that if they're not doing that, how do they keep up with, you know, consumer trends and consumer behaviors out there?
当你思考更广泛的供应链，以及公司之间如何相互共享信息时，这完全是全新的。
Now when you think about the broader supply chain and you think about how companies share information with each other, that's net new.
公司过去并没有很好地做到这一点。
Companies have not been doing a
很长时间以来，它们在这方面做得并不好
very good job at that for a lot of
你之前提到的那些原因，格兰特。
the reasons you kinda talked about, Grant.
我的意思是，他们害怕这么做。
I mean, it's it's they're scared of doing that.
如果我分享这个知识产权，它会被如何利用？
If I share this IP, how does it get exploited?
谁会从中获益？
Who can take advantage of it?
但我们开始看到这种旧思维模式出现了裂痕。
But we are starting to see cracks in that old way of thinking.
所以我们还有另一个客户叫Neurion，他们把我们的——它叫Ask Albert。
And so we've got another customer called Neurion, and they have put our our it's called Ask Albert.
这是我们的LLM，放在了他们的网站上。
It's our our our LLM and and on their website.
现在他们的所有客户都可以与之交互。
And all their customers now can interface with it.
但你仍然需要穿过一堵墙。
Now there's still a wall that you have to go through.
你仍然需要注册。
You still have to sign up.
他们会验证你是否是真正的合法客户，而不是竞争对手之类的人。
They're gonna validate that you're an actual legitimate customer and not maybe a competitor or something like that.
但这非常令人兴奋，因为我们现在不只是在改变单个公司的创新方式。
But that's super exciting because now we're not just changing how a single company invents.
你可以开始思考整个供应链的运作。
You can start to think about how the entire supply chain events.
对我们消费者而言，这才是我们真正需要的。
And for us as consumers, that's what we really need.
解决一家公司的问题，并不能推动整个制造业共同前进。
Solving one company's problem isn't gonna solve how the entire manufacturing industry moves forward together.
哇。
Wow.
你如何看待美国本土制造业回流这一努力，以及这到底有多大的影响呢？
How are you thinking about that in terms of like the efforts to onshore manufacturing in The US and, you know, like how how much of yeah.
总的来说，你对这个话题有什么看法？
Just in general, what are your thoughts on that topic?
我的意思是，这是一个很大的话题。
I mean, it's a it's a big topic.
是的。
Yeah.
是的。
Yeah.
是的。
Yeah.
当然。
For sure.
是的。
Yeah.
我觉得，你知道，我非常支持制造业。
I think, you know, I'm a big fan of of manufacturing.
对吧？
Right?
我是在加州后院的一个实验室工作台和拖车里创办了阿尔伯特公司。
I I started Albert in a lab bench and a trailer in the backyard in California.
对吧？
Right?
所以，如果我们不相信亲自动手、自己制造东西，这家公司根本不可能存在。
And so if we didn't believe in, like, getting your hands dirty and and making something for yourself, I don't think this company would ever exist.
所以，是的。
And so Yeah.
我认为，努力实现本地化制造是有重大意义的。
I I think that, you know, there's a big reason to try to have localized manufacturing.
不把产品运往世界各地，也有可持续发展的原因。
There's sustainability reasons of not shipping things all over the world.
还有竞争优势、国家安全等问题，所有这些都很重要。
There's also competitive advantage, national security issues, all that good stuff.
所以我认为我们正看到一个巨大的趋势。
So I think we're seeing a big trend.
我希望我们不会看到公司或国家之间开始筑起壁垒的情况。
I hope we don't see it where we start to build walls between companies or between countries.
这对全球经济体系不利。
That's not good for the global economic system.
但我认为我们并没有看到很多这样的情况。
But I don't think we're seeing a lot of that.
至少从我所在行业的视角来看，我们看到人们正在重新投资于以前可能没有投资的领域，比如本地制造。
At least my vantage point in the industry is we're seeing people just reinvest in areas where they maybe haven't invested before, like in local manufacturing.
这让我非常兴奋，因为就业增长很好，我们希望优秀的科学家能在美国本土工作。
And that's really exciting to me because job growth is great, and we want smart scientists to be located here in The US.
当然。
For sure.
你知道吗，我看到你们的一个客户提到，以前需要三个月的项目，现在在使用Albert的情况下，有些已经缩短到仅需两天。
You know, I saw that, one of your customers had said projects that used to take three months, they've now been able to knock down to as little as two days in some cases with Albert.
你能为我们详细解释一下，究竟是什么导致了如此显著的速度提升吗？
Could you walk us through what's actually happening there to create that kind of speed up?
一般来说，主要有两个方面。
Generally, there's there's two things.
如果我们回到我之前提到的科学方法，你会经历一个从想法到获得数据的迭代过程，而这个过程通常以失败告终，因此需要多次迭代。
So if we go back to kind of that scientific method I was talking about before, you've got one iteration of how do you go from an idea to something where you have data, and generally it's a failure, so you do multiple iterations there.
我们的目标是缩短每次迭代的时间，并尽可能减少迭代次数。
And so our goal is to reduce the time per iteration and collapse the number of iterations down to as few as possible.
如果我们能做好这两点，科学进展就会更快。
And if we can do those two things, then science gets faster.
在这个例子中，我认为在使用Albert之前，他们每次迭代可能需要三到四天。
And so in that example, I think each one of their iterations previous to Albert maybe took, you know, three or four days.
对吧？
Right?
因此，他们通常会进行大约十次迭代，整个项目可能需要几个月才能完成。
And so then generally they run 10 iterations or so, and it's a couple of month long project that they'd have to run.
对吧？
Right?
这就是过去的做法。
And that's the old, you know, old way of doing it.
有了Albert，因为我们能够更轻松地协作，他们可以将迭代时间缩短到两天甚至一天。
With Albert, because we are able to collaborate easier, they can collapse that down to maybe a two day iteration or maybe a one day iteration.
而且，由于我们可以将AI叠加在他们的历史数据之上，能够推荐那些能带来最高信息密度的实验，供科学家们进行。
And then because we can layer the AI on top of their historical data, we can start to recommend the experiments that the scientists can go run that give the highest information density per experiment.
这听起来可能有点奇怪，但这就是科学的本质。
And that sounds a little weird maybe, but it's like, that's the point of science.
当你进行实验时，你希望站在前沿。
When you run something, you wanna be on the bleeding edge.
如何才能获得最多的信息，以指导下一次实验？
How do you get the most information to inform the next experiment?
尽可能多地学习。
Learn as much as possible.
是的。
Yeah.
每个月。
Per month.
没错。
Exactly.
所以如果
And so if
你这样做，那么你
you do that, then you
可以进行两次迭代，而不是十次。
can take two iterations instead instead of 10.
因此，如果你缩短每次迭代的时间，就能从十次减少到两次。
And so if you collapse the time per iteration, you go from 10 to two.
现在你只需要两天时间，就已经推出了产品，或者获得了具有商业可行性的产品，这非常非常令人兴奋。
Now you're at two days and you've launched a product, or you've gotten a product that is now commercially viable, which is super, super exciting.
你是通过某种模拟来实现这一点的吗？
Is that are you doing that with the like, in simulation at some point?
还是你就只是觉得，好吧，不错。
Or are you just, okay, cool.
是的。
Yeah.
所以实际情况是，我们基本上就是这样做的，而这正是机器学习发挥作用的地方。
So what what happens is we basically so and that's where the machine learning comes in.
