第10集 - AI如何加速当今科学发现及未来展望

大家好，我是安德鲁·梅恩，这里是OpenAI播客。

Hello, I'm Andrew Main, and this is the OpenAI Podcast.

今天我的嘉宾是OpenAI科学部门负责人凯文·威尔，以及范德堡大学物理学教授兼OpenAI研究科学家亚历克斯·卢霍斯卡。

Today, my guests are Kevin Weil, Head of OpenAI for Science, and Alex Luchowska, who is an OpenAI Research Scientist and Professor of Physics at Vanderbilt University.

我们将探讨人工智能如何影响科学界、一篇即将发表的研究论文，以及未来五年科学发展的可能方向。

We're going to be discussing how AI is impacting science, an upcoming research paper, and where science may be headed in the next five years.

或许人们感受通用人工智能最深刻的方式就是通过科学。

Maybe the most profound way that people are going to feel AGI in their lives is through science.

有了ChatGPT，我可以同时向这个方向、那个方向以及其他多个方向展开研究。

With ChatGPT, I can just launch it in that direction, in that direction, that direction.

这些工具带来的加速效应将彻底改变科学研究。

The acceleration that is going to come from these tools is going to change science.

那么你正在负责'OpenAI科学计划'的推进工作。

So you're running the Opening for Science initiative.

能否请你介绍一下这个项目的具体情况？

Could you explain what that's about?

是的，Opening Eye for Science的使命是加速科学发展。

Yeah, the mission of Opening Eye for Science is to accelerate science.

所以问题是，我们能否帮助科学家们在五年内完成接下来大约25年的科学研究和发现。

So the question is, can we help scientists do the next, say, 25 of scientific research and scientific discovery in five years instead.

科学支撑着我们生活和工作的方方面面。

Science underpins so much of, you know, what we do and how we live.

如果我们能将最先进的模型交到世界顶尖科学家手中，从而加快科研进程，我们就应该这么做。

And if we can make progress go faster by putting our most advanced models into the hands of the best scientists in the world, we should do that.

这正是我们正在努力实现的。

That's what we're trying to do.

你可能会问，为什么是现在？

You could ask like, why now?

为什么我们一年前没有这样做？

Why didn't we do this a year ago?

为什么我们不等到一年后再做？

Why aren't we doing this a year from now?

其中一个重要原因是，我们刚刚开始看到我们的前沿AI模型能够进行新颖的科学探索。

One of the big reasons is we're just starting to see our frontier AI models being able to do novel science.

我们开始看到GPT-5能够实际证明新事物的案例。

So we're starting to see examples where GPT-five can actually prove new things.

也许还不是人类无法做到的事，而是人类尚未完成的事。

Maybe not yet things that humans could not do, things that humans have not done.

这些小小的存在证明表明，GPT-5能够突破人类知识的边界，进入未知领域。

So these little existence proofs of GPT-five being able to break out past the frontier of human knowledge and into the unknown.

如果说我在OpenAI这一年半左右的时间里学到了什么，那就是从模型完全不会做某事，到勉强能做某事，这个过程往往快得惊人。

And if there's one thing that I've learned from now, you know, a year and a half or so at OpenAI, it's that you go very quickly from the model can't do something to the model can just barely do something.

虽然目前还不太擅长，但你已经能看到这些早期的案例。

And it's not great at it yet, but you see these, these, these early examples.

然后，六个月或十二个月后，你会突然发现没有AI简直无法想象如何完成这些事。

And then, you know, six months later, twelve months later, all of a sudden you couldn't imagine doing this thing without AI.

我认为科学正处于这个初始阶段——我们看到使用AI的科学家正经历着真正的加速发展。

And I think science is in that initial phase where we're seeing real acceleration for scientists that are using AI.

有时是新颖的，你知道，可能还不是重大突破，可以称之为小突破。

Sometimes novel, you know, not yet maybe large breakthroughs, call them small breakthroughs.

这正说明这个领域潜力巨大。

And that just says that there's so much potential in this space.

我们已经看到一些例子，比如AI在数学证明方面的帮助。

We've seen examples of, let's say, AI helping with mathematical proofs.

你能举个例子说明它在其他领域如物理学中可能做的事情吗？或者短期内我们可能会看到什么样的进展？

Could you give me an example of how it might do things in some other areas like physics or whatever kind of things we might see in the short term?

是的。

Yeah.

我的意思是，我们每天都能看到例子，它们涵盖了各种科学前沿领域。

I mean, we're seeing examples every day and they're across the range of sort of the scientific frontier.

你可以在数学、物理、天文学、生命科学如生物学中看到这些例子。

You see examples in mathematics, in physics, astronomy, life sciences, like biology.

Alex，我是说，你曾经参与过其中一些工作。

Alex, I mean, you've worked on some of these.

也许现在是个好时机，来聊聊你见过的一些物理学方面的应用。

Maybe it's a good time to talk about some of the physics stuff that you've seen.

是的，我想回到凯文提到的这个特殊时期，我也有同感，因为我在2025年初时还觉得，ChatGPT虽然很酷。

Yeah, I think coming back to Kevin's point about how this is a special time, that's very much how I feel as well, because I started the year, 2025 thinking, yeah, ChatGPT is cool.

和所有人一样，它一推出我就用了，觉得是个很棒的聊天机器人，但我确信要等很久才能真正对我的工作产生实质影响。

Like everybody, I used it when it came out and I thought it's a great chatbot, but I was sure it would take a very long time before it would become really relevant for my own work.

所以年初时我可以说是个AI怀疑者，因为我习惯在看到证据后才被说服。

So I started the year, I would say, an AI skeptic, because I like to see evidence before I'm convinced of something.

后来我看到人们用它辅助写作，我也开始尝试这个用途。

And, I saw people using it to help in their writing and I started to use it for that as well.

