本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
我给我弟弟订了一份《纽约时报》的订阅。
I gave my brother a New York Times subscription.
我们会交换文章。
We exchange articles.
因此，读了同一篇文章后，我们可以一起讨论。
And so having read the same article, we can discuss it.
她送了我一份为期一年的订阅，这样我就能访问所有游戏了。
She sent me a year long subscription so I have access to all the games.
《纽约时报》增进了我们共度的优质时光。
The New York Times contributes to our quality time together.
它丰富了我们的关系。
It enriches our relationship.
这真是一个
It was such a
很酷又贴心的礼物。
a cool and thoughtful gift.
我们在读同样的东西。
We're reading the same stuff.
我们在做同样的食物。
We're making the same food.
我们想法一致。
We're on the same page.
了解更多关于将《纽约时报》订阅作为礼物的信息，请访问 nytimes.com/gift。
Learn more about giving a New York Times subscription as a gift at nytimes.com/gift.
辛西娅·里沃是世界上最好的歌手。
Cynthia Rivo is the best singer in the world.
她太棒了。
She's incredible.
我不知道她的声音有什么魔力，但每次听到，都会让我泪流满面。
I don't know what it is about her voice, but it, like, brings me to tears, like, every single fucking time I hear.
她的声音有着极其动人的情感。
She has the most incredibly, like, emotional voice.
我不知道。
I don't know.
我当时在想那到底是什么，但就是她太着迷了。
I I was, like, trying to figure out what it was, but it's just like, she's just obsessed.
因为显然她有那种力量，但里面还有很多层次感。
Like because, like, she obviously she has the power, but there's all these, like, textures in there.
你看过她回母校的那段设计成病毒式传播的视频吗？
Did you see the designed to go viral clip of her visiting her old school?
大家都
Everyone's
我当然彻底疯了。
I obviously lost my shit.
我当时是
I was The
最棒的是学生们开始唱歌，但他们唱得简直难听极了。
absolute best is that the students start singing, and they just sound like shit.
这简直是我的噩梦。
That's like my nightmare.
太棒了。
Sweet.
想象一下你是这些孩子中的一个。
Just imagine you're one of these kids.
你根本不是说这只是课后课程。
You're not that it's just it's just like after school class.
你知道的，这只是一个小小俱乐部。
You know, it's a little club.
你只是为了稍微丰富一下生活而参加，就像在慢慢熬着，只想熬过这一天。
You're just doing it for a little bit of enrichment, and, like, you're just kinda plotting along trying to get through the day.
然后该死的辛西娅·里瓦出现了，他们就说：好了，孩子。
And then fucking Cynthia Riva shows up, and they're like, alright, kid.
该你了。
You're up next.
你有什么？
What do you got?
不。
No.
谢谢。
Thanks.
不。
No.
太棒了。
It was so sweet.
我会表现得很好。
I would do well up.
太棒了。
So sweet.
我是凯文·拉塞，《纽约时报》的科技专栏作家。
I'm Kevin Russe, the tech columnist at the New York Times.
我是来自Platformer的凯西·纽特。
I'm Casey Newt from platformer.
这是《硬核》节目。
And this is Hard Fork.
本周，未来之家首席执行官萨姆·罗德里格斯来到演播室，为我们厘清人工智能科学中的炒作与现实。
This week, Future House CEO Sam Rodriguez joins us in the studio to separate the hype from the reality of AI science.
好吧，凯西，该来点科学了。
Well, Casey, it's time for some science.
嗯。
Yeah.
等我一下，凯文。
Give me a second, Kevin.
我先穿上我的实验服，点上本生灯，看看你今天为我们准备了什么好东西。
I'm just gonna put on my lab coat here to get up my Bunsen burner and see what you've got cooking for us today.
我一直痴迷于这个问题：人工智能在科学和科学发现中究竟做了什么，又没做什么。
So I have been obsessed with this question of what AI is and isn't doing for science and scientific discovery.
显然，我们经常从大型AI公司的领导者那里听到这种说法，比如达里奥·阿马德、萨姆·阿尔特曼、德米斯·哈萨比斯。
Obviously, this is something we hear a lot about from the leaders of the big AI companies, people like Dario Amade, Sam Altman, Demis Esabas.
最近几个月，他们都表示，他们认为我们离用这些新AI工具解决科学问题、治愈疾病和应对气候变化已经非常接近了。
They have all been saying things in recent months about how close they believe we are to solving new scientific problems and curing diseases and fixing the climate with all of these new AI tools that they're building.
其中一些显然是炒作，或者至少带有炒作的痕迹。
And some of that is obviously hype or at least has the sort of markings of hype.
但事实上，AI与科学领域正在发生很多真正的事情，而我个人却缺乏评估这些进展的能力。
But there's actually a lot of real stuff going on in AI and science that I just do not feel personally qualified to evaluate.
是的。
Yeah.
我还要说，科学已经成为这些科技公司领导者希望我们评价他们的主要方式之一。
And I would also say that science has become one of the main ways that the leaders of these tech companies want us to evaluate them.
因为每当他们的模型出了什么糟糕的事情时，我们得到的回应基本上都是：别担心，我们马上就要治愈癌症了。
Because whenever one of their models does something horrible, the message we basically get back in response is, don't worry, we're about to cure cancer.
再耐心等一等。
Just hang on tight.
我知道这个聊天机器人可能让你快疯了，但如果你能再给我们几次更新，我们一定会做些真正出色的事情。
I know that this chatbot might be driving you to madness, but if you could just give us a few more releases, we're gonna do some really good stuff.
是的。
Yes.
而且我们现在也从美国政府那里听到了类似的说法。
And this is something that we're also hearing now from the US government.
白宫在感恩节前宣布了‘创世计划’。
The Genesis Mission was announced by the White House just before Thanksgiving.
他们称这是旨在释放一个新时代的、由人工智能加速的创新与发现的国家级协调努力，以解决本世纪最具挑战性的问题。
That is what they're calling a dedicated coordinated national effort to unleash a new age of AI accelerated innovation and discovery that can solve the most challenging problems of this century.
不。
No.
我还以为‘创世计划’只是他们想请菲尔·柯林斯去白宫圣诞派对演出呢。
I thought the Genesis Mission was just them trying to get Phil Collins to play the White House Christmas party.
看来不是这样。
I guess not.
因此，今天我们请来了一位真正的科学家，帮助我们辨别哪些科学发现和可能性是真实的，哪些不是。
And so today, we have brought in a bonafide scientist to help us understand which of the sort of scientific discoveries and possibilities out there are real and which are not.
我们需要一位视野广阔的专家，不仅关注人工智能对生物技术或药物发现的影响，还要关注其对不同科学领域的整体影响。
We need an expert with a broad focus, someone tracking the impact of AI, not just on biotech or drug discovery, but across the different sciences.
凯西，我们找到了最合适的人选。
And, Casey, we have found the perfect person.
让我们来了解一下他。
Let's hear about him.
萨姆·罗德里格斯是未来之家和爱迪生科学公司的联合创始人兼首席执行官，这是一家位于旧金山的机构，我想它既是非营利组织也是营利性企业，未来之家也是。
Sam Rodriguez is the cofounder and CEO of Future House and Edison Scientific, which is a San Francisco based I guess it's both a nonprofit and a for profit, not Future House as well.