我认为这一点对你们这样的技术受众来说非常重要。
I think that this is important, especially for a technical audience that you have.
大语言模型并不是解决所有问题的万能钥匙。
LLMs are not just the root, like, or the solve for every problem out there.
我认为它们在探索发现方面非常出色。
I think they're really good at discovery.
它们非常擅长向广大受众揭示信息。
They're really good at, you know, exposing information to broad audiences.
它们擅长帮助提供上下文并推理可能复杂的科学问题。
They're good at helping give contextual and reasoning through what might be complex scientific problems.
但当你想要进行优化，认为自己已经接近目标时，实际上有比大语言模型更好的工具来进行高通量模拟。
But when you wanna go to an optimization where you think you have something that's pretty close, there's actually much better tools than an LLM than to do high throughput simulation.
在这种情况下，与该客户合作时，他们在进入实验室前，用我们的软件在几分钟内运行了数十万次模拟。
And so in this case, with that customer, they were running hundreds of thousands of simulations in minutes of time before they went into the lab with our software.
然后他们能够说：嘿，在十万次或百万次模拟中，我认为这两个实验最有可能成功，这才是价值所在。
And then they were able to go there and say, hey, out of a 10,000 or a 100,000 simulations, I think that these two experiments are the most likely to be successful, and that's where the value comes in.
因此，关键在于为正确的工作找到合适的工具。
So it's really about finding the right tool for the right job.
我认为在当今世界，如果大语言模型是一把锤子，那么一切看起来都像钉子，但事实并非如此。
And I think in today's world, if an LLM's a hammer, everything looks like a nail, and that's not necessarily the case.
完全正确。
A 100%.
他们说，你可以将一切进行标记化。
Well, they say you can tokenize everything.
对吧？
Right?
你可以标记分子，但但但但如果你需要一个万亿参数的模型，这真的是最实际的方法吗？
You can tokenize molecules, molecules, but but but but is is that that really really the most practical way to do it if you have to have like trillion trillion parameter?
没错。
Exactly.
是啊，怎么说呢。
How yeah.
因为一个拥有1500万种分子结构的模型，那应该不是个大语言模型，我猜。
Like, because you're a 15,000,000 molecular structure model, that's not an LLM, I'm assuming.
它里面确实有一个Transformer，但同时上面还叠加了一个神经网络。
It's it there's a transformer sitting inside of it, but it's it's also there's a neural net that sits on top of that as well.
此外，我们还有其他方式来向量化化学结构，也就是如何将化学结构编码到一个潜在空间中，从而代表这种化学特性？
So there's and there's other ways that we basically vectorize what that how do you encode a chemical structure into a latent space that basically represents what that chemistry is?
但说实话，这是一个巨大的数据集。
But, honestly, that's a huge dataset.
1500万个化学结构，这涵盖了全部公开知识。
15,000,000 chemical structures, that's the entire public knowledge.
我们刚才提到的那家公司，从三个月缩短到两天，他们基于30个历史实验来进行推荐模拟。
That company we were just talking about that went from three three months to two days, they had 30 historical experiments that we were basing that recommendation simulation on.
因此，你需要处理这种小数据，并仍能提供可信的模拟实验结果。
So this is small data that you have to be able to take and still give credible simulated experiments for.
所以你们自己构建了基础模型。
So you guys built your own foundational models.
模型。
Models.
对吗？
Is that correct?
而不是基于其他现有的模型？
Rather than building off of another?
是的。
Yeah.
所以我们使用了所有现有的基础大语言模型。
So we use all the foundational LLM models out there.
我们并不从事构建大语言基础模型的业务，而是专注于化学领域的专用基础模型。
We're not in the business of building LLM foundational models, but when we're in the business of doing chemistry domain specific foundational models.
因此，这正是你可以将这些内容结合起来的地方。
And so that's, again, where you get to pair these things up.
你知道，我认为这个问题之所以一直没有被解决，是因为还没有人集中精力去针对某一特定领域攻克它。
It's, you know, think that a lot of the reason that this problem hasn't been tackled is because there hasn't been this like concerted effort to go solve it for one domain.
你知道，已经有很多人说，‘科学领域，我们来试试用人工智能解决科学问题吧。’
You know, think there's been a lot of people that are like, hey, science, let's go try to solve, you know, AI for science.
但即使‘人工智能用于科学’这个目标对我来说也太宽泛了。
And even that to me was too broad.
我们只是专注于化学和材料领域。
Like, we are just doing chemistry and materials.
所以生物技术、制药领域，我们会稍微涉足一下这些领域。
So biotech, pharma, know, then we might dabble a little bit in those types of areas.
对吧？
Right?
如果它们旁边有类似的东西的话。
If there's like a a butt up next to them.
但大部分情况下，这些是不同的。
But for the most part, those are different.
比如，蛋白质序列与小分子截然不同，而小分子又与DNA序列大不相同。
Like, a protein sequence is much different than a small molecule, which is much different than a DNA sequence.
所以如果你试图构建一个覆盖所有这些领域的通用AI，就会像把花生酱抹得太薄一样，无法发挥数据真正能带来的价值。
And so if you try to build some generic AI across all of it, you end up spreading the peanut butter, so to speak, and you don't get the value that the data actually can provide.
我认为可能会有一个时候——我们已经在其他行业看到这种情况了——行业领导者会联手打造这种高度专精的基础性方法。
Do think there might be a time where I assume and we're seeing this in other industries as well, where industry leaders are coming together and building this foundational approach that's hyper specific.
我推测这种事在生物制药领域也会发生，或者已经在发生了。
And I would assume that's going to happen or is happening in biopharma.
这在物理学中正在发生，并且也将继续发生。
That will be and is happening in physics.
你觉得将来让这些模型能够相互交流会有好处吗？
Do you feel like there is a time where having those models all able to talk to each other could be beneficial in the future?
我
I
希望如此，老兄。
hope so, man.
我希望如此。
I hope so.
我讨厌说这个
That's I hate to
对不起。
I'm sorry.
我刚才
I went
有点跑题了
a little off the deep
但确实如此，是的。
end there, but it was, yeah.
还没问过。
Hadn't asked.
既然我们已经跑题了，作为这部分内容，你对云实验室有什么看法？
Since we're talking off the deep end, as part of this, what what are your thoughts on Cloud Labs as well?
我很好奇。
I'm curious.
比如，完全自主地进行所有实验，就是这样。
Like, the idea of, like, doing all the experiments, like, completely autonomously, like, yeah.
你对此有什么想法
Just what are your thoughts
呢？
on that?
所以我会试着，也许我会尝试同时回答这两个问题。我希望这个行业，我们所服务的这个行业，多年来一直缺乏紧迫感。
So I'll try to, maybe I'll try to address both of So I hope that the industry, the industry has not had a lot of urgency for many, many years, this industry that we serve.
但现在情况正在改变。
And now that's changing.
我们开始看到紧迫感的出现，这是由于利润率的侵蚀，以及来自世界其他地区的竞争压力。
We're starting to see urgency because of margin, margin erosion, because of competitive pressures in other regions in the world.
而且，现在出现了其他公司，它们通过全自动、高通量的实验室工作与传统的化学行业展开竞争。
And because there's other companies now that are popping up that are doing a fully autonomous, laboratory work with high throughput lab testing that are competing with the more traditional chemical industry.
因此，我认为作为解决这一问题的一部分，像Albert这样的公司正在发挥作用，帮助这些公司稍微跃入未来，或者从过去跃入现在。
And so, I think as part of the solve to that, you know, companies like Albert are playing a role to help these companies jump into the future a little bit, or jump maybe into the present from the past.