它在校对方面非常有用，但我当时想，要让它涉足我真正专精的领域还需要很长时间。

It's very useful for proofreading, but I thought, oh, it's going to be a while before it gets to do the special stuff that I'm really a specialist at.

你是指黑洞研究这类领域对吧？

You're like black holes, right?

没错，就是黑洞物理学这类领域。

Like black hole physics, exactly.

今年早些时候我有过这样的经历：当时我正在尝试寻找一种描述脉冲星周围磁场的解，脉冲星是一种具有极强磁场的旋转恒星。

And I had this experience early this year where I was trying to find this magnetic field solution that describes what happens around a pulsar, which is a rotating star with very powerful magnetic fields.

我当时正致力于寻找这个非常特殊的解。

And I was going for this very particular solution.

我需要解一个偏微分方程。

I had to solve a partial differential equation.

我成功将该解识别为一种称为勒让德多项式的特殊函数乘积的无穷级数。

I was able to identify that solution as an infinite sum over products of special functions called Legendre polynomials.

如果你读过物理研究生院，这正是那种需要花大量时间熟悉的内容。

And if you go to physics grad school, this is the kind of thing that you spend a lot of time getting familiar with.

而且我也很喜欢这类难题。

And I also like these puzzles.

我在摆弄这个级数时，感觉它应该能化简为一个简单公式。

And I was playing around with the sum and I felt like there should be a simple formula that it evaluates to.

我当时想，好吧，我有个朋友有ChadGPT03Pro，虽然那时我自己还用不上。

And I thought, okay, I have this friend who has ChadGPT03Pro, which I didn't have access to at the time.

我想，好吧，我就把这个发给他，看看会有什么结果。

And I thought, okay, I'm just going to send it to him and see what comes out of it.

然后他给我发回了这个输出。

And he sends me back this output.

它思考了十一分钟，这在当时我从未见过，因为我用的是免费版本，思考时间没那么长。

It thought for eleven minutes, which at the time I'd never seen it do because I was using the free version, which doesn't think for as long.

它给出了一个绝妙的答案，能够理解这个求和式并将其分解成可处理的部分。

And it gave this beautiful answer where it was able to understand what the sum was and break it down into pieces that it could tackle.

接着它还得去找到这个特殊的恒等式，这个等式发表在1950年代挪威数学期刊的一篇论文里。

And then it had to go and find this special identity that was published in one paper from the 1950s in the Norwegian Journal of Mathematics.

所以它理解了问题所在，知道这个随机出现的恒等式正好适用，然后运用它们，给出了这个精彩的输出。

And so it understood what the problem was, and it knew about this random identity that was just the thing for the job, and it used them, and it gave this beautiful output.

最后，答案还是错的，因为它犯了个愚蠢的拼写错误。

At the end, the answer was wrong because it made the silly typo.

它在计算中多加了一个因子。

It added an extra factor in fraud.

这几乎就像人类在最后犯了个愚蠢的笔误，但验证推导过程却非常简单。

It was almost kind of like a human making a silly typo at the end, but it was very easy to check the derivation.

我仔细检查后意识到，这里多了一个因子。

And I went through it and I realized, okay, this is an extra factor.

但除此之外，它完成了所有工作。

But aside from that, it did the work.

这真的让我震惊，因为我觉得这不能算是人类独有的能力。

And that really sent me reeling because I thought, okay, I wouldn't say that's a uniquely human ability.

我曾以为这是理论物理学家特有的才能。

I thought that's something that makes theoretical physicists special.

你知道，在2025年的今天，它们显然能做到我认为不可思议的事情。

You know, now in 2025, clearly they're capable of doing things that I would consider amazing.

是啊。

Yeah.

最酷的一点是，就像亚历克斯的例子，虽然他自己最终也能完成，但GPT能更快做到。

And one of the cool things, so you got examples like Alex's where it was probably not something that he, like he could have done it himself over, you know, eventually, but GPT was able to do it faster.

这本身就是一种加速。

That's acceleration on its own.

而且这其中还有质的不同，因为如果你能在一小时内并行探索10条路径，而不是花一周时间探索两条路径，突然间你就能尝试更多想法，这也是一种加速。

And there's something qualitative about that even as well, because if you can explore instead of exploring two paths over the course of a week, if you can explore 10 paths in parallel in, you know, an hour, all of a sudden there's a lot more ideas that you can try, and that's also acceleration.

我们也在文献检索中看到类似的例子，虽然你可能不认为这是深刻的科学创新，但能够了解是否有人研究过这个问题确实非常重要。

We also see examples in like literature search, which you don't think of as maybe like deep scientific innovation, but it's really important to be able to understand, you know, has somebody worked on this problem before?

如果有的话，我是否能从中学习以加速自己的工作？

And if so, is there something I can learn to speed up my own work?

我们见过一个有趣的例子——我可能记错细节了——当时我们和一位研究人员交谈，他说他正在研究高维优化的某个特定想法。

So, and we've seen interesting examples where there was one, I might get the details of this wrong, but we were talking to this researcher and he was saying, he was exploring this particular idea in like high dimensional optimization.

他说：'老兄，我正在研究的这个东西很有趣，但肯定有人研究过这个。'

And he was like, man, you know, this thing I'm working on, it's interesting, but somebody must have worked on this before.

我不可能是第一个有这个想法的人。

I can't be the first person to have had this idea.

绝对不可能，但我就是找不到任何先例。

I just can't, but I can't find any examples.

然后他向GPT-5描述了自己的研究内容，GPT-5从经济学或其他完全不同的领域找到了一个例子，那个领域使用了完全不同的术语。

And then he had given it, he'd sort of given a description of what he was working on to GPT-five and GPT-five found an example from, I think it was like economics or something, a completely different field that used completely different terminology.