以前听过吗？
Heard that before?
听过。
Yes.
等他
Come back when he
他有个董事会合作。
has his board coop.
Future House 是非营利组织。
Future House is the nonprofit.
Edison Scientific 是从中衍生出的营利性机构。
Edison Scientific is the for profit that spun out of it.
我去过他们在 Dogpatch 的办公室。
I've been to their office in Dogpatch.
那里真的很有趣。
It's really fun.
它感觉像一个疯狂的科学家实验室。
It's it sort of feels like a kind of wacky mad scientist lab.
他们有各种我都不懂的实验室设备。
They've got all these, like, you know, sort of lab machines that I don't understand.
人们穿着白大褂跑来跑去，都在谈论人工智能，感觉真是个酷地方。
You know, people running around in lab coats, they're all talking about AI, and it just feels like kind of a cool place to be.
他们正在构建萨姆所称的‘AI科学家’，这是一种能够执行科学研究某些环节的AI代理。
And they are building what Sam calls an AI scientist, which is an AI agent that can do sort of parts of the process of scientific research.
萨姆本人也是一位科学家。
And Sam is also himself a scientist.
他拥有麻省理工学院的物理学博士学位。
He has a PhD in physics from MIT.
在创办Future House之前，他曾多年领导一个应用生物技术实验室。
And before he launched Future House, he spent several years running an applied biotech lab.
因此，他从多个角度见证了这些发展。
So he has sort of seen this stuff happening from a couple different angles.
是的。
Yeah.
今天，我们想和他聊聊他目前正在做什么，同时也了解他对整个领域的看法。
And today, we wanna talk to him about what he is up to, but also kind of get his vision of the entire landscape.
告诉我们哪些方面有效，哪些无效，哪些是炒作，哪些是真正的成果。
Tell us what is working, what isn't, where's the hype, where's the real stuff.
萨姆对此有很多话说。
Sam has a lot to say about it.
是的。
Yes.
我认为可以说，萨姆在关于人工智能将如何推动科学发展的观点中，属于较为乐观的一端。
And I think it's fair to say that Sam is on the more optimistic end of the spectrum of beliefs about what AI will do for science.
但正如你将在我们的对话中听到的，他比那些声称我们能在五到十年内治愈所有疾病的人更为谨慎。
But as you'll hear in our conversation, he's more skeptical than some of the most optimistic people who are claiming that we'll cure all disease in five or ten years.
是的。
Yeah.
如果你一直渴望听到一些对最狂野预测的泼冷水，他正好能提供一些这样的观点。
If you've been craving a little bit of cold water for the wildest projections, he has some of that
给你。
to offer you.
所以我们来请他出场。
So let's bring him in.
我们回来后，将邀请萨姆·罗德里格斯加入。
When we come back, we'll be joined by Sam Rodriguez.
你知道吗？印度是全球最大的加密货币采用国，而爱沙尼亚在所有选举中都提供在线投票。
Did you know that India is the biggest adopter of crypto globally and that Estonia offers online voting in all its elections?
我是凯瑟琳·本霍尔德，纽约时报全新每日通讯《世界》的主持人。
I'm Catherine Benhold, host of The World, a new daily newsletter from The New York Times.
我曾用二十年时间在十多个国家进行报道，有一天我突然想到：我会想读什么样的通讯呢？
I spent twenty years reporting from more than a dozen countries, and it occurred to me one day, you know, what kind of newsletter would I like to read?
我不住在美国内地。
I don't live in The US.
我希望有一份专门为全球读者撰写的通讯，能帮助我理解正在发生什么以及为什么重要，最好还能不让我情绪低落。
I want something that's written especially for a global audience, something that helps me understand what's going on and why it matters, and ideally something that doesn't just get me down.
《世界》正是如此。
The world is just that.
每个工作日早晨，我们会为您带来最重要的新闻、我的同事们发自现场的报道，以及一些令人愉悦的视频惊喜。
Each weekday morning, we bring you the biggest stories, dispatches from my colleagues on the ground, and a few delightful surprises with video too.
《世界》新闻简报，来自《纽约时报》。
The World Newsletter from The New York Times.
立即在 nytimes.com/theworld 注册，每天工作日早晨直接发送到您的邮箱。
Sign up now at nytimes.com/theworld to get it in your inbox each weekday morning.
萨姆·罗德里格斯，欢迎来到《硬核》。
Sam Rodriguez, welcome to Hard Fork.
你好。
Hello.
谢谢。
Thank you.
今天我们请来您作为专家，为我们指引近期科学领域中最重要的AI突破。
So we have brought you here today to be expert, our guide to the biggest recent AI powered breakthroughs that are happening in science.
这个领域我大致知道它很重要，也在发生大事，但我们俩都不是科学家，虽然我小学时做过一个超棒的小苏打火山。
This is an area that I sort of understand in an ambient way is important, and there are big things happening, but neither of us are scientists, although I did make a killer baking soda volcano in in elementary school.
今天我们有太多话题要聊了。
So we have so much to talk about today.
但在深入具体细节之前，我想问问你正在做的项目。
But before we get into some of the particulars, I want to ask you about your project that you've been working on.
嗯。
Yeah.
上个月，你所在非营利组织的商业分支——爱迪生科学公司——推出了一款名为‘宇宙’的AI科学家，你表示它单次运行就能完成相当于六个月博士或博士后科学家的工作量，没错。
Last month, the commercial arm of your nonprofit, which is called Edison Scientific, launched a new AI scientist called Cosmos that you say can accomplish work equivalent to six months of a PhD or postdoctoral scientist in a single run Yeah.
这个模型。
Of this model.
给我们讲讲‘宇宙’是如何工作的，以及这个‘六个月’的数字是怎么来的。
Tell us about how Cosmos works and where that six month number comes from.
嗯。
Yeah.
嗯。
Yeah.
没错。
Exactly.
实际上，我先说一下，当我得到这个六个月的数据时，我最初的反应是：这不可能是真的。
And actually, will just, like, start out by saying that when I got that six month number, my reaction originally was like, there is no way that this is true.
对吧？
Right?
我们现在已经用多种方式对它进行了测量。
And we've now measured it in a bunch of different ways.
我可以带你们详细了解一下。
I can walk you guys through that.
但总的来说，先退一步说，我们花了两年时间研究如何构建一个AI科学家。
But, basically, just to take a step back, so we've been working for two years on figuring out how to build an AI scientist.
这里的理念是，我们可以做的科学工作远远超过我们拥有的科学家数量。
And the concept here is there's so much more science that we can do than we have scientists.
对吧？
Right?
那么，我们该如何扩大科学的规模呢？
And so how do we scale up science?
关于Cosmos，真正酷的一点是，它是我们迄今为止打造的、第一个让你在使用时真正感觉像一个AI科学家的东西。
And the thing that is that happened with Cosmos that is that is pretty cool is Cosmos is, like, the first thing that I think that we've made that actually really feels like an AI scientist when you're working with it.
对吧？
Right?
也就是说，你进去后，给它一个研究目标。
Which is to say that you go in, you give it a research objective.