因此，云实验室在这里将发挥重要作用，但它永远不会取代实际的实验结果。
And so, you know, there's going be a big role to play in Cloud Labs there, but it will never replace the actual experimental result.
我认为，这就是科学的美妙之处。
That's I think is the beauty of science.
永远不会出现一个完全可以通过数字方式做科学的世界，我认为是这样，或者我们离那个时代太远了，以至于谈论它已经无关紧要。
There will never be a world where you will just purely be able to do science digitally, I think, or maybe we're so far away that it doesn't matter about talking about that.
我认为，你将能够以数字化方式完成95%的失败实验，希望如此。
I think that you're going to do 95% of the failed experiments, hopefully digitally.
而你真正去实际运行的实验，是那些具有我之前提到的信息密度的实验——它们可能不是正确答案，但至少能测试一个你完全没有历史经验的领域。
And then the ones that you actually go run are the ones that have that information density I was talking about that may not be the right answer, but they're at least going to test a space that you have no historical knowledge around.
我喜欢给人们举个例子，以便让大家理解这个问题的难度有多高。
And I like to give people an example, just to kind of like put in perspective how challenging this problem is.
如果我今天给你们一个任务，比如说：你们有100种不同的原料。
If I give you guys a task today, and I was like, okay, you've got a 100 different ingredients.
我要你们设计一个蛋糕的配方。
I want you to make a recipe to make a cake.
而我只要求你们从这些原料中挑选出10种。
And all I want you to do is pick a set of those 10 ingredients.
不用担心每种原料的比例和用量。
So don't worry about the ratios and how much each one has to be.
你们觉得，从100种原料中选出10种，能组成多少种不同的组合？
How many unique sets of 10 ingredients out of a 100 do you think that there are that you could come up with?
数量级上是百万亿分之一。
On the order of MAGA, it's like one to the one hundredth.
对吧？
Right?
是17万亿。
It's it's 17,000,000,000,000.
而且
And
这就是原因。
that's why.
不。
No.
所以我完全猜错了。
So I'm way off.
是的。
Yeah.
所以，仅仅10种成分就有17万亿种组合。
So it's it's it's 17,000,000,000,000 combinations of just 10 ingredients.
现在，如果你考虑这些成分的比例，那么这个数字还会变得更大。
Now, if you think about what the ratios of those ingredients should be, then that blows up even bigger.
当然，作为科学家，你可以选择的成分远不止10种。
And of course, there's way more than 10 ingredients that you can choose from as a scientist.
因此，如果没有相关领域的知识，想通过暴力穷举来解决这个问题，根本是不可能的。
And so the idea that you could even brute force this problem without having domain knowledge around it, it's just impossible.
化学可能的组合数量比宇宙中的原子总数还要多。
There's more possible combinations of chemistry than there are atoms in the universe.
所以，这是一个非常有趣的问题，因为它与其他领域截然不同——在那些领域，你可以探索每一种可能的路径，最终找到一个明确的最佳答案。
And so that's a super fun problem to work on because it's much different than other domains where you can kind of explore every possible route and just come up with this is the definitive best answer.
有没有打算保存那些目前不符合实验标准、成功率较低的组合？也许它们低于某个阈值，但未来仍可通过AI进一步研究，看看是否仍存在潜在的机会，尽管可能性较小？
There any intention to save those that maybe don't make the cut for experimentation right now, but maybe have a lower chance of success a little bit that maybe is whatever's below the threshold and and still maybe have AI research those in some way in the future still to see if there aren't underlying opportunities that while less likely might still exist?
是的。
Yeah.
我认为这就是科学的美妙之处。
I think that's that's the beauty of science.
对吧？
Right?
因为一次失败的实验，明天可能就成为成功的实验。
Because one failed experiment could be a successful experiment tomorrow.
需求会变化。
The requirements change.
所以，是的，所有这些数据都有价值。
And so, yeah, there's val there's value in all that data.
关键在于，如果你要走向现实世界，走向实验台去进行实验，你希望这个实验是针对今天问题最有价值的。
The key is if you're gonna go put, if you're gonna go into the real world, you're gonna go into the benchtop and run that experiment, you want that to be the most valuable experiment for today's problem.
当然。
For sure.
没错。
Right.
当然。
For sure.
对。
Right.
百分之百。
A 100%.
这说得通。
That makes sense.
嗯，这让我想到了。
Well, that that makes me think.
如果我们讨论的是一个可能有17万亿个潜在研究方向的问题集。
So if we're talking about a problem set that could potentially have 17,000,000,000,000 possible, you know, areas to go.
那么，你是在和OpenAI竞争计算资源吗？
What type of like, are are you competing with OpenAI for compute here?
你们需要多少计算资源来实现这个目标？
Like, how much compute do you need for your Yeah.
对于这种东西？
For this type of stuff?
对，没错。
Like yeah.
嗯。
Yeah.
我认为你最好尽可能避免玩这种游戏。
I think you wanna try to not play that game as much as possible.
对吧？
Right?
我认为你会看到，现在我要谈一些我未必有资格谈的事情。
And and I think that you're gonna see, and now I'm talking about stuff that I don't have a right to necessarily talk about.
但我认为，即使是基础大语言模型提供商，也必须遵循物理定律。
But I think even with the foundational LLM providers, like, there is the laws of physics that you have to run.
你不可能拥有一个和地球一样大的数据中心。
Like, you cannot have a data center the the size of the planet Earth.
展开剩余字幕（还有 377 条）
对吧？
Right?
所以最终你必须想清楚，好吧，我们该如何提高这些算法的运行效率，以便能够直接处理——有17万亿种组合。
And so at some point you have to figure out, okay, like how do we get more efficient about how we run some of these algorithms so that we can just okay, there's 17,000,000,000,000 combinations.
你如何在不模拟它们的情况下，直接排除其中16.9万亿种组合？
How do you exclude 16,900,000,000,000 of those right off the bat without having to simulate them?
这样你就可以专注于可能仅剩的十亿种。
So then you can focus on maybe just a billion.
这就是科学领域知识发挥作用的地方。
That is where scientific domain knowledge comes in.
这就是人类介入问题的地方。
That's where the human comes into the problem.
他们的作用是说：好吧，这是一个无限的空间。
Their role is to say, okay, this is the infinite space.
我知道，如果我在配制油漆，它至少需要含有50%的水。
I know if I'm making a paint, it needs to have at least 50% water.
好的。
Okay.
就在那里，你已经将这个问题的范围约束得比任何可能的组合都要小得多。
Right there, you've now constrained that problem much more than any possible combination.
我知道它需要具备这些特性，而且不能有100种成分，因为我无法合理地生产那么多。
And I know it needs to have these type of attributes and it can't have a 100 ingredients because I can't manufacture that reasonably.
因此，科学家会介入，基于现实世界的约束来构建限制条件，并开始设定这些约束。
So the scientist comes in with a real world, like, constraint building and they start to put constraints on it.
这正是你首先更快解决问题的方式。
And that's how you, first of all, solve the problem sooner.
这也是减少计算量、避免用暴力方法解决这个问题的方式。
It's also how you reduce the amount of compute to go like brute force this problem.
我认为这引出了我们经常被问到的问题：你们是在取代科学家吗？
And I think it comes into, you know, the question we get all the time is, you know, are we replacing scientists?
当然，我们正试图取代科学家在过程中所执行的某一部分工作。
For sure, we're trying to replace a piece of the scientific process that they're going through.