所以任何关键词检索都无法发现这个关联。

So no keyword lookup would have ever worked.

GPT-5进行的是概念层面的文献检索。

GPT-five did sort of a conceptual level literature search.

没错。

Yeah.

它找到了一篇德语的博士论文。

Found somebody's PhD thesis in German.

而且还是完全不同的语言，这篇论文基本上已被时间遗忘，但作者确实做过与他研究相关的有趣工作。

So also a completely different language, you know, it was like basically lost to time, but this person had done really interesting sort of related work that helped him in his research.

所以这又是另一个领域。

And so, you know, that's another area.

你可以讨论由新颖证明带来的加速，或者GPT-5独立或在专家指导下完成工作的能力，但还有这些计算加速和文献检索的例子，它们都在推动科学加速发展。

So you can talk about the acceleration that comes from just like novel proofs and GPT-five being able to do something on its own or guided by an expert, but there's also these examples of acceleration in calculations and literature search, all of them contribute to accelerating science.

是啊。

Yeah.

同样的事情也发生在我身上。

And the exact same thing happened to me.

当时我正试图推导黑洞的这个特性，得到了描述我所研究现象的方程，其中包含一个三阶导数项，这相当罕见。

I was trying to derive this property of black holes, and I got this equation that described this phenomenon I was after, and it had a three derivative term, which is pretty unusual.

我看着它，认出这是数学中出现的特殊概念——施瓦茨导数。

And I looked at it and I recognized it's something called the Schwarzene derivative, which is a special thing that appears in math.

我当时想，哇，这居然会出现真是太奇怪了。

I thought, wow, this is really strange that this would show up.

我就直接把方程复制粘贴到ChadGPT里问：你见过这个吗？

And I just copy pasted the equation into ChadGPT and said, have you seen this before?

它回答说：哦，见过，这是共形桥方程。

And it said, oh, yes, this is the conformal bridge equation.

当时我完全不知道共形桥是什么。

I had no idea what a conformal bridge was at the time.

它说，哦，去查查这篇论文。

And it said, Oh, just look up this paper.

这太神奇了，因为事实证明，我工作中出现的这个方程已经在其他研究中被探讨过了。

And that was amazing because it turns out that this equation that showed up in my work had already been studied in some other works.

我听很多从事物理研究的同事说，这种情况经常发生。

And I've heard from a lot of colleagues doing research in physics that there's a lot of that going on.

在知识的最前沿，一切都变得如此细分，以至于很难了解相邻领域的最新细节，GPT在这方面提供了极大的帮助。

And at the forefront of knowledge, everything becomes so niche that it's very hard to know the latest details in neighboring fields, GPT is an amazing help with that.

是啊。

Yeah.

这也是我们采访的教授和研究人员提到的另一点——如今你必须非常专业化。

That's another thing that we've heard from, professors, researchers that we've talked to is there's so much, you have to be so specialized today.

所以有时候要探索专业领域之外的领域会很困难。

And so sometimes it gets hard to explore an area outside of your main area.

我们曾与一位数学家交谈，他说：'你知道吗，我最近的一篇论文中，有个方向我很想深入研究，但那不是我的专长，要花很长时间。'

There's one particular mathematician we were talking to who said, know, one of my last papers, I knew there was an area that I wanted to go follow it off in this direction, but it wasn't my specialty and it would have taken me a long time.

我最终觉得，那可能不是我最该花时间的地方。

And I just kind of ended up feeling like, maybe that's not the most efficient place for me to spend my time.

现在有了GPT-5，我要回去探索那个方向了，因为我相当于有了一个读过几乎所有科学论文的同事兼合作者，可以说在你想讨论的任何领域都是相当资深的专家。

Now with GPT-five, I'm gonna go back and explore that because I've got a coworker effectively, a collaborator who has read just about every scientific paper that's out there and you know, a pretty meaningful expert on just about any topic you want.

我觉得通过ChatGPT，我能比独自研究时更好地探索这些相邻领域。

And I think I'm gonna be able to go explore these adjacencies in a far better way with ChatGPT than I could have on my own.

所以这也是个令人着迷的新视角，对吧？

And so that's also a fascinating new take, right?

它能帮助每个人——既能帮你深入钻研，就像你刚才说的，也能帮你拓宽视野。

It helps every, it can help you go deeper, like you were saying, and it can also help you go more broad.

文献检索特别有意思，我有个奇怪的爱好就是追溯那些早期被发现但后来没怎么被应用的科研成果。

Literature search is pretty interesting because like one of my weird hobbies is I like to go back and look at when was some early scientific discovery made that didn't get utilized too much later on.

著名的例子就是碳灯丝。

You know, famous one was carbon filaments.

你知道爱迪生费尽周折寻找的解决方案，其实早在二十年前就有论文发表过了。

You know, when Thomas Edison spent all that effort to try to find it, you know, had been published in like twenty years before.

是的。

Yes.

要知道，杜威十进制分类法就是那一年发明的，所以不能怪他。

You know, Dewey Decimal System was invented that year, so you can't blame him.

还有像硅辅助半导体这类发明。

Other things like silicon assisted semiconductor.

如果有人当时查阅了文献，我们可能早五到十年就掌握这项技术了。

You know, if somebody would read in the literature, we might have had that five to ten years earlier.

DNA复制技术其实在被人发现前已经发表了十到十二年。

The ability to replicate DNA that had been published like ten or twelve years before somebody figured that out.

还有我们使用的DNA鸟枪法测序技术，最早在1982年就发表了。

And then the shotgun technique we use for DNA, you know, understanding, you know, figuring out like the DNA sequencing that was first published like 1982.