然后它离开，再回来时带来一些真正深刻、有趣，有时甚至错误的见解，但大约80%的时间是对的，这就像你让一个人去完成某项任务，他回来时正确率也差不多。
It goes away, and it comes back with insights that are actually, like, really deep and and interesting and sometimes wrong, but but, you know, about about 80% of the time right, which is, like, kind of similar to, like, if you ask a human to go away and do something, comes back, like, similar percentage of the time is right.
而且，和它一起工作是一种全新的体验。
And and it's, like, it's a kind of new experience working with it.
所以这一切都非常令人兴奋。
So that's all that's that's very exciting.
具体到六个月这个数字，我们是这样测量的：我们有一群学术合作者，也就是之前做过不少研究但尚未发表成果的科学家。
The six month number specifically, the way that we measured this was we had a bunch of academic collaborators, you know, scientists who had done a bunch of science previously that they had not published yet.
我们把同样的研究目标和同样的数据集交给AI——Cosmos，让它去探索并发现新的成果。
And we basically gave the same research objective and the same dataset to the AI, to Cosmos, and we asked it, you know, to go away and just make new discoveries.
它会回来，并且找到了研究人员一夜之间发现的相同结果。
And it would come back, and it had found the same things that the researchers had found overnight.
然后你去问研究人员，你们最初找到这个结果花了多长时间？
And then you go and you ask the researchers, you know, how long did it take you to find this in the first place?
他们会说，大概三个月、五个月，或者六个月之类的。
And they would say, like, three months, five months, like, six months, whatever.
所以这个数字就是这么来的。
And so that's where it comes from.
这正是他们得出这一发现所花费的时间。
And it's like, that's the amount of time that it took them to come up with the finding.
明白了。
Got it.
所以让我问你几个问题，让我更好地理解一下。
So let me just ask you a couple of questions so I can ground myself here.
这个工具是不是像其他聊天机器人那样，是一个你可以输入文字的框？
Is is this tool kind of a box you type into like the other chatbots?
如果是这样，那它背后是由什么驱动的？
And if so, what is powering it?
你们是从零开始构建了自己的模型吗？
Did you guys sort of build your own model from scratch?
你们是不是对其他公司的模型进行了微调？
Did you sort of, you know, make fine tune mints fine tunings to another company's model?
或者，是的。
Or Yeah.
所以，它确实是一个你可以输入内容的框体。
So it is it is indeed a a box that you basically type into.
你输入一个研究目标。
You you ask a research objective.
它不是一个聊天机器人。
It's not a chatbot.
对吧？
Right?
它会运行大约十二个小时，然后才把结果返回给你。
Like, it runs for twelve hours or or so before eventually coming back to you with with its findings.
在构建方式上，我们基于来自OpenAI、Google和Anthropic的多个不同语言模型进行开发。
In terms of how it's built, it we build on top of a bunch of different language models from OpenAI, from Google, from Anthropic.
在每次运行中，我们都会使用来自所有不同提供商的模型。
Like, in any given run, we use models from all the different providers.
我们还拥有自己内部训练的专用模型，这些模型在我们专门训练的任务上，表现远优于前沿提供商提供的模型。
We also have, like, our own models for specific tasks that we've trained internally where those models are, like, much better for the specific task that we train them on than the models that the Frontier providers make.
明白了。
Got it.
而Cosmos的关键洞察在于，我们使用了一种称为结构化世界模型的技术。
And then the key insight in Cosmos is basically this use of what we call, like, a a structured world model.
如今AI系统的主要局限之一是，它们在执行任务的长度和复杂性上存在限制，一旦超出这个范围，就容易偏离轨道。
So one of the main limitations with AI systems today is that they're just limited in the length of the task and the sophistication of the task that they can carry out before they kind of go off the rails.
它们会忘记自己正在做什么。
They, like, you know, forget what they're doing.
它们不再专注于任务了。
They no longer are on task.
我们发现了一种方法，让它们不断贡献到一个随着时间积累的全局世界模型中，这个模型描述了它们所执行任务的完整知识状态，这使得我们能够协调数百个代理并行或串行运行，共同朝着一个连贯的目标努力。
And what we figured out was a way to have them contributing to this world model that gets built up over time that basically describes, like, the full state of knowledge about the task that they're working on, which then means that we can orchestrate hundreds of, like, different agents running in parallel, running in series, and have them all working towards a coherent goal.
这才是真正的突破。
And that was, like, the real unlock.
对。
Right.
我对Cosmos的另一个感兴趣的地方是它的成本。
Another thing that I found interesting about Cosmos is the cost.
这个模型每次提示收费200美元。
This model costs $200 per prompt.
是的。
Yeah.
所以每次你给它一个任务，都要支付200美元。
So every time you give it a task, you're paying $200.
为什么这么贵？
Why is it so expensive?
我的意思是，它消耗了大量的计算资源。
I mean, it it uses a lot of compute.
我的意思是，根本原因就是它消耗了大量的计算资源。
I mean, that's like the fundamental answer is it uses a lot of compute.
对吧？
Right?
能给我们一个具体的数字吗？
Like, give us a sense of how much.
一个单独的Cosmos运行平均会编写42,000行代码，并阅读1,500篇研究论文。
Well, so as an individual run from Cosmos will will write 42,000 lines of code and read 1,500 research papers on average.
如果你运行Claude，它可能只会写几百行代码。
Like, if you run Claude, it might write, like, a few 100 lines of code.
对吧？
Right?
这样你就有点概念了。
So that gives you some sense.
这意味着投入了大量计算资源。
It's like there's a lot of compute that is going into this.
你有没有遇到过这样的情况：一位科学家的猫踩到了键盘，不小心按了回车，结果一下子花了600美元？
Have you ever had, like, a scientist whose cat walks across the keyboard and accidentally hits enter and and all of a sudden spends, like, $600?
这是个问题。
This is a problem.
这是
This is
一个问题。
a problem.
而且你必须明白，如果你是一名科学家，去做一个实验，就会得到一些数据。
And we are, like, right so the thing that you have to understand, right, is that if you are a scientist and you go and do an experiment, you get some data back.
你会花五千到一万美元来收集这些数据。
You're gonna spend 5 or $10,000 gathering that data.
所以科学家们想要的是他们所能获得的绝对最佳性能。
And so what scientists want is they want the absolute best performance that they can get.
而且，使用过Cosmos的科学家通常会回来告诉我，他们简直不敢相信我们只收200美元。
And, like, scientists who have used Cosmos generally come back to me and are like, they can't believe we're only charging $200 for it.
对吧？
Right?
我知道，现在200美元是一个促销价格。
And, you know, I I will say, like, you know, $200 right now is a is a promotional price.
我们最终实际上不得不提高价格。
We we actually have to eventually charge more.
哦，价格要上涨了。
Oh, it's going it's going up.
所以在圣诞节前赶紧提交你的提示吧，耶。
So get those prompts in before Christmas yay.
赶紧提交你的提示。
Get those prompts in.
没错。
Exactly.
但说实话，如果你得花几千美元来收集数据，那么最终的成本其实并不是限制因素。
But, like, but, really, you know, it's like if you have to spend thousands of dollars gathering the data, like, the cost at the end of the day is not the limitation.