我们是要取代整个科学家吗？
Are we trying to replace the entire scientist?
绝对不是。
Absolutely not.
因为科学家仍然需要进来，为AI如何解决这个问题设定界限。
Because the scientists still needs to come in and give those guardrails for how the AI should go approach the problem.
是的。
Yeah.
嗯。
Well.
我们也差不多达成了这个共识。
We sort of landed on that too.
对吧，科里？
Right, Corey?
看起来被取代的是具体任务，而不是整个工作。
Where it seems like it's it's specific tasks that are getting replaced, not jobs.
目前，没有任何人工智能能够取代任何一个单独的工作。
At this point, no no AI can replace any single job at this point.
是的。
Yeah.
我不确定是否能这么绝对地说。
I don't know that I would say it in an absolute like that.
说得有道理。
Fair enough.
是的。
Yeah.
说得有道理。
Fair enough.
我我我得说，我会说得更大胆一点。
I I I've I I will make more bold.
绝对地说。
In an absolute.
是的。
Yeah.
这公平。
That's fair.
这似乎不太可能。
It seems unlikely.
总体方法。
General approach.
是的。
Yeah.
说实话，这正是我们所看到的。
Mean Honestly, that is what we're seeing.
如果我们正朝着一种巨大繁荣的方向发展，也就是说，我们开始自动化这些任务，那么在某个时刻，我们可能会达到每个人的工作都被完全取代的境地。
If we're headed to a place of, like, huge amounts of prosperity, right, where we're starting to automate these tasks, you know, at some point we may get to a point where everyone's job has fully been replaced.
所以问题不再是，未来某个时候，这是否是不可避免的最终状态？
So then the question isn't, you know, are, is that the inevitable end state at some point in the future?
问题是，最后剩下的工作会是什么？
The question is like, what's the last job it's going to be out there, right?
我认为科学家和一些类似领域的职业将会是最后剩下的工作，因为这些角色非常具有挑战性，很难完全实现自动化。
I think like scientists and some of those types of fields, those are gonna be some of the last jobs that are out there because they're they're very challenging roles to be in to just fully autonomous, fully automize the entire process.
所以，如果我是一个科学家在听这场对话，我今天不会太担心。
And so I think if I was a scientist listening to this conversation, I wouldn't worry too much about it today.
对于这一代人来说，我认为我们会有很多酷炫的工具可以使用，而不是会被完全取代。
These generations, I think we're gonna have a lot of cool tools to play with and not something that's just gonna completely replace it.
是的。
Yeah.
那你怎么看待OpenAI的立场？他们现在表示，到2026年，他们有信心我们会拥有能够做出类似——我忘了他们确切的说法——但像是有意义的、小规模的
Well, what are your thoughts on OpenAI's stance that they're basically saying now in 2026, they think that they're gonna have they're confident that we're gonna have AI that can make, like, I forget the exact phrasing that they said, but, like, like, contri like, meaningful, small,
复杂的、新颖的
complicated Novel
科学突破的AI
scientific breakthroughs
或者什么？
or something?
或者，是的。
Or Yeah.
但在2027年或2028年，他们认为AI将对新颖的突破性成果产生重大影响。
But then in 2028 or 2027, they're thinking that it'll have significant impact on novel, breakthroughs.
你有什么想法？
What what are your thought?
你觉得这些时间线和你所看到的相符吗？
Do you think that those timelines make sense with what you've seen?
你怎么看？
What are your thoughts?
我认为，如果你把技术进步画在一张坐标图上，横轴是时间，纵轴是进展，你会发现，嗯。
I think that if you look at the progress of technology on a x y axis, where x is time and and y is progress, you know, you see that Mhmm.
那条曲线，那个指数级的上升趋势。
That curve, that exponential ramp.
所以，如果你从这个角度来看，你会说，天啊，短短五年，我们就要直接跃升到下一个阶段了。
And so if you look at it from that perspective, you say, well, shoot, two five years, my god, we're gonna immediately jump to this next stage.
但如果你仔细观察这张图表，它根本不像平滑的曲线。
If you zoom in though on that chart, it doesn't look smooth at all.
它看起来像是一段段平缓，然后突然大幅跃升，接着可能是缓慢进步，再迎来另一次跃升。
It looks like a bunch of kind of flat, and then a huge step change, and then maybe a slow progress, and then another step change.
因为像Transformer这样的技术，正是那种跃升，对吧？
Because things like transformers, right, were that step change, right?
五年前，如果我们根据当时技术进步的速度来预测，能想到今天我们达到的水平吗？
Five years ago, would we have predicted where we are today if we had looked at the rate of technology improvement?
我们根本想不到。
We wouldn't have.
实际上，我们现在可能远远超出了之前的预期。
We actually are probably way ahead of where we were.
所以，但我也认为，假设我们今后会以当前的速度永远持续进步，这种想法也是错误的，对吧？
So then, but I think it's also false to say if you assume that we're gonna continue to progress at the current rate forever, right?
必须要有另一个飞跃。
There has to be another step change.
因此还会出现更多的飞跃。
And so there will be more step changes.
会在2028年前的两年内发生吗？
Does it happen in two years by 2028?
我不知道。
I have no idea.
我认为没人知道。
I think nobody has any idea.
如果他们知道，今天就已经做到了，我们早就拥有了。
If they did, they would have done it today and we would have already had it.
对吧？
Right?
是的。
Yeah.
说得通。
Fair enough.
所以，我对这个问题没有答案。
And so I don't know the answer to that question.
在某个时刻，由于人工智能，将会出现极其重要的突破，但我认为你不应该贬低那些在控制、约束和引导人工智能的人类。
At some point, there will be incredibly meaningful, you know, advances because of AI, but I think that you shouldn't discredit the humans that are controlling it, constraining it, guiding it, that type of stuff.
我认为你会看到越来越多的诺贝尔化学奖由人类与模型共同获得，而不是仅由其中一方获得。
And I think that you're gonna see more and more, you know, Nobel Prizes in chemistry won by humans plus models, versus just one or just the other.
你会在论文上把模型列出来吗？
Would you would you list the model on the paper?
你会说，比如，由我和Grock共同撰写，或者由我撰写吗？
Would you say, like, and and authored by, you know, me and Grock or me.
不错。
Nice.
是的。
Yeah.
为什么不呢？
Why not?
我觉得是的。
I think so.
该给谁的荣誉就给谁。
Credit where credit's due.
对吧？
Right?
嗯。
Yeah.
嗯。
Yeah.
嗯。
Yeah.
嗯。
Yeah.
我也这么认为。
I think so too.
而且我认为这也能体现可信度，因为两年后，我们可能想记住当时用的是哪个模型，以及了解它的局限性。
And and I think it would speak to the credibility too because, you know, two years down the road, we might wanna remember what model that was and understand what its limitations might have been.
而且可能仍然需要进一步的研究和重复验证。
And maybe it still requires, you know, further study and repetition.
是的。
Yeah.
我觉得这是个很好的观点。
I think that's that's a good point.
我的意思是，你得注明出处。
Mean, you gotta you gotta cite your sources.
对吧？
Right?
而且这些AI基础模型正逐渐成为一种信息来源。
And and these AI foundational models are becoming a source.
你知道，这也是我们公司长期愿景的一部分，我希望看到我们能为人类知识体系做出重大贡献。
You know, that's that's part of our our long term vision also as a company, right, is I would love to see us contribute significantly into the the, you know, human knowledge sphere.