但那时候还没有能运行它的超级计算机。

But at that time, there weren't supercomputers that could run it.

没错。

Right.

光是想象能拥有一个强大的工具，可以搜索所有这些资料并提取你需要的答案，就令人兴奋不已。

And that's exciting just to think of just having a really good tool that can search through all of this stuff and pull up these answers you have.

是啊。

Yeah.

而且我认为，现在一些最有趣的研究往往出现在两个领域的交叉点上。

And I think especially some of the most interesting research now happens at the intersections of two fields.

再说，一个人要成为两个领域的专家已经很难了，更不用说三四个甚至五个领域。

And again, it's hard for one person to be an expert in two fields, let alone three or four or five.

有时人类之间的协作也很困难。

And sometimes it's tough for humans to collaborate.

你不一定能找到合适的人选。

You don't necessarily find the right person.

对方也不可能有无限的耐心。

The person doesn't have infinite patience.

而现在有了GPT，你可以选择这样一个合作者：它能7×24小时工作，拥有无限耐心，而且基本上阅读了过去若干年里发表的每一篇科学论文。

And here with GPT, have now the option to have a collaborator that will work 20 fourseven, has infinite patience, you know, has read substantially every scientific paper written in the last however many years.

因此，这是一种新型的合作方式，本身就是一种加速形式。

And so it's just, it's a new kind of collaboration that is its own form of acceleration.

想想克劳德·香农的妻子是位数学家，这对他的成就有多大帮助。

Think You about Claude Shannon's wife was a mathematician and how much that to help what he was able to do.

我认为我们常常忘记合作在其中扮演了多么重要的角色。

And I think we forget how much collaboration really is a factor of that.

但我要说，听到这里可能有人会反驳：是啊，但它去年连‘草莓’都拼不对。

But I would say some people hearing this might go, yeah, but it couldn't spell strawberry last year.

它连数学题都解不了。

It couldn't do math.

那我们凭什么指望它能搞科研呢？

So why are we going to have it do, you know, science?

确实。

Yeah.

其实，我都不确定是否跟你提过这件事。

So actually, I don't even know if I've told you this.

关于GPT-5能力的起源故事，我记得大约是一年前的事了。

Own sort of origin story with appreciating what GPT-five could do, in this case it was, I think, oh, this was almost a year ago.

当时是在'O1预览版'会议上，我初次见到劳伦斯利弗莫尔实验室的物理学家布莱恩·斯皮尔斯，他当时在华盛顿特区。

So it was 'one But preview I was meeting with this guy named Brian Spears, who's a physicist at Lawrence Livermore that was in DC and we'd never met before.

所以我完全不知道会面临什么情况。

So I didn't know sort of what to expect.

原以为见面后要向他介绍新功能，解释他该如何使用O1预览版以及为何值得尝试。

I thought maybe I was gonna go in and be talking to him about what was new and what he could do with O1 preview and why he should give it a try.

没想到刚坐下他就主导了谈话，说'让我告诉你我能用你们模型做什么'。

Little did I know, I sat down and he immediately took control of the conversation and said, let me tell you what I can do with your models.

他当时说'这些对科学界来说是最惊人的突破，这将改变世界'。

And like, these are the most amazing things for science and this is gonna change the world.

接着他表示'好，让我带你看看这个'。

And he was like, okay, let me take you through this.

他打开笔记本电脑——你知道他是研究核聚变的吧？

And he opened up his laptop and, you know, he works on fusion, right?

劳伦斯利弗莫尔实验室是首个实现大规模聚变并产生正能量输出的，超级激动人心。

Lawrence Livermore was the first to do large scale fusion with positive energy, like super exciting.

他说，好，我们就以聚变为例。

So he's like, all right, we're going to take a fusion example.

首先我要从这个问题的本科版本开始讲起。

And first I'm going to start with the undergrad version of this problem.

他给我看了这段对话，说，你看，这里有一根铜棒，我们要用超高压力波轰击它。

And so he shows me this conversation and he's like, all right, so you've got, you know, a copper rod and we're going to bombard it with super high pressure waves.

会发生什么？

What happens?

然后，你看，他回答后，预览版给出了不错的答案。

And, you know, he's like, so he answered and one preview gives a good answer.

就像，好吧，很酷。

It's like, okay, cool.

所以它答对了，答对了本科水平的问题。

So it got the, it got the, got the undergrad problem right.

那么现在让我们来探讨这个问题的研究生版本。

And then now let's, let's ask the graduate version of this.

当你这样做时，铜棒内部会发生什么变化？

Now what happens inside the rod itself as you're doing this?

你知道要满足什么条件才能产生这类特定的冲击波吗？

And you know, what needs to be true in order for it to generate these certain kinds of shock waves?

他继续讲解，然后说：好，这部分答对了。

And he goes through and he's like, okay, so got that right.

好的。

All right.

现在让我们来问博士后级别的问题。

Now let's ask the postdoc level question.

好的。

All right.

这时我在想，尽管我有物理背景，但只能勉强跟上，因为他讲的内容已经超出了我的能力范围。

Now let's and ask at this point I'm like, you know, despite having a physics background, I'm just following along because for the he's beyond anything I can do.

好吧，就这样。

Like, all right.

现在让我们问一个‘你刚加入劳伦斯利弗莫尔实验室时’那种类型的问题。

Now let's ask the, you just joined Lawrence Livermore and you, you know, kind of question.

你已经完成了博士后阶段。

You've gone through your postdoc.

你是一名核物理学家。

You're a nuclear physicist.

他继续深入提问。

And he keeps going.

而这位预览者不断答对问题。

And o one preview keeps getting the answer right.

然后他说，好吧。

And then he's like, all right.

现在让我问你一个‘你在劳伦斯利弗莫尔工作了二十年’级别的问题。

Now let me ask you the, you've worked at Lawrence Livermore for twenty years question.