我们必须对退款非常宽容，因为人们会，你知道，
We do have to be very generous with refunds because people have, you know,
犯错，我打错字了。
make mistakes I made a typo.
对吧？
Right?
是的。
Yeah.
没错。
Exactly.
是的。
Yeah.
是的。
Yeah.
是的。
Yeah.
是的。
Yeah.
你刚才提到的那些测试，你们通过这些测试来确定这个系统能运行多久，能为科学家节省多少时间，这实际上是在复制现有的研究。
So what you just mentioned about the sort of the tests that you all ran to figure out how long this thing could run for, how much time it was saving scientists, that's about, like, sort of replicating existing research that's that's out there.
但我们从那些运营大型AI实验室的人那里听到的，是AI很快就会开始做出新颖的科学发现的可能性。
But a lot of what we hear from the people who are running these big AI labs is the possibility that pretty soon AI will start making novel scientific discoveries.
我们将开始做现有科学方法和流程无法做到的事情。
We'll start doing things that existing scientific methods and processes can't do.
我们离那一天还有多近？
How close are we to that?
实际上，那已经发生了。
I that's already happening, actually.
所以，如果你去阅读我们发布的关于Cosmos的论文，我们得出了七个结论，其中三个是现有发现的重复，另外四个则是对科学文献的新贡献，也就是新的发现。
So if you go and you read the paper that we put out about about Cosmos, we put out seven conclusions that it had come to, three of which were replications of existing findings, four of which are net new contributions to the scientific literature, like new discovery.
在这些发现中，哪一个最令人印象深刻？
And of those, what's the most impressive?
我们特别喜欢其中一个，那就是人类基因组包含数百万个遗传变异。
So I like, one of the ones that that we really like, the human genome contains millions of genetic variants.
对吧？
Right?
这些变异是不同人之间DNA的差异，与疾病相关。
These are differences between different people's DNA that are associated with disease.
但大多数情况下，我们知道某种变异与疾病有关，却完全不知道原因。
And for the most part, we know that a variant is associated with a disease, but we have no idea why.
对吧？
Right?
于是我们向Cosmos提出了这个问题。
And so we asked Cosmos.
我们给了它大量关于众多不同遗传因素的原始数据。
We gave it a bunch of raw data about a huge number of different genetic factors.
比如这些变异是什么、哪些蛋白质会结合在变异附近，诸如此类的所有信息，然后让它去识别与其中一种变异相关的2型糖尿病机制。
So, like, what the variants are, what proteins bind near the variants, right, like, all these kinds of things, and just asked it for type two diabetes to go and, you know, identify a mechanism associated with one of these variants.
它返回的结果显示，这个变异并不位于基因内部。
And it came back, and it identified this was a variant that was not in a gene.
Cosmos识别出这实际上是另一种蛋白质的结合位点。
And Cosmos identified that this is actually somewhere where a different protein binds.
它成功识别出结合的是哪种蛋白质、哪个基因被表达，并将这些与该基因的实际机制联系起来——即参与胰腺分泌胰岛素的SSR1基因。
It was able to identify what protein binds and what gene is being expressed and connected that to the actual mechanism of that gene, s s r one, which is involved in the pancreas in secreting insulin.
对吧？
Right?
好的。
Okay.
所以我的理解是，在这个案例中，你们的模型能够对现有数据进行非常复杂的推理，发现了一些人类科学家尚未注意到、甚至可能很长时间都发现不了的东西。
So so in this case is what I'm hearing that your model was able to do some very fancy reasoning over some existing data and identify something that sort of no other human scientist had had gotten around to and might not have for a really long time.
是的。
Yeah.
没错。
That's right.
好的。
Okay.
我认为科学通常包括决定收集哪些数据、收集这些数据，然后得出结论。
And and I think science generally consists of deciding what data to gather, gathering that data, and then drawing conclusions.
所以目前来说，Kosmos 主要针对的是第三步，你知道的，而且还有
And so at this point, basically, it's like step number three that Kosmos is aimed at, you know, and there's
更多的工作是第零步，也就是让特朗普政府解冻你的资金。
more work step zero, which was getting the Trump administration to unfreeze your funding.
但其他一切都对。
But everything else was right.
是的。
Yeah.
那么，当Kosmos取得这样的发现时，会发生什么？
So what happens when you get a discovery like this from Kosmos?
你们之后必须去验证它吗？
Do you have to then go validate it?
你们会把它交给一组研究人员，让他们确保这个发现确实有效吗？
Do you hand it to, like, a team of researchers who then have to, like, make sure it works?
或者接下来会发生什么？
Or, like, what happens next?
是的。
Yeah.
当然。
Absolutely.
你们必须去验证它。
You have to go and validate it.
因此，这也是论文中提到的一件事，我们详细描述了如何验证这个特定的变异体。
And so that's actually one of the things also, you know, in the paper, actually, we describe how we went and validated that particular variant.
一般来说，当人们使用它时，确实如此。当你运行一次常规分析时，首先要理解它告诉你什么，因为它刚刚完成了一项科学家们认为需要六个月才能完成的工作，而你需要花很长时间去阅读和理解它。
In general, when people are using it, yeah, you go in mean, actually, literally, when you run a customs run, the first thing you have do is you have to understand what it's telling you because it has just done something that scientists think is, like, six months worth of work, and you're gonna sit there for a long time just, like, reading and understanding it.
一旦你读完并理解了它，那么是的，你确实需要去开展各种实验，进行自己的分析，并交叉验证，以说服自己这个结果是真实的。
Once you've read it and understood it, then, yes, indeed, you're gonna go and you're gonna run, you know, various experiments, do your own analysis, cross reference to try to, like, convince yourself that this is true.
然后，根据你的研究侦探的发现，你会决定下一步行动。
And then based on what your research detective is, you'll decide next steps.
对吧？
Right?
在这种情况下，我认为这个特定发现不太可能带来一个新的药物靶点。
You know, in this case, I think it's probably low likelihood there's a new drug target, like, from this particular finding.
对吧？
Right?
但你可以将它应用于其他发现，最终或许能发现新的药物靶点，从而启动一个药物研发项目。
But you could go and you could run this on other findings, and then eventually maybe you find new drug target, you start a drug program.
你知道的？
That's you know?
我听到一些人对像Cosmos这样的模型表示担忧，认为问题并不出在研究方法上，我们之所以没有更多由AI发现或设计的药物来治愈疾病，其实并不是因为我们缺乏发现这些药物的研究手段。
So one concern that I've heard people express about models like like Cosmos is that there's this is just, like, sort of not where the roadblocks are, that the the sort of reason that we don't have more AI discovered drugs and design drugs out there curing diseases is not actually because, like, we don't have the research methods to discover those.
而是因为你需要进行临床试验，招募受试者，还要获得FDA的批准。
It's because there's, like, you gotta go to trials and you gotta recruit human subjects and you gotta get FDA approval.
这些流程所花的时间，远比药物本身的发现要长得多。
Like, all that stuff just takes a lot longer than the actual discovery of the drug.
那么，像这样的模型目前在我们的科学流程中究竟解决了哪些问题呢？
So what problems are models like these helping to solve in our scientific process right now?
确实如此。
So so absolutely.