但我们最终是通过客户来实现这一点的。
But we're gonna do that ultimately through our customers.
你知道，我们的客户才是推动这些创新的人，我们只是试图用最好的工具来赋能他们，我认为这使我们在行业中成为了一个倍增器，而这正是我做这件事的根本原因——这是成为倍增器的最佳方式。
You know, we our customers are the ones that are gonna that make those innovations, and we're just gonna try to empower them with the best tools possible, which I think makes us a multiplier in the industry, which is that's fundamentally why I am doing this is it's the best place to to be a multiplier.
那么，你会说像Albert Invents这样的模型是具有代理性的吗？
So do you would you say that the, like, Albert Invents models are agentic?
也就是说，你们系统的任何部分具有代理性吗？
Like, does any part of your system agentic?
人们是如何与这些模型互动的？你是怎么看待的？
Like, how how do you how do you think of the way that people are working with that?
他们是把这当作一个科学合作伙伴，还是仅仅把它当作一个特定工具？
Are they working with that as a co scientist, or is it just specific tool?
你怎么看待这个问题？
How do you look at it?
是的。
Yeah.
所以我们把技术栈分为三个不同的组成部分。
So we look at our tech stack as three different components.
我们之前提到过的那个记录系统。
We've got that system of record that we were kinda talking about before.
我们如何用一个能真正记录你工作流程的工具来替代纸质笔记本？
How do we replace the paper notebook with something that you actually collect your workflow in?
然后我们再看构建在其上的记录系统。
Then we look at our system of record that layers on top of that.
而机器学习正是在这里发挥作用。
And that's really where like the machine learning comes in.
规则引擎也是在这里应用。
That's where the rule engines come in.
就像我说的，更传统但仍然复杂的软件工程是在此基础上应用的。
Like, I'd say the more traditional, but still complex, you know, software engineering gets applied on top.
像监管信息这样的内容，你并不希望一个代理系统去猜测，比如，你打算放在客机上并运输的物品的下一个词可能是什么。
And certain things like regulatory information, you don't want an an agentic system guessing, you know, or guessing what the next token could be for, like, something that you're gonna put on a passenger plane and ship around.
这是一个明确的规则。
That's a discrete rule.
你必须要有这个规则。
You need to have that.
然后我们有我们的工作系统，这是最顶层的第三层。
Then we have our system of work, which is that third layer that's on the top.
在这里，你拥有基于检索增强的大型语言模型和位于其上的代理网络。
And that's where you've got the Rag LLM, the the agentic network that sits on top of it.
在这里，我们开始实现某些工作环节的完全自动化。
And that's where we're starting to do like fully automate, you know, certain parts of the work.
那么，人们是否愿意手动将数据输入系统呢？
So does people, do people want to type data into a system?
没有人愿意手动将数据输入系统。
Nobody wants to type data into a system.
好的。
Okay.
所以你需要以某种方式将数据输入到Albert中。
So you've got get the data into Albert somehow.
我们如何部署一个代理来为你完成这个任务？
How do we put an agent there that maybe does that for you?
它可以从机器中获取数据并自动将其输入，或者从在线来源的PDF中提取数据，并以清晰、结构化的格式存入。
It takes it from the machine and automatically puts it in there, or it takes it from a PDF on some online source and puts it in there in a clean and structured format.
因此，我们的平台顶部有一个完整的代理层。
And so there's a whole agentic layer of our platform that sits on top.
我们的真正竞争优势在于，我们同时实现了这三个层级。
And our real competitive advantage, think, is that we're doing all three of those layers.
我们不仅仅是一个代理层。
We're not just an agentic layer.
我们拥有作为基础的系统记录，位于所有其他层级之下。
We have the system of record as the foundation under lease, underneath everything else.
他们的神经网络位于中间，
And they have the, the neural network kind of in the middle of
没错。
Exactly.
没错。
Exactly.
是的。
Yeah.
酷。
Cool.
不错。
Nice.
我有个关于知识产权数据的问题。
I have a question dealing with data around IP, for example.
我知道你提到过，它一直在从Kinvue的数据中学习。
I know you mentioned it's been learning from some of Kinvue's data.
那么，如何在不侵犯他们知识产权的情况下实现这一点呢？
How does that work without compromising their IP, for example?
我知道公司长期以来一直对这类东西，比如配方研究等，保护得非常严密。
I know companies are protective of that kind of stuff for ages and ages, formulation research and whatnot.
你们双方如何利用这些数据进行创新，同时确保保护客户宝贵的知识产权呢？
How do you both use that to innovate and ensure that at the same time you're protecting you know, valuable clients IP?
是的，这是一个每个客户都会问到的基础性问题，尤其是在当今这个时代。
Yeah, that's a foundational question that every single customer also asks us, especially in today's age.
我们的回答非常简单。
And our answer is super simple.
他们的数据就是他们的数据。
Their data is their data.
基于他们的数据构建的模型，就成为他们的模型。
The models that are built on their data become their models.
我们绝不会将这些数据与任何人共享。
We would never share that data with anybody else.
那么，我们能否从在线公开资源中抓取数据供所有人使用？
Now, could we scrape things from online public sources to use for everybody?
当然可以。
Absolutely.
我们正是用这种方式处理了那1500万个化学结构。
And that's what we did with the 15,000,000 chemical structures on that.
但一旦你将自己的数据加入其中并开始从中获得洞察，这些数据和洞察就属于你了，你拥有它们。
But the moment you add your data into it and you start getting insights out of it, that becomes yours and you own it.
我认为，如果没有这一点，我们根本无法做到我们现在所做的事情。
And I think, you know, we couldn't do what we were doing without that.
现在，回到你之前提过的一个问题：当公司希望彼此共享信息时，情况会怎样？
Now, back to an earlier question you had is like, what about when companies want to start sharing information with each other?
对吧？
Right?
我们非常乐意促成这一点，但这必须由他们主动来找我们，由行业主动向我们提出：嘿，我们希望更广泛地共享信息。
We would love to facilitate that, but that has to be them kind of coming to us and the industry coming to us saying, Hey, we want to share information more broadly.
我们还设置了限制机制，使得可能只有他们知识产权中非常小的一部分被决定为出于某些目的而公开。
And then we set up ways to constrain that as well, where maybe it's only a very small subsegment of their IP that they've decided to make public for x, y, and z purposes.
可能吧。
Probably
他们所制造的一切。
everything they've ever made.
是的。
Yeah.
没错。
Exactly.
但这必须是明确的自愿参与。
But that has to be a clear opt in.
我们不能有默认状态。
There can't be a default, status for us.
这说得通。
That makes sense.
是的。
Yeah.
没错。说到Kenview，你们现在在做些什么类型的工作？
That makes Speaking of Kenview, I mean, what type of stuff are you working on?
你们贡献过的东西，有没有已经被产品化了？
Are you like, has any of your stuff, that you've contributed to been been, productized yet?
或者怎么样
Or how's
是的。
it Yeah.
所以，Kenview是我们较早或最近建立的关系之一，我们就在大约一个半月前刚刚公开宣布了这个合作。
So, I mean, Kenview's, you know, one of the earlier or the more recent, you know, relationships that we have, we just publicly announced that just, think a month and a half ago or something at this point.
是的。
Yeah.
恭喜。
Congratulations.
恭喜。
Congrats.
谢谢。
Thank you.
是的。
Yeah.
他们是一家了不起的公司，生产着我们每天都在使用的产品。
They're, they're an amazing, amazing company, you know, doing, making products that we all use every day.