它继续前进并答对了。

And it goes and gets it right.

不仅如此，它还暗示前进的唯一途径是使用这套模拟工具，这些工具部分属于机密，或者只有劳伦斯利弗莫尔实验室才有。

And then not only that, but it like suggests that the only way to go forward is to use these set of simulation tools that are like partially classified or that only Lawrence Livermore has.

就像在说，你知道，虽然你们无法接触这些工具，但如果可以，你们会想用它们。

And it's like, you know, don't have access to these, but if you did, you would want to use these tools.

他接着说，听着，这里展示的内容没有什么是我不可能完成的，但会花费我数天时间。

And he's like, look, nothing in here, nothing that I just showed you was something that I couldn't do, but it would have taken me days.

当然，实验室里不是每个人都能做到这一点。

And certainly not everybody at the lab can do this.

这些工具带来的加速将改变科学。

Like the acceleration that comes, that is going to come from these tools is going to change science.

于是我从原本以为要和这个人讨论AI的价值，变成了被他彻底震撼——关于AI的潜力。

And so I went from like sitting down with this guy who I thought maybe I was gonna be sort of talking to him about the value of AI, to him just completely blowing my mind about the the potential of AI.

这是一年前的事了。

And this is a year ago.

这只是一个预览，你知道吗？

This is o one preview, you know?

自那以后我们取得了巨大进步。

We've come leaps and bounds since then.

我总是试图提醒大家，我们今天使用的AI模型，尽管GPT 5.1 Pro已经很优秀了，但它们将是我们余生中使用的最差的AI模型。

And the thing that I always try and, and like remind everybody, the AI models that we're using today, as good as GPT 5.1 Pro is, these are the worst AI models that we will ever use for the rest of our lives.

当你想到这一点，我们现在的成就本身就意味着未来非常光明。

And when you think about that, the fact that we're here just implies that the future is very bright.

你的同事们是如何使用这些工具的？

How have your colleagues been using these tools?

是的，我认为有很多不同的使用方式。

Yeah, there's a lot of different usages, I think.

文献检索，这是我正在研究的内容。

Literature search, here's what I'm working on.

它与其他任何事物有联系吗？

Does it connect to any other thing?

这是我们作为科学家投入大量时间研究的事情，就是理解当工作中出现新事物时，它与其他事物的关联。

And this is something that we spend a lot of time on as scientists, just understanding when something new shows up in our work, how it connects to other things.

好吧，让我自己彻底接受AI的那个经历，

And okay, my own experience that made me become AI pilled,

我想。

I think.

这就是你加入OpenAI的原因吗？

Is this the reason you came to OpenAI?

是的。

Yeah.

当GPT-5 Pro发布时，我认识了在OpenAI工作的马克·陈。

And when GPT-five Pro came out, I met Mark Chen, who works here at OpenAI.

他是首席研究官。

He's Chief Research Officer.

他给了我一个挑战。

And he gave me a challenge.

他非常自豪。

He was very proud.

他说，为什么不给它一个难题呢？

He said, Why don't you just give it a hard problem?

我想，他想要一个难题。

And I thought, He wants a hard problem.

于是我就提出了这个问题。

So I gave it this question.

量子引力。

Quantum gravity.

没错。

Right.

我刚发现了黑洞的这些新对称性，这种情况并不常见。

So I had just found these new symmetries of black holes, which is something that doesn't happen that often.

六月份我在预印本网站上发表了相关论文，对此我感到非常高兴。

And I'd written up a paper that came out in June on the archive, I was very happy about that.

我想，好吧，让我们看看GPT Pro如何处理这个新问题。

And I thought, okay, well, let's see how GPT Pro handles this new question.

于是我把方程给了它，但没提它有任何对称性。

And so I gave it the equation and I didn't say that it has some symmetries.

我没有给出引导性的问题。

I didn't give it a leading question.

我只问了句：'对称性是什么？'

I just said, what are the symmetries?

它思考了五分钟，然后说：'没有对称性'。

And it thought for five minutes and it said, no symmetries.

我说：'它还没达到那个水平，不过还是比AI强'。

And I go, it's not there yet, still better than the AI.

马克·谢德明显很沮丧，他说：'好吧，那就给它个简单点的问题'。

Mark Shed is visibly crestfallen, he goes, okay, well just give it an easier question then.

于是我想，好吧，我准备给它这个问题的热身简化版——找出这个方程的对称性，不是在整个复杂的黑洞时空中，而是在时空为空的平坦空间极限情况下。

And so I think, okay, I'm going to give it the warm up baby version of the problem, which is find the symmetries of this equation, not in the full black hole spacetime, which is complicated, but in the flat space limit where the spacetime is empty.

然后我按下回车，它思考了九分钟，给出了这个绝妙的答案：'哦，这个方程具有共形对称性'——完全正确，还列出了三个生成元，简直太完美了。

And hit enter, it thinks for nine minutes, and it comes back with this beautiful answer, Oh this equation has conformal symmetry, which is the correct thing, and here are the three generators, it was very beautiful.

这个版本的方程，我确信过去几十年里已经被研究过无数次。

And this version of the equation, I'm sure it has been studied many times over the decades.

虽然不清楚他具体做了什么，但他得出了正确答案，我觉得这非常棒。

So I don't know what he did exactly, but he came up with the answer and I thought, okay, this is very good.

这真是个了不起的成果。

This is a great outcome.

接着马克说：'既然它已经在热身例题上做过准备，现在在这个聊天窗口里试试更难的题目吧'。

And then Mark said, okay, but now that it's been primed on the warm up example, try again in this instance of chat, the harder problem.

我当时想：'好，来吧'。

And I thought, okay, let's go.