我其实非常同意你的观点，归根结底，解决医学问题的瓶颈在于临床试验。
I I actually, like you know, I I really agree that, like, the bottleneck at the end of the day in solving medicine is basically, you know, clinical trials.
最直观的例证就是，我们目前已知能在小鼠身上治愈的疾病数量多得惊人，因为显然我们可以随意进行实验。
I mean and the easiest way to see this is if you look at the number of diseases that we, like, know how to cure in mice, right, it's, like, astronomical because, obviously, you can just, like, run experiments.
但在人类身上，事情就慢得多。
And in humans, things are just slow.
话虽如此，如果你认为制药公司目前进行的每一项实验——每一项临床试验——都是在充分掌握现有知识的前提下最优规划和设计的，那你简直是疯了。
That said, if you think that every experiment that is being run right now by pharma companies, like every clinical trial that's being run is, like, optimally planned and optimally, you know, conceived given the full state of knowledge, you are off your rocker.
嗯。
Mhmm.
对吧？
Right?
这根本不可能。
There's, like, no way.
而这些实验的花费高达数亿美元。
And those experiments cost hundreds of millions of dollars.
嗯。
Mhmm.
所以归根结底，我们确实必须进行临床试验。
And so the question is, like, we do, at the end of the day, have to run clinical trials.
那么，如何确保这些实验是我们基于所有已知信息和数据所能进行的最优秀的实验呢？
How do we make sure that those experiments are the best experiments we could possibly be running given all the knowledge that we have, given all the data we have?
我们手头有大量数据，其中蕴含着许多有待发现的洞见，但就是没人去挖掘它们，而这些最终将推动更优的实验和试验。
There's so much data that we have that has insights in it that are waiting to be found, where we just, like, do not have people to go and find them, and that's ultimately gonna feed into better experiments, better trials.
对吧？
Right?
那么，我很好奇，你觉得你的工具如何融入当今科学家的工作流程？
Well, so then I'm curious how you see, like, your tool fitting into the the workflow of today's scientists.
是那种我做完实验后，才需要你帮忙做分析的类型吗？
Is it the sort of thing where, like, I have completed my experiments, and now I want some help doing some analysis?
还是说我有一些旧实验，只做了很少的分析，想看看能不能从中再榨出更多价值？
Or is it I have all these old experiments that I only did a little bit of analysis on, and I'm curious if I can, like, sort of squeeze any more juice out of them.
或者，AI 目前在哪些方面对一线科学家特别有帮助？
Or, like, like, what other ways are you seeing the AI being, like, really good right now for a working scientist?
嗯。
Yeah.
嗯。
Yeah.
很好的问题。
Great question.
回到2019年，那时我刚完成博士学业，我有一个庞大的数据集，而我想毕业，因为我是博士生，年薪只有大约四万美元，而且外面有大量绝佳的机会，让我可以不再当博士生。
So so going back to me in 2019, which is when I was wrapping up my PhD, right, I had this gigantic data set, and I wanted to graduate because I was a PhD student, which meant that I was making, like, you know, $40,000 a year or something on and, like, there were a ton of great opportunities to go out and, like, don't be a PhD student anymore.
好的。
Okay.
所以我花了整整六个月，就坐在桌前，试图分析数据、得出结论、阅读论文。
So I spent six months literally just, like, sitting at my desk, like, trying to analyze the data and drawing conclusions, reading papers.
对吧？
Right?
目前，Cosmos 就处在这样的位置。
For right now, that's what that's where Cosmos sits in.
你可以直接拿那个数据集。
It's like, you know, you would just take that data set.
把它交给 Cosmos。
You give it to Cosmos.
它会生成大量发现。
It comes up with a lot of findings.
目前，你还需要手动去做大量工作来验证这些发现等等。
Right now, you need to go and do a bunch of manual work to validate those findings and so on.
很快，它就会直接给出发现，而你会觉得：太棒了。
Pretty soon, it's gonna come with findings, and you're gonna be like, great.
萨姆，我很想知道你能否帮我们和听众们梳理一下当前人工智能科学的整体状况。
Sam, I'm curious if you could help sort of give us and our listeners a state of the world of AI science right now.
最近，白宫宣布了一项名为‘创世计划’的举措，这是一个联邦层面的努力，旨在整合并利用联邦政府所掌握的所有数据集，开展新的科学探索。
Recently, the White House announced what it's calling the Genesis Mission, which is a federal effort to kind of corral and harness all of these datasets that the federal government is sitting on and use them to do new scientific exploring.
我们还看到许多其他努力，包括你们的项目，以及科技行业、生物技术行业在AI用于材料科学等领域的大量工作。
We also have lots of efforts, including yours, but lots of things going on in and around the tech industry, the biotech industry, people doing AI for materials science.
给我们介绍一下，目前人工智能科学领域哪些方向最热门？
Give us a sense of, like, the lay of the land of, like, what's hot right now in AI science?
目前的精力和资金都投向了哪些方向？
Where is the effort and money going?
对。
Right.
要理解人工智能与科学的现状，首先你必须从根本上明白，人工智能是关于构建模型的。
In order to understand the landscape of AI and science, the first thing, like, fundamentally that you have to understand is that AI is about building models.
对吧？
Right?
比如，什么是语言模型？
So for example, right, like, a language model like, what is a language model?
语言模型本质上是对人类语言的建模。
A language model is fundamentally a model of human language.
恰好的是，当你构建一个人类语言的模型时，它就会以某种方式学会像人类一样思考，因为人类将他们的思想编码在语言中。
It just so happens that when you build a a a model of human language, it, like, learns how to think like a human in some sense because humans, like, encode their thoughts in language.
这可以说是二十一世纪乃至有史以来最伟大的发现之一。
This is, like, one of the greatest discoveries, right, certainly of the twenty first century maybe of all time.
同样，当我们谈论人工智能与科学时，你需要思考的是你在对事物进行建模。
So similarly, when we talk about AI and science, what you have to think about is that you are modeling things.
这就是AI所做的。
That is what AI does.
而且大致分为两类。
And there are kind of two fundamental categories.
一类是模拟自然世界。
There's modeling the natural world.
对吧？
Right?
另一类是模拟科学活动的过程。
And there's modeling the process of doing science.
这两者本质上是不同的。
These things are fundamentally different.
做出这种区分的原因是，我们正在模拟科学活动的过程。
And the reason to make this distinction is because, you know, what we are doing, right, we are modeling the process of doing science.
AI用于科学的另一面是构建能够预测蛋白质结构、生成新抗体、从零开始创造新生物的模型，这些在2025年都已实现，并且势头迅猛。
The other side of the AI for science world is building models that can, for example, predict the structure of proteins, that can generate a new antibody, that can create a new organism from scratch, which are all things that have kind of, like, happened in 2025 where there's just a huge amount of momentum.
是的。
Yeah.
这说得通。
That makes sense.
我的意思是，在模拟自然世界这一部分，你提到了蛋白质折叠、新型生物。
I mean, of the things that are happening in the part of the sort of process of modeling the natural world, you mentioned protein folding, novel organisms.
作为科学家，你看到的最让你兴奋的是什么？
Like, what has most excited you as a scientist that you've seen?