他们目前正在经历一次重大的数字化转型，而我们是其中的核心部分。
So they're in the middle of a big digital transformation and we're a core piece of that.
因此，我们现在正将Albert部署到他们整个设施中的每一个实验室。
And so, we are right now deploying Albert to every single laboratory inside of their entire facility.
到明年年中，所有来自Kenview的产品都将由Albert参与，我们将帮助这些科学家推动他们的创新。
By middle of next year, every product coming out of Kenview is gonna be touched by Albert, and we're gonna be helping all those scientists with their innovation.
所以，是的，我认为明年你去CVS或塔吉特购物时，可能会看到瓶身上写着“由Albert发明”，这真的很酷。
And so, yeah, I think, you know, you should next year be able to go to a CVS or a Target, and, you know, I wish they would go for it, but invented by Albert would be, you know, invented with Albert would be really cool to have on the bottle there.
所以，你想看到阿尔伯特的名字出现在获得诺贝尔奖的论文上吗？
So, do you wanna see Albert listed on, papers that win the Nobel Prize?
还是说这才是目标？
Or is that is that the goal?
那不是目标。
That's not the goal.
那会是一个不错的成果。
That would be a nice outcome.
对吧？
Right?
那不是
That's not
不过，你的感受呢，不是吗？
the your feelings, though, would it?
不是。
No.
我认为最有意义的事情是，当一位科学家主动联系我们，说：谢谢你们让我的生活更轻松了。
I think what would be the most meaningful thing is when a scientist, you know, reaches out to us and just says like, thanks for making my life easier.
对吧？
Right?
我曾经长期为论文和Excel文件所困扰，无法像我希望的那样高效工作。
I've been struggling for a long time with paper, Excel, haven't been able to move as fast as I as I think I could.
而现在你们让我的生活变得轻松了。
And now you guys are making my life easier.
我们经常收到这样的反馈。
And we have that type of feedback all the time.
这就像，我最初是为自己做的。
And that's like, I built this for myself.
我就是所有这些工具的第一个用户，当时是在肯的拖车里。
I was the first user of all of this, in, you know, the Ken's trailer.
所以我才开发了它，而能够与全世界分享它，这才是我们真正的使命。
And so that's why I built it and to be able to share that with the rest of the world, that's really what we're here for.
我们想问一两个关于你们商业模式的问题。
Let's, like to ask a question or two about kinda your business model.
你们本质上并不是在销售某个特定模型的访问权限？
Are you essentially so you're not selling access to a specific model.
你们是在帮助他们打造适合他们的模型，对吧？
You're helping them craft a model that is correct for them, essentially?
这么说对吗？
Is that correct?
首先，我们帮助他们管理数据，因为没有数据，我们就无从下手。
So first, we help them manage their data because without data, we can't do anything.
所以这成了我们的首要目标。
So that becomes, like, objective number one.
然后我们帮助他们的科学家每天为自己构建模型。
And then we help them, and we're really helping their scientists build models for themselves every single day.
在过去十二个月里，我想我们平台上运行的模拟实验大约有1.2亿次。
And so, you know, we've had a last in the last twelve months, I think we've had something like a 120,000,000 simulated experiments running through our platform.
这可不是小事。
And so that's nontrivial.
对吧？
Right?
有这么多模拟实验在运行。
That's a lot of simulation that's being run.
未来几年，这个数字将达到数十亿甚至上万亿。
It's gonna be in the billions and in the trillions in the years to come.
这些全部都是科学家们基于自己的数据以及所在机构的数据构建的模型，目的是避免进行大量不会带来良好结果的实验。
And so those are all models that scientists are building for themselves based on their data plus their organization's data to help not run so many experiments that wouldn't lead to good results.
哇。
Wow.
太棒了。
That's awesome.
我的意思是，我刚刚在通话前向科里坦白了。
I mean, I'm I'm personally so I I just, confessed to Corey right before we got on the call.
我确实逃了我的高中化学课。
I actually skipped my high school chemistry class.
天哪。
Oh, man.
我逃掉了。
I got out of it.
但我选了一门数字科学课，学到了很多现在对我作为成年人很有帮助的酷炫工具。
But it's because I did a digital science class, learned a lot of really cool tools that help me now as an adult.
但我确实逃了化学课。
But I did skip chemistry.
不过从那以后我学到了很多化学知识，因为我曾经为一家从事材料科学、致力于研发新型电池技术的公司撰稿。
But I've learned a lot about chemistry since then because I used to write for a company that's in the material sciences trying to invent new battery technology.
太棒了。
Awesome.
因此，电池材料是我最关注的领域之一，因为能量密度、寻找容易获取且不稀缺的材料——你知道，不能依赖任何一个国家供应这些材料——而且你还能在极小的体积内储存更多的能量，这对我来说就像是一个巨大的突破。如果我们在这方面取得重大进展，就能在其他领域实现指数级的进步，因为这能大幅提升我们能够储存的能量。
And so battery materials is one of the biggest things that I'm focused on because energy density and finding the right materials that are easy to source, not scarce, you know, not just you you don't become reliant on any one country for them, and you can also, you know, pack so much more power in a tiny form factor is is like that to me feels like one of the biggest unlocks that if we can make a lot of progress there, we can exponentially increase progress elsewhere, you know, because it just increases the amount that we can, you know, amount of power that we can store.
如果非要说的话，你在电池和材料领域，比如能源材料方面，现在在做些什么呢？
If anything, are you doing in the battery space and materials, like energy materials, I guess, let's say.
是的。
Yeah.
我的一些工作内容还不能公开，但我能肯定谈论的一个很好的公开例子是，有一家叫Chemours的公司，总部位于东海岸，他们在那儿有一个出色的电池实验室。
I mean, I some of this stuff isn't public, but one of the great public examples that I can definitely talk about is there's a company called Chemours, which is based in in the East Coast, and they they have an amazing battery lab out there.
他们正在电池领域进行一些令人惊叹的科学研究。
They're doing some incredible science around battery.
现在这些工作都通过Albert系统进行。
That's all now being run through Albert.
电池化学过程的一个很棒之处在于，这是一个数据密集型的过程。
And one of the great things about the battery chemistry process is it is a pretty data intense process.
对吧？
Right?
当你考虑对新型软包电池或纽扣电池等进行循环测试时，你会收集到海量的数据。
And you think about cycle testing of a new pouch cell or coin cell or whatever it may be, you're collecting a tremendous amount of data.
所以，纸质笔记本在这里根本行不通。
And so a paper notebook actually doesn't work there.
对吧？
Right?
因此你不能用它。
And that's you cannot use it.
所以你需要一些其他的工具来替代。
So you need some tool to use, instead.
他们取得了巨大的成功，是一家非常棒的公司。
And so they're they're having a a tremendous amount of success, fantastic company.
我们非常自豪能够支持他们，当然也支持更广泛的可持续发展和电气化努力。
And, you know, we we are really proud to be supporting, you know, them and then, of course, the larger sustainability and kind of electrification efforts that are happening beyond that.
化学和材料科学是一个庞大的产业，但普通人日常并不会想到它。
Like chemistry and material science is a massive industry, but not one the average person thinks about every day.
你能帮我们稍微理解一下这里的规模吗？
Can you kinda help us understand the understand the scale here a little bit?
比如，有哪些问题正在被更快地解决？这对消费者日常使用的产品意味着什么？
Like, what types of problems are being solved faster, and what does that mean for products consumers use every day?