于是我们再次输入那道难题，按下回车，它开始不停地思考。

And so we give it the hard problem again, hit enter, and it thinks and it thinks.

那是我第一次见到它思考这么久。

And that was the first time I saw it, I think, for so long.

我想大概有十八分钟。

I think it's like eighteen minutes.

然后它给出了这个完全正确的完美答案。

And it comes out with this beautiful answer that was completely correct.

这让我非常震惊，因为我为此已经研究了很长时间。

And that blew my mind because I had been working on this for a very long time.

我得说这个计算已经达到了我能力的极限。

And I would say that that calculation is at the edge of my abilities.

我认为很少有人能像我这样完成这个计算。

I think it's something that very few people could have done the way I did it.

所以我真的很震惊，因为你花费数年时间训练自己成为顶尖专家，研究黑洞对称性和这类方程。

And so I was really shocked because you spend years of your life training to be best in class or something and finding symmetries of black holes and these kinds of equations.

这就是我的专长。

That's my jam.

我当时想，好吧，看来事情就这样发生了。

And I thought, okay, so I guess that just happened.

这真的让我思绪万千。

And it really sent my mind reeling.

我震惊了好几天，之后就一直忍不住反复思考这件事。

And I was a little bit shell shocked for a few days and then I just couldn't stop thinking about it.

后来我意识到，我必须参与其中，因为眼看着这种能力在当下世界出现却不参与其中，对我来说简直太疯狂了。

And after that I realized, okay, I have to become involved in this because to see this capability emerge into the world right now and to not be involved with this just seemed crazy to me.

我其实正想说，你刚才提到一个非常重要的观点：当你给它难题时它没答对，

I was going to actually think you made made a really important point in the middle of that around the fact that you gave it the hard question, it didn't get it right.

但给它简单些的问题就能答对，然后你才能继续给它更难的问题。

You gave it an easier question, it got that right, and then you're able to give it a harder question.

尽管我们显然对未来的前景充满热情，但同时也存在一个非常现实的认知——

Is still, you know, as excited as we clearly are about the future here, there's also a very real sense.

当你给GPT-5或任何这类AI模型提出处于能力边界的前沿问题时，它们往往还是会经常出错。

Like when you're giving GPT-five or any of these AI models a problem that's on the frontier, that's at the limit of their capabilities, they tend to still be wrong a lot.

对吧？

Right?

就像任何人在其能力边界上操作时都会遇到的情况一样。

Kind of like any human would be at operating at the level of, at the frontier of their capabilities.

而且你知道，这还不是自动完成的。

And it takes, you know, it isn't just automatic yet.

希望未来能够实现——输入任何难题，模型都能给出答案。

Hopefully in the future it will be, you know, enter in any hard question and the model answers it.

但如今需要大量来回互动，那些最擅长从模型中获取最大价值的研究人员，都具备与它们反复沟通的耐心。

But today there's a lot of back and forth and the people that are best, the researchers that are best at getting the most out of the models have a sort of patience to go back and forth with them.

我认为这很自然。

I think that's natural.

这大概就像你和任何两个在能力极限边缘工作的人合作时的模式。

It's probably the way that you would work with any, you know, any two people operating at about the limit of their capabilities.

但我觉得很重要的一点是，尤其是对正在用模型做研究的听众来说，要知道这不是一次尝试就能永远奏效的。

But I think it's important, especially for folks listening to this who are doing research with the models to know that it's not, it isn't just one shot and it always works.

确实需要反复沟通，需要某种耐心。

It, there really is a back and forth and sort of a patience that it takes.

我们正在花大量时间思考的一个有趣研究问题是，如何帮助人们减轻这种认知负担。

And one of the interesting research problems that we're spending a lot of time thinking about is how we, how we help people with, how we sort of help reduce that cognitive load.

因为当你处理某个问题时，假设模型在该问题上的通过率只有5%。

Because when you're working on a problem, say the model is, has a 5% pass rate on some problem.

从技术上讲，模型每20次尝试中能答对1次，但这确实已接近其能力极限。

So technically the model can get it right once out of 20 times, but it's really at the frontier.

所以它不可能做到几乎每次都答对，甚至连接近都谈不上。

So it's not gonna get it right nearly, you know, even close to every time.

如果你在ChatGPT里直接输入这个问题，可能需要输入10次才有概率得到正确答案。

If you're sitting inside ChatGPT and just entering in this question, you're gonna have to enter it in, you know, what, 10 times before you have the odds that it's gonna get the right answer.

而大多数人根本不会这么做。

And that's most people aren't gonna do that.

因此存在大量模型本可以解决的问题，但人们尝试几次后就会放弃——'试了三次都没答对'。

And so there's a whole host of problems that the model can solve that people probably try and are like, oh, well, after three tries, it didn't get it right.

所以干脆就'算了，我换下一个问题吧'。

So let's, I'll move on.

模型目前还不够好。

The model's not good enough yet.

但实际上它是可以的，只是很难区分低通过率的问题和过于困难的问题。

And actually it is, but it's just very hard to tell apart low pass rate problems from problems that are too hard.

我认为这对帮助研究人员和数学家突破非常重要，因为当前最有趣的问题将是那些模型通过率极低但非零的情况。

And I think that's actually a really important thing for us to help researchers and mathematicians get past because the most interesting problems right now are gonna be the ones where the model has a very low, but non zero pass rate.

这些将是模型能解决的最难问题，也是它能加速科学发展的最佳途径。

Those are gonna be the hardest problems that the model can solve, the best ways that it can, that it can help accelerate science.

因此这是一个非常有趣的研究课题，我们正在努力使其变得更自动化，减少基础工作。

And so that's a really interesting research problem that we're taking on to try and make that a little more automatic, a little less groundwork.