所以，我认为目前最令人兴奋的，毫无疑问，是朝着我们所说的生成模型发展的趋势。
So it's absolutely what's most exciting right now, I think, without a doubt, is this trend towards what we call generative models.
这些模型能够从零开始生成具有所需特性的蛋白质、抗体或其他任何东西。
So these are things where these are models that can produce examples of, you know, proteins or antibodies or or whatever that have desired characteristics basically from scratch.
这是我们以前从未拥有的新能力，而且意义重大。
This is a new capability that we have never had before, and it's huge.
我在想，当你进行所有这些实验时，可靠性方面如何？
I I'm curious about the reliability piece as you're running all of these experiments.
你知道吗？
You know?
这周我在社交媒体上看到了这个。
I saw this going around on social media this week.
我自己复现了它。
I reproduced it myself.
如果你问谷歌，2026年是明年吗？
If you asked Google, is 2026 next year?
它说：不是。
It said, no.
2026年不是明年。
2026 is not next year.
它是后年。
It is the year after next.
所以在这样的世界里，萨姆，有些人可能会对我们将所有数据分析都交给人工智能感到担忧。
So in such a world, Sam, some people might get concerned at the idea that we're now entrusting the AI with all of our data analysis.
展开剩余字幕（还有 351 条）
那么，科学家们需要花多少时间回去重新检查人工智能的工作呢？
So how much time are scientists having to spend go back and essentially rechecking the work of the AIs?
这对他们的工作造成了多大的负担？
And what kind of tax does that place on their work?
是的。
Yeah.
这真的很有趣。
This is very funny.
我的意思是，你看，你确实得花很多时间回去检查。
I mean, look, you have to spend a lot of time going back and checking.
是的。
Yeah.
但要明确的是，无论这是人工智能做的，还是你请朋友做的，情况都一样。
But, like, to be clear, this is true regardless of whether or not an AI does it or whether you ask a friend to do it.
如果你要发表论文，你最好回去仔细检查，确保自己有信心。
If you're gonna publish a paper, you damn well better go back and check it and, like, be sure that you are confident.
而且这永远不可能达到100%。
And it's never gonna be a 100%.
对吧？
Right?
你所能达到的最好状态，就是和你自己亲自做时差不多好，而你自己做也不是100%的，因为你也不是完美的。
The best you're gonna do is you're going to get to a place where it is similarly good to if you were doing it yourself, which is not a 100% because you're not infallible.
对吧？
Right?
检查工作总是比最初生产它要快得多。
And checking the work is, like, always gonna be faster than producing it in the first place.
明白了。
Got it.
对吧？
Right?
快很多。
By a lot.
我们历史上许多重大的科学突破都源于这些奇怪的意外。
A lot of our biggest scientific breakthroughs in history have come from these kind of strange accidents.
是的。
Yeah.
这些偶然的时刻，比如青霉素在培养皿中生长出来，我们就会惊呼：天啊。
These moments of serendipity, you know, penicillin starts growing in a petri dish, so we just go, oh my god.
这真是太棒了。
It's you know, it it it this is great.
人工智能能保留这种偶然性、这些意外吗？还是会将其优化掉？
Does AI preserve that kind of serendipity, those kinds of accidents, or do they sort of optimize it away?
是的。
Yeah.
这是个很好的问题。
This is a great question.
事实上，我们目前真的还不知道。
And the fact of the matter is we just, really don't know yet.
这将成为一个非常重要且被很多人关注的核心问题。
This is gonna be a a, like, really important core question that a lot of people are asking.
你对此的直觉是什么？
What's your intuition on it?
我的意思是，我认为它们很可能确实会，因为
I mean, I I think that they probably will because
它们很可能确实会
They probably will
很可能保留
probably will preserve
保留它？
Preserve it?
保留它。
Preserve it.
因为青霉素，据我了解，基本上是某个没有抗生素的琼脂培养皿的窗口被打开了。
Because penicillin, my understanding is that, basically, like, the window was left open on some agar with, like, no antibiotic in it.
显然，当时还没有抗生素，因为这是第一个抗生素的发现。
Obviously, they didn't have antibiotics since this was the discovery of the first one.
对吧？
Right?
所以当时窗户开着，培养皿里有琼脂，一些孢子飘了进来并开始生长，他们观察到细菌的生长被抑制了。
So the window was left open with some agar and, like, you know, some spores flew onto it and began growing, and they observed that the bacteria was inhibited.
对吧？
Right?
这是一个错误。
That's a mistake.
有人搞砸了。
Someone screwed up.
对吧？
Right?
而这个错误却带来了惊人的发现，我想，你也会犯错。
And that mistake led to something fantastic, and you will have mistakes, I think.
这将被保留下来。
That will be preserved.
但与此同时，科学家们应该总是让窗户开着。
But in the meantime, scientists should always leave their windows open.
你永远不知道会发生什么
You never know what's
事情。
gonna happen.
你知道，认真说吧，当一个研究生进入学术界时，尤其是第一年的研究生，他们完全不知道该做什么。
You have no you know, seriously, though, like, there's so much when you get a graduate student in academia, right, when you get graduate students, first year graduates, they have no idea what to do.
他们完全不知道该做什么，而这正是科学进步的重要来源，因为他们会做些最随机、最古怪的事情，而任何有经验的人都不会想到去做这些。
They have no idea what to do, and that is a huge source of scientific progress because they just do the most random, kooky stuff that no one who knew anything who knows anything would ever think to do.
这实际上真的很重要。
And it's actually it's actually really important.
这就像你几乎希望你的AI科学家模型能稍微产生一些幻觉。
It's like You almost want your, like, AI scientist model to hallucinate a little bit
完全正确。
Totally.
它不会失去这种特质。
That it doesn't lose that quality
噪音。
noise.
对，没错。
Of Yeah.
对吧？
Right?
我们把这称为通过添加噪音来实现，这实际上对生物进化也很重要。
We talk about this as just like adding noise in order to this is actually important for, like, biological evolution also.
对吧？
Right?
比如，基因组中有很多噪音，这正是进化随机产生新事物的方式。
Like, the genome has a lot of noise, and that's how the the evolution randomly comes up with, like, new stuff.
嗯。
Mhmm.
有一种蛋白质，完全是随机的，什么用都没有。
Is that there's a protein that, like, is just totally random, doesn't do anything.
然后有一天，突然间，嘿，它开始起作用了，这太棒了。
Then one day, all of a sudden, oops, it does something, and that's great.
对吧？
Right?
那你怎么看待各大AI实验室的领导者，比如德米斯、达里奥和萨姆·阿尔特曼，他们说AI将在未来十年或二十年内帮助我们治愈所有或大多数疾病？
So What do you make of the leaders of the big AI labs, people like Demis and Dario and Sam Altman who are saying, you know, AI is going to allow us to cure all diseases or most diseases within the next decade or two.
十年太疯狂了。
Decade is crazy.
而且我很乐意对此持非常明确的立场，因为如果我错了，那反而是好事。
Oh, and and I'm I'm happy to take a very strong stance on this because if I'm wrong, it's a great thing.
对吧？
Right?
但如果我错了，每个人都会受益。
But if I'm wrong, everyone wins.
但十年太疯狂了。
But, like, a decade is crazy.
为什么疯狂？
Why is it crazy?