是的。
Yeah.
人们想不到这一点很有趣，就连我这样天天身处其中的人，每周都会偶然发现一家我以前从没听说过的十亿美元营收公司。
It is funny that people don't think about it, and even somebody like myself who, like, lives in it, I there's, like, a billion dollar revenue company that I stumble across every week that I've never heard of before.
就是这样。
And it's like that
这就像在人工智能领域工作一样。
That's what it's like working in AI too.
没错。
Yes.
所以是十亿美元的营收，不是十亿美元的市值。
So a billion of revenue, not not billion of of market.
是的。
Yeah.
但情况并不是这样的。
And that's not what it's like.
是的。
Yeah.
不是市值。
Not market cap.
是的。
Yeah.
没错。
Yes.
化学的奇妙之处在于它是物理世界的基础。
And so it's it's the cool thing about chemistry is the foundation of the physical world.
所以，如果你们环顾一下现在所处的环境，我看到背景里有一把吉他。
So if you guys just look around where you are right now, I see a guitar in the background.
从琴弦的金属材料到塑料部件，制造这把吉他涉及了无数材料科学项目，没错。
The number of material science projects that went into making that from the material, the the metal that's of the string to the plastic Yeah.
吉他的涂层，还有我肯定有一些非常特殊的材料用于声学阻尼之类的，你知道的，声学……
Of the guitar, to the coating on top of it, to I'm sure there's actually some very special materials that are some sort of acoustic dampening or, you know, acoustic Oh,
是的。
Yeah.
你有琴枕，还有背面的所有塑料部件。
You know, you've got the nut, all of the plastics on the back.
所有这些东西。
All of that.
旋钮里的化学物质。
The chemicals in the knobs.
你会有
You would have
所有这些。
Everything.
电线。
The wiring.
电线本身。
The wiring itself.
你身上穿的衣服。
The the clothes on your on that you're wearing.
这些都是纤维。
These are fibers.
对吧？
Right?
这些是由某种合成棉或聚合物或其他材料纺成的纤维。
These spun fibers from some sort of synthetic cotton or maybe polymer or material.
所以物理世界中的所有东西都是化学。
So everything in the physical world is chemistry.
这意味着它也是最古老的行业之一，对吧？
And it's that means it's one of the oldest industries as well, right?
这是人类从事时间最长的活动之一。
It's one of the things that humans have been doing for the longest.
所以我认为，最好的理解方式是，每当你去商店，看着货架上的一切，对吧？这些东西几乎都是化学制品，也许那些两英寸乘四英寸的木板除外，它们纯粹来自木材，但其他所有东西都是。
And so I think, you know, best way to think about it is anytime you go to a store and you look at everything on the shelf, right, that's chem maybe not the two by fours, right, that were purely coming from a piece of wood, but everything else Yeah.
可能并不是简单的单一成分。
Is probably not just a simple, you know, single component.
因此这非常令人兴奋，因为这意味着我们正在帮助——我们常认为自己在帮助更快地创造物理世界，或者我们的客户正利用我们的技术来做这件事。
And so that's that's super exciting because then you're helping to, you know, we like to think of ourselves as helping to invent the physical world faster, or our customers are using our technology to do that.
所以，是的，世界并没有放缓。
And so, yeah, it's it's and and, you know, it's not like the world slowing down.
对吧？
Right?
我们都想要更好的产品。
We all want better products.
我们想要更便宜的产品。
We want cheaper products.
我们想要更可持续的产品。
We want more sustainable products.
这种来自消费者的压力会传导到像苹果和特斯拉这样制造这些产品的人身上，但他们立刻会转向供应链，说：帮我解决这个问题。
And that pressure from the consumers trickles to the people like the Apples and the Teslas who are making those products, but they immediately then turn around to their supply chain and say, go solve this for me.
这就是我们介入的地方。
And that's where we operate.
我们深入到这条供应链的内部运作。
We operate deep into that supply chain there.
是的。
Yeah.
嗯，我刚才本来就想提一下，比如以苹果为例。
Well, that's what that's what I was gonna bring up a second ago was that, like, for let's look at Apple, for example.
他们的限制因素实际上是电池寿命和材料性能。
Their limiting factor is actually battery battery life on their their materials.
所以，他们想做这个，已经公开泄露了。
So, like, they wanted to do this is publicly leaked and everything.
他们早在三年前就想做AR眼镜了，没错。
They wanted to do AR glasses, like, three years ago Yep.
或者去年发布，但它们受限于实际的供电能力。
Or or last year, release them, but they were limited by the the actual, like, power to power them.
受限于化学技术。
By the chemistry.
是的。
Yeah.
化学技术是瓶颈。
Chemistry was the limiter.
没错。
Yep.
没错。
Yep.
所以这就是我关注的焦点，你知道的，电池寿命和能量，因为这似乎是许多技术的瓶颈。
So that that's where I focus on, you know, battery life and and energy in particular is because that seems like that's a limiting factor for so much technology.
我认为这是几乎所有事物的瓶颈。
I think it's a limiting factor for almost everything.
对吧？
Right?
我的意思是，埃隆想登陆火星。
I mean, like, Elon wants to go to Mars.
很棒。
Great.
对吧？
Right?
那是一个非常酷且鼓舞人心的使命。
That's that's a super cool inspiring mission.
对吧？
Right?
多星球生存。
Multiplanetary.
这可不是软件问题。
That is not a software problem.
我们非常清楚如何发射火箭并将其送达目的地。
We very much know how to launch a point a rocket and get it there.
在很大程度上，这是一个纯粹的材料问题。
It is a pure materials problem for the most part.
对。
Right.
你如何制造一种烧蚀材料，能够承受进入火星大气层或返回地球大气层时的极端条件？
How do you make an ablative material that can withstand the, you know, coming in through either the Martian atmosphere or returning back into The US atmosphere?
你如何让钢材具备承受太空中的温度和压力，同时也能应对地球表面环境的性能？
How do you have, the properties of steel that can withstand, you know, the temperatures or pressures of space plus, you know, on, on the planet earth?
就像那样，那边正发生着大量的创新，对吧？
Like that's what that's, there's a huge amount of innovation happening right over there, right?
仅仅围绕化学和材料来解决这个问题。
Just around chemistry and materials to go solve that problem.
我们必须更快地解决这些问题。
We have to solve these problems faster.
如果我们想在有生之年完成我们所有人都希望实现的目标，就不能花十年、二十年、三十年来解决这些问题。
We can't take ten, twenty, thirty years to solve these problems if we want to get done what I think all of us want to in our, in our lifetimes.
关于这一点，现在是不是一个数据问题？
To your point on that, is it, is it a data problem at this point?
这是一个计算问题吗？
Is it a compute problem?
那么，真正限制我们解决这些问题的因素到底是什么？
Like, what is the limiting factor actually to, to, to solve a lot of this stuff?
我认为，大部分情况下这是一个数据问题。
I think it is a, a data problem for the most part.
人们无法有效利用他们的知识和数据。
People can't leverage their knowledge and their data.
一旦你开始解决这个问题，接下来就会变成一个物理定律的问题：如何像我之前说的那样，尽可能快地收集正确的数据和进行下一个实验。
And then once you start to solve that problem, then it becomes a, like a laws of physics problem of how do you, like, kind of, like I was saying before, you know, collect the right data, the right next experiment as quickly as possible.
这时候，高通量实验室自动化测试就派上用场了。
And there's where high throughput laboratory automated testing comes in.