但目前来说，投入时间与模型反复交流确实能产生成果。

But for now, like putting in the time and really going back and forth with the model does yield results.

嗯，感觉我们现在就像从GPT 3.5过渡到ChatGPT的那个时刻。

Well, it feels like we're at a moment kind of like when we went from GPT 3.5 to ChatGPT.

3.5是个非常强大的模型，但它本质上仍是个基础模型。

3.5 was a model, extremely capable model, but it was still effectively a base model.

我当时是一名提示工程师，知道如何通过提示让它给出出色的结果。

And I was a prompt engineer at the time and knowing how to prompt that I could get great results for it.

但需要掌握所有这些小技巧才能真正理解上下文。

But it took all those little tricks to sort of understand the context.

后来我们转向ChatGPT，明白了：我们了解人们试图解决的问题类型。

Then we went to chat GPT and we understood, Okay, we know the kind of problems people are trying to solve.

让我们帮助他们更容易地达成目标，而无需掌握那些技巧。

Let's make it a little bit easier for them to get there without having to do that.

感觉这有点像我们现在进入科学领域的方式——既然有像Alex这样的人在解释你们试图解决的问题和正在做的事情，我们可能会看到这方面的巨大加速。

It feels like that's kind of where we're heading into a science though, that now that you have people like Alex explaining the problems you're trying to solve and what you're doing that we may see like a big acceleration with this.

我认为这可能是任何处于模型能力边界问题的共同特征。

I think it's probably just a characteristic of any question that's on the frontier of, or sort of at the limit of what the models can do.

回到GPT-3.5和早期4.0版本时，处于模型能力边界的问题要基础得多。

And back with GPT 3.5 and early versions of four, the questions that were at the limit of what the model can do were much more basic.

现在这些问题变成了科学研究级别，但当你处于前沿领域时，通过率仍然会很低。

Now they're questions of, you know, scientific research, but you still, when you're operating at the frontier, the pass rate will be low.

所以你得坚持尝试几种不同的方法，抓住模型做对的部分进行优化，同时指出它在哪些地方出了错，这本身就很有价值。

And so you gotta kinda, like there's value in sticking and with trying a few different things and taking the parts that it gets right and refining them while telling the model where it got other things wrong.

在我提到的这个例子中，它需要一个热身步骤，但这个热身就像人类会做的那些显而易见的热身准备。

In this example I mentioned, it needed a warm up, but the warm up was the obvious warm up that you would do as a human.

对。

Right.

因为实际上，当我着手解决这个问题时，最初并没有考虑黑洞的情况。

Because actually, when I was attacking this problem, I wasn't thinking about the black hole case first.

这个平坦空间极限才是显而易见的起点，而我也正是从这里入手的。

This flat space limit was the obvious place to start, and that is where I began.

所以我认为这些模型其实非常出色，但我们可以进一步改进，让它们能自主想到热身问题，从而直接进入正题。

And so I think the models are actually really good, but we could get better at making them think of the warm up problem themselves so they can go there directly.

但更广泛地说，我认为我们必须牢记一点：作为科学家，我们的职责就是推动认知边界。

But more generally, I think there's this thing we have to bear in mind, which is that as scientists, our role is to push the edge of knowledge.

总有些知识存在于边界之外，而我们的目标就是通过理解它们，将这些未知领域纳入认知边界之内。

There are things that are just beyond the edge, and our goal is to bring them before the edge of knowledge by understanding them.

但这个边缘非常参差不齐。

But this edge is very jagged.

宇宙中存在一些非常基础的问题，比如为什么空间是三维的？

So there are very basic questions about the universe, like why are there three dimensions of space?

或者大爆炸时发生了什么？

Or what happened to the Big Bang?

这些都是每个人都想知道答案的问题。

These are things that everybody wants to know the answer to.

然而，尽管这些都是简单的问题，我们对此确实说不出什么有见地的话。

And yet, even though they're simple questions, there's really nothing intelligent to say about this.

我们就是不知道。

We just don't know.

实际上，这些都是非常棘手的问题。

They're very hard problems, actually.

而与此同时，有些你认为我们根本不可能回答的极其困难的问题，我们却已经有了极其详尽的答案。

And then meanwhile, there are these very hard questions that you would think we wouldn't be able to answer at all, to which we have extremely detailed answers.

我们可以预测电子的偶极矩，精确到小数点后12位，这简直不可思议。

We can predict the electron dipole moment to, I don't 12 decimal places, something crazy.

因此人类知识的边界本身是参差不齐的，需要多年研究生学习才能了解这个边界在哪里。

So the edge of human knowledge itself is very jagged and it takes many years of graduate school to learn where the edge is.

而我们在这些AI模型中发现，它们的知识边界同样非常不规则。

And I think what we're finding with these AI models is that the edge of their knowledge is also very jagged.

你提到有些基本问题这些模型无法回答。

So you mentioned there's some basic questions that the models can't answer.

确实如此。

That's true.

但同时，现在就有一些非常困难的问题它们已经非常适合解答了。

At the same time, there are some very hard questions that they're very well suited for already today.

我认为令人兴奋的是，它们的知识边界以一种与我们不同的方式呈现不规则性。

And I think what's exciting is that their edge of knowledge is very jagged in a way that's different from ours.

显然，随着时间的推移，我认为这些模型的能力边界将会不断扩展。

So obviously, as time goes on, I think the edge of ability for these models is going to keep expanding.

但只要它以与我们不同的方式扩展，这同样非常有趣，因为在它能超越我们的交界处，我们可以领先于它，我认为那里将会发生许多有趣的事情。

But as long as it expands in a way that is slightly different from our edge, that's also really interesting because at the intersection where it can go farther than us so we can get ahead of it, that's where a lot of interesting things are going to happen, I think.