因为正如我们之前讨论的，你必须进行临床试验。
Because for the reason that we were talking about before, you have to run clinical trials.
对吧？
Right?
如果我们现在有一种药物能阻止衰老，完全停止人类在25到65岁之间的衰老，你十年内都不会知道，因为在那个年龄段，你根本无法检测出人类是否在衰老，至少需要五到十年的时间。
If we had a drug right now that prevented aging, completely halted aging in humans are on you know, between the ages of, like, 25 and 65 or something, you would not know for ten years because you can't detect in humans in that age range whether or not they're aging for, like, at least, like, you know, five or ten years.
你无法从一年到下一年察觉到自己在衰老。
You don't detect from one year to the next that you're aging.
所以你无法知道这个药物是否有效。
So you won't know if the thing is working.
不知道。
Don't know.
我高中毕业十年聚会上，有些人看起来已经很憔悴了。
Some people at my ten year high school reunion were already looking pretty rough.
这说得通。
This is fair.
真不想这么说。
Hate to say it.
我确实说了25岁。
I did say 25.
嗯。
Yeah.
25岁。
25.
好吧。
Okay.
说得通。
Fair enough.
说得通。
Fair enough.
但但但对。
But but but right.
我的意思是，你知道，所以我们必须做实验。
I mean, you know, so we have to conduct experiments.
这些实验需要时间。
Those experiments will take time.
现在，我认为三十年是非常有可能的。
Now will we like, thirty years, I think, is very plausible.
我们不知道未来可能实现什么。
We don't know what is gonna be possible.
我们不知道是否有可能阻止衰老。
We don't know if it's possible to halt aging.
我们不知道是否有可能治愈所有疾病之类的。
We don't know if it's possible to, like, cure all diseases or whatever.
但在现在和未来三十年之间，我认为你会看到巨大的进步，
But there like, between now and thirty years from now, I think you should expect to see a humongous leap forward in
在这一点上，让我深入探讨一下，因为我觉得有些人听到这个可能会认为，这本质上是一个监管问题，即我们没有像FDA那样能够衡量这些的体系。
terms of Let me drill in in on that a bit there because I think some people might hear that in saying that, like, this is essentially a regulatory issue that, like, we just don't have, you know, the the FDA set up to to measure this.
不过，我对实验方面很感兴趣。
I'm curious about the the experimental side of it, though.
对吧？
Right?
因为我的理解是，我们实际上没有足够的生物学家来开展所有可能的实验，可能也没有足够的资金来资助这些实验。
Because my understanding is, like, we don't really have enough biologists to run all the experiments that we we might not have, like, the funding to to fund the experiments.
你确实提到了，有些实验实际上需要很长时间才能完成。
And you did raise the point that some of these experiments does actually take a long time to run.
对吧？
Right?
所以，在你看来，有哪些因素会让这件事变得特别困难呢？
So, like, what are all of the factors that in your mind are just gonna make it so hard to
天哪。
My gosh.
你得去实际操作，即使你已经有了一个想在人体上测试的分子，并且知道要选哪些人做测试，你还是得去生产它。
You have to go and you have to, like you know, even supposing you have a molecule that you wanna test in a human and you know which humans you wanna test it in, you have to go and make it.
对吧？
Right?
人体是庞大的。
Humans are big.
它们需要大量的这种物质。
They require, like, a lot of it.
你得确保它的纯度足够高，才能真正用于人体。
You have to make sure it's, like, high enough grade that you can actually put it into a human.
你得找到病人，这意味着要与医生建立关系。
You have to find the patients, which means forming relationships with the doctors.
对吧？
Right?
实际上，你知道，你得等到有足够的愿意参与的患者。
Actually, you know, waiting until you have enough patients who are willing to do it.
对于许多疾病来说，患者根本没多少。
For many diseases, like, there just aren't that many patients.
所以找到患者很难。
And so finding the patients is hard.
对吧？
Right?
然后你还得真正给他们用药。
And it just and then you have to actually dose them.
你得等待并观察会发生什么。
You have to wait and see what happens.
对吧？
Right?
即使没有监管，这个过程也会很慢。
Even with no regulation, it would be slow.
而且，目前几乎没有任何AI可以缩短这些步骤。
And There's no AI shortcut for almost any of that, at least not right now.
没错。
No.
是的。
Yeah.
比如，AI能让我们做到的是，发现那些我们其实已经拥有足够信息去发现的东西。
Like, there the there what AI will allow us to do is it will allow us to discover a lot of things where we already have the information to discover it.
我们只是还没搞明白而已。
We just haven't figured that out yet.
你不应该指望有一天，比如，你问GPD七号如何治愈阿尔茨海默病，它就能直接告诉你答案。
You should not expect that you're one day gonna, like, get g p d seven and just, like, ask it how to cure Alzheimer's, and it will just tell you.
我的预期是，目前的知识储备还不足够。
My expectation is that there is not enough knowledge.
对吧？
Right?
即使拥有无限的智慧，我们也没有足够的知识从根本上解决它。
We do not have enough knowledge to solve it in principle even with infinite intelligence.
对吧？
Right?
即使拥有无限的智慧，世界上仍有一些事情是我们尚未知晓的，必须通过实验来发现。
Like, with infinite intelligence, there would still be some things that are just not known about the world where we have to conduct the experiments to see.
你可以根据已知的所有信息规划出最佳的实验，但你不可能凭空直接推导出答案。
You'll be able to plan the best possible experiment given everything that's known, but you will not just be able to, like, you know, de novo kind of figure it out.
对吧？
Right?
凯西，我学过拉丁语。
Casey, I I took Latin.
这意味着‘从头开始’。
That means from new.
哦，谢谢你。
Oh, thank you.
是的。
Yes.
谢谢
Thank
你。
you.
这省去了我一次谷歌搜索的步骤。
That's saved me a step of googling.
回来后，我们将和我们的嘉宾萨姆·罗德里格斯玩一个‘过度炒作还是被低估’的游戏。
When we come back, we'll play a game of overhyped or underhyped with our guest, Sam Rodriguez.
这不算严格意义上的科学，但我很好奇你怎么看，萨姆。
This isn't quite science per se, but I'm curious what you make of this, Sam.
所有大型人工智能实验室都痴迷于数学。
All of the big AI labs are obsessed with math.
是的。
Yeah.
赢得国际数学奥林匹克竞赛，获得金牌成绩，解决那些未被证明的数学定理。
With winning the International Math Olympiad, with putting up a gold medal score, with solving these unproven math theorems.
我对这一点有个看法，那就是我相信这些实验室里的人大多在高中时是参加数学竞赛的佼佼者，参加过IMO并取得了不错的成绩。
And I have a take about this, which is that I believe that this is because these labs are filled with people who were themselves competitive mathletes in high school and took part in the IMO and did pretty well.
而很多这些人认为，人工通用智能（AGI）不过是他们自己的一种更聪明的版本。
And a lot of those people think that, like, AGI will just sort of be, like, a slightly smarter version of them.
但我很好奇。
But I'm curious.
为什么这些地方如此痴迷于数学，把它当作最先取得重大进展的领域之一？
Like, why are these places so obsessed with math as being one of these sort of first places that they wanna make a lot of progress?