人们正在那里发明各种技术，以帮助解决我们想要捕获的下一组数据问题。
There's all sorts of technologies that people are inventing there to help to solve the next set of data we wanna capture.
但如果我们连今天的数据都未能很好地捕获，那就没有任何依据来指导我们下一步该做什么。
But if we're not capturing today's data very well, you know, there's nothing to inform us of what we wanna do next.
好吧，我这里还有一个完整的问题要问你。
Well, I got one last full question for you here.
如果Albert如你所愿那样运作，那么在物理世界中，哪些今天不可能实现的事情将成为可能？
If if Albert works exactly as you hope, what becomes possible in the physical world that isn't possible today?
在未来的十年里，我们可能会看到哪些产品或材料，而这些在今天看来似乎是不可能的？
What products or materials might we see in, you know, another decade that maybe seem impossible now?
我将通过给出一个略有不同的答案来回应这个问题。
I'm gonna answer that question by giving a slightly different answer.
我认为
I think that
这很公平。
That's fair.
这个这个
The the
我刚刚确实让你预测未来了。
I did just ask you to predict the future.
所以
So
嗯。
yeah.
嗯。
Yeah.
什么
What
我希望能看到的是，在未来任何可能的时刻，你都能用笔记本电脑进行发明。
I would love to see in in the future, at at whatever point is is possible, is that you can invent with a laptop.
你可以用笔记本电脑发明物理世界。
You can invent the physical world with a laptop.
今天，我们可以在数字世界中很好地做到这一点。
Today, can do that in the digital world very well.
以前这是不可能的。
Before that wasn't possible.
二十年前，你不可能随便拿出笔记本电脑，登录AWS，就能创造出一个全新的奇妙产品。
Twenty years ago, you couldn't just take out your laptop, log into AWS and make some new amazing product.
你得有一台服务器，而且很可能你得在一家拥有大型服务器的公司工作。
You have a server and you'd probably work for a company that had access to a big server.
我希望在未来，无论你身处世界何处，只要你有一个关于如何发明实物的好点子，就能通过一次按键和一次点击实现它。
I would love in the future where regardless of where you are in the world, if you have a good idea of how to invent something physical, you should just be able to do that with a, a keystroke and a click.
这是一个令人鼓舞的世界，因为这样一来，你就把今天可能数以百万计的实物创新者，变成了我们数十亿人中的任何人都可以成为的人。
And that is an inspiring world to live in because then you've taken what may be millions of physical world innovators today, and you've made it so that anybody out of the billions of us can be.
那么，这些人会发明出什么样的产品呢？
And then what products those people invent?
我都不敢猜测，但这一定会非常令人兴奋。
I won't dare to to to guess, but it will be really exciting.
好吧，你想像一下，现在世界上有多少伟大的想法，正存在于那些完全不知道从何下手、该做什么的人脑子里。
Well, you know, imagine how many great ideas float around the world right now that are in the hand the heads of people who have absolutely no idea where to start or what
对这些想法不知所措。
to do with them.
没错。
Exactly.
你知道，这可能会让这些问题变得尤为突出。
You know, this this brings a lot of that to the to the forefront potentially.
是的。
Yeah.
是的。
Yeah.
这是一个值得向往的世界。
And it's it's a world that is worth is worth wanting.
对吧？
Right?
而且我认为，在当今世界这么多事情发生的情况下，你希望对未来感到兴奋。
And I think with all the other stuff happening in today's world, you wanna be excited about the future.
所以让我们努力创造一个我们都相信的令人兴奋的未来。
So it's let's let's try to make an exciting future that we all believe in.
Molekule 最初是一家三维打印公司，对吧？以前是这样。
Molekule started as a three d printing company, right, back in the day.
你对今天的三维打印有什么看法？
What are your thoughts on three d printing today?
我们谈到了云端实验室，但你觉得三维打印会在开发者使用的笔记本电脑中发挥作用吗？比如，你可以设计实验，在笔记本电脑上运行模型，然后在家打印出来进行测试？
We talk kind of about Cloud Labs, but like, is it what like, do you think three d printing is gonna play a role in in the laptop, like developer where, you know, you can, you know, design the experiment and run the models on your laptop, and then print it in your in your house to test it out?
你怎么看？
What what are your thoughts?
是的。
Yeah.
我认为可能不会在家打印出来。
I think all may may not print it in your house.
我认为那有点像一时的潮流，尤其是在工业类应用方面。
I think that was a little bit of a fad, you know, especially for like industrial type applications.
但3D打印是一项了不起的技术，关键在于如何将一个想法转化为一种数字形式的化学应用？
But, I mean, three d printing is an amazing technology to, again, how do you take an idea and use something that's a digital application of of of chemistry?
关键是找到合适的使用场景。
Just gotta find the right use case.
对吧？
Right?
用3D打印制作红色一次性纸杯可能不是个好主意，但为你的口腔和使用需求量身定制牙科设备，那就是绝佳的应用场景。
Making like red solo cups with three d printing is probably not a good use case, but making a dental device that's personalized to your mouth and your use case, incredibly good use case.
制作我们新想法的首个原型。
Making our first prototype of a new idea.
我不认为当今世界上任何被制造出来的产品，其研发过程中会完全没有3D打印的参与。
I don't think I I would doubt that there's really any prototypes anymore in the world of anything that's being made that three d printing isn't touching today.
随着任何技术获得更多投资，成本会下降，产量能够扩大，你就能开始玩转那个单位经济效益的游戏了。
And as any technology gets more investment, the costs go down, the volumes can start scaling, and you start to play that, you know, unit economics game there.
所以，3D打印在我心中占有非常特殊的位置，也希望它在未来物理世界中占据重要地位。
So three d printing has a has a very, near and dear place in my heart, and hopefully a big place in the future of the physical world as well.
太棒了。
Awesome.
酷。
Cool.
谢谢。
Thank you.
好了，尼克，非常感谢你今天做客我们的Neuron节目。
Well, Nick, thanks so much for joining us today on the neuron.
人们可以去哪里了解更多关于Albert、Vint以及你们那边正在进行的精彩工作呢？
Where can, people go to learn more about Albert and Vint and some of the amazing work you all are up to over there.
请访问我们的网站，albertinvent.com。
Check out our website, albertinvent.com.
我们网站上有大量优质的资源，而且我们现在正在疯狂招聘。
We got a bunch of great resources on there, and, we're also hiring like absolutely crazy right now.
我们有很多非常适合关心这类事情的人的有趣职位。
We got a lot of really interesting roles for people who care about this type of stuff.
所以，来和我们一起发明物理世界吧。
And, so come help us invent the physical world.
这是一个令人兴奋的时刻。
It's an exciting time.
我同意。
I agree.
此外，感谢所有今天观看的朋友们。
Well, to everyone watching, thanks so much to you for joining us today as well.
我们非常感谢你们。
We really appreciate you.
如果你们喜欢今天的节目，请点赞、订阅，并做所有帮助我们持续制作这些视频和邀请精彩嘉宾的事情。
If you enjoyed today's episode, please like, subscribe, and do all of the things that help make us keep being able to bring you these videos and these amazing guests.
同时，别忘了访问 neuron.ai，注册每日 Neuron 新闻通讯，加入每天早晨阅读它的 60 多万人行列。
Also, make sure you take a moment to pop by the neuron.ai and sign up for the daily neuron newsletter and join more than 600,000 people reading it every morning today.
但在下次之前，再见了，人类。
But until next time, farewell humans.
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。