是的。

Yeah.

人类与AI的结合，正是如此。

Human and AI together Exactly.

比单独的人类强大得多，确实如此。

Are Much more powerful than human alone Exactly.

或AI。

Or AI

我想进一步探讨这一点，但首先请告诉我关于那篇研究论文的情况。

I want to explore that a little bit more, but first tell me about the research paper.

是的，我们已经讨论了很多Alex从他与同事相处时收集的轶事案例，这些案例我们几乎每天都能在Twitter上看到。

Yeah, so we've talked a bunch about these anecdotal examples that Alex has gotten from the time that he spent with his colleagues that we see coming in across Twitter, you know, on a semi daily basis at this point.

我们想把这些案例汇集起来，撰写并发表一些内容，阐述当前GPT-5在科学领域的现状。

And we wanted to sort of bring them together and just write something, publish something about, that lays out the current sort of state of GPT-five with respect to science.

我们集结了OpenAI内部的几位合作者，以及来自数学、物理、天文学、计算机科学、生物学、材料科学等多个领域的八、九位外部学者。

And so what we've got, it's a handful of collaborators from Inside OpenAI and I think eight or nine academics from beyond our walls across a bunch of different fields, math, physics, astronomy, computer science, biology, material science.

这篇论文大约有12个章节，每个章节都展示了GPT-5如何以不同方式加速他们的研究工作。

And the paper is something on the order of 12, each one highlighting a different way that GPT-five is accelerating their work.

我们的目标不是夸大其词，宣称所有问题都已解决。

The goal is not to be, you know, hype y and say, everything is solved.

而是想实事求是地说，

It's, you know, it's really to say,

看——人人都能用的悬浮滑板。

look- Hoverboards for everybody.

对。

Yeah.

比如，这些方法行之有效，

Like, this is what works.

而那些则行不通。

This is what doesn't work.

这是我尝试过的。

Here's what I tried.

在很多情况下，我们分享了ChatGPT的完整对话链接，这样你可以看到科学家与模型的互动过程。

In many cases, we're sharing the ChatGPT, you know, the full share links, the So you can see the back and forth that the scientist has with the model.

这更像是记录当下时刻，展示我们目前达到的水平。

And it's meant to be kind of a moment in time to say, this is where we are today.

我想半年或一年后再回顾时，我们会走得更远，这很令人兴奋。

And I think we'll look back in six months, twelve months and, you know, we'll be much further and that'll be exciting.

但即便是现在，我们在论文中专门设置了文献检索的各种案例章节，还有关于计算加速等不同应用场景的章节。

But even where we are today, we've got a section in the paper on the different, a bunch of different examples around literature search, a section in the paper with a bunch of different examples around acceleration, whether it's calculations and other things like that.

还有一个章节我们实际贡献了四五个非平凡的数学新发现。

And then a section where we actually contribute four or five new non trivial results in mathematics.

其中有些成果虽小，有几个甚至本可以单独成文发表。

And a couple of these are small, a couple of them probably could have been papers on their own.

所以从平凡实用的加速应用，到更具深远意义的——GPT-5实际上正在突破人类知识的现有边界。

And so you go from kind of the mundane, very pragmatic and real bits of acceleration to the more sort of profound GPT-five actually pushing past the current frontier of human knowledge.

我们对这篇论文感到无比兴奋。

And so we're super excited about this paper.

我认为未来还会有更多成果涌现。

It's, you know, I think there'll be a lot more to come.

顺便说一句，我们不是唯一做出杰出工作的实验室。

We're not the only lab doing great work by the way.

谷歌在这方面已经深耕多年，我对Demis及其团队在AlphaFold等项目上的成就深表敬意。

Google has been doing this for a while and I have a ton of respect for what Demis and the team have done with AlphaFold and more.

我只是觉得我们正处在一个激动人心的时代。

I just think we're at a really exciting time.

科学领域的创意往往会在多人同时产生相同想法时迎来爆发期，就像Alex提到的量子力学，或是电灯泡的发明。

You know, ideas in science often have their moment when you have multiple people coming with the same idea, whether it's quantum mechanics like Alex was talking about or the light bulb.

现在很明显，AI才刚刚开始改变科学领域，未来几年将会非常精彩。

Right now it's very clear that AI is just beginning to change science and it's going be an exciting few years.

对于理科学生和研究生，你有什么建议？

What advice do you have for students and grad students in the sciences?

因为我听到有人说，哦，我们以后不需要科学家了，这听起来简直荒谬。

Because I hear people talk about like, oh, we're not going to need scientists anymore, which sounds absolutely crazy.

望远镜并没有让天文学家失业。

It's not like the telescope got rid of the astronomer.

实际上创造了...你对这个怎么看？有什么建议？

Actually created How the do you feel about that and what advice do you have?

我认为首先必须承认，学术界目前存在许多与AI无关的焦虑。

I think first of all, it's important to acknowledge there's a lot of anxiety in academia right now that is unrelated to AI.

这与我国科研组织方式的诸多变革有关，而我们仍在经历这些变化。

It has to do with lots of changes in the way that science is organized in this country, and we're still going through these changes.

因此我认为在与年轻人交流时，围绕这一点存在许多焦虑情绪。

So I think that talking to young people, there's a lot of anxiety surrounding this.

实际上我认为人工智能是一种令人振奋的新工具，它正在到来并变得可用，将极大地帮助科研工作，因为它能显著提升每个人的效率。

I actually think AI is a really exciting new tool that's coming, that's becoming available, that is going to help a lot because it's just going to make everybody just so much more efficient.

正如凯文早些时候提到的，当你进行一个研究项目时，往往并不清楚具体该往哪个方向走。

As Kevin was mentioning earlier, when you work on a research project, oftentimes you don't know which way exactly to go.