有两个原因。
There are two reasons.
我认为其中一个原因正是你刚才说的。
I think that one of the reasons is exactly what you just said.
只是因为熟悉。
It's just familiar.
对吧？
Right?
但另一个原因是你可以衡量进展。
But the other reason is that you can measure progress.
对吧？
Right?
所以从根本上说，推动机器学习进展的因素，很大一部分就是基准测试。
So ultimately, like, what drives progress in machine learning well, a big part of what drives progress is benchmarks.
在数学中，你可以判断你的证明是否正确，而且有无穷多的问题可以去证明。
With math, you can tell whether or not your proof is right, and there's kind of, like, an infinite number of things to go and prove.
所以很容易判断你是否在进步。
So it's just, like, really easy to tell whether or not you're getting better.
像国际数学奥林匹克竞赛这样的赛事，就提供了绝佳的机会。
And things like the IMO just present, like, great opportunities.
相比之下，如果你看看最近一些最大的突破，比如今年人工智能在生物学领域最重要的突破，像Chai Discovery和NABLA这样开发出全新抗体生成模型的成果。
By contrast, if you look at, like, some of the biggest breakthroughs recently, you know, biggest breakthroughs this year in AI for biology, right, things like, you know, Chai Discovery, NABLA coming up with these, like, extremely good models for producing antibodies de novo.
对吧？
Right?
巨大的突破。
Huge breakthrough.
但最终，它们的成功将体现在这些成果获得人体批准时，而这可能还需要五年左右的时间。
But, like, ultimately, the win for them is going to be, like, when it's approved in a human, and that might be another five years or something.
Arc研究所首次实现了从零开始设计生物体。
Arc Institute putting out, like, the first time anyone has designed an organism from scratch.
他们设计了一种噬菌体。
They designed a bacteriophage.
这是一种感染细菌的病毒。
It's a kind of virus that infects bacteria.
令人难以置信。
Incredible.
对吧？
Right?
但更难评估。
But, like, just harder to evaluate.
它到底有多好？
Like, how good is it?
你不会把它释放到自然界中，等等。
Like, you're not gonna release it into the wild, and so etcetera.
它更难评估，而IMO就非常清晰。
Like, it's harder to evaluate, whereas, like, the IMO is just, like, super clean.
因此，我们认为的一个重要问题是：我们如何才能建立明确的基准，来衡量我们在科学上是否做得好？
And so I think that's one thing that we think about a lot is just, like, you know, how do we get really clear benchmarks that we can pursue to measure whether or not we're doing a good job at science?
我有一个答案。
I have an I have an answer here.
国际癌症治愈奥林匹克竞赛。
International Cancer Curing Olympiad.
我喜欢这个主意。
I like that.
我们开始吧？
Should we start this?
我觉得这太棒了。
I think that would be great.
我们可以给获胜者颁发奖牌。
We can give people a medal if they win.
我们行动起来吧，实验室的各位。
Let's get on it, Labs.
当这些公司的首席执行官或领导者做出声明，说我们将在未来十年、十五年或他们给出的任何时间框架内用人工智能治愈所有疾病时。
So when the CEOs or leaders of these companies make these statements about how we're gonna cure all disease using AI in the next ten years or fifteen years or whatever that whatever timeline they give.
他们是不了解瓶颈所在吗？
Are they doing that because they don't understand the bottlenecks?
我的意思是，这些人都是非常聪明的。
I mean, these are very smart people.
那么，他们没看到的是什么？还是说这仅仅是一种营销手段？
So what are they not seeing, or are they just doing this as sort of a marketing exercise?
这是否是为了让那些原本对AI感到恐惧的人对它产生兴奋？
Is this an attempt to get people excited about AI who might otherwise be freaked out about it?
他们为什么要做这些预测？
Why are they giving these projections?
不。
No.
听好了。
Look.
我的意思是，理性的人对此可能有不同看法。
I mean, I think that they are reasonable people could disagree.
有很多理由可以论证，比如模型真的会变得极其聪明，它们会找到方法在进行临床试验前衡量我们是否取得了进展，从而加快迭代周期。
There are lots of reasons why you could argue that, like, actually, the models will get super smart, and they will figure out ways to measure whether or not we're making progress before you run a clinical trial, and that will increase the iteration cycle.
对吧？
Right?
我的意思是，关于这一点，是有合理论点的。
Like, there are reasonable arguments to be made about that.
对吧？
Right?
我的意思是，我们可能再也不做完整的临床试验了。
Like, you know, that we are just gonna not do full clinical trials anymore.
我们只会使用生物标志物。
We'll just, like, use biomarkers.
这并不疯狂，这也是我可能出错的一个方面。
Like, that's not crazy, and that's one way that I I could be wrong.
也许十年后，我们真的能治愈所有疾病。
And maybe in ten years, we do have cures for all diseases.
所以这也是其中一部分。
So that's part of it.
很明显，其中一部分原因是他们想炒作这件事。
Like, obviously, there's there's part of it, which is that they want to hype the the thing.
其中一部分是，你知道，萨姆·阿尔特曼真的能深刻理解，把一个小分子药物从研发到临床生产并规模化制造，到底需要什么吗？
Part of it is that, you know, does Sam Altman, like, really intimately understand, like, what it takes to go and manufacture, like like, scale up manufacturing for a small molecule to put into the clinic?
大概不能。
Like, probably not.
对吧？
Right?
所以这其实是一种混合情况。
So there's it's a mixture.
我不认为其中有任何一方是出于恶意。
I I don't think any of it's in bad faith.
只是大家都太兴奋了。
It's just people are very excited.
总有一天，现实会与之发生碰撞。
There will be a little bit of a collision with reality at some point.
我们会看到这种碰撞究竟发生在哪儿。
We're gonna see exactly where that is.
但无论如何，未来将会非常精彩。
But regardless, the future is gonna be awesome.
对吧？
Right?
在2025年的今天，你觉得AI工具对科研人员的生活改变了多少？一年后，你预计会有多大不同？
At this moment in 2025, how much do you think AI tools have changed the life of a working scientist, and how different do you expect that will be a year from now?
我认为你会对AI工具至今尚未产生多大影响感到震惊。
I think that you'd be shocked to the extent that they have not yet.
科学家总体上是非常保守的人，因为当你在做实验时，尤其是在生物学领域，你从来无法完全确定结果的原因。
Scientists in general are extremely conservative people because if you're running experiment, you, like, never actually fully know in in biology, at least.
你通常并不完全理解，为什么实验成功了，或者为什么失败了。
You usually do not, like, fully understand, like, why the experiment works and why not.
有些做法是你从过去的实验方案中继承下来的，大家就这么做。
There are some things that you've inherited from protocols that you've run-in the past and where it's like, we do it this way.
你可以去验证，但可验证的事情太多了。
You could go and test it, but there are way too many things to test.
所以你就被锁定在自己的方法里了，这就是有效的方法。
So you're just kind of, like, locked in in your methods, and and it's what works.
你只想做有效的事情。
And you just wanna do what works.
因此，出于这个原因，生物学家们采纳新方法的速度很慢。
And so for that reason, like, biologists just adopt new methods slowly.
我认为，全球大多数实验室仍然在用以前的方式做科学，而且在可预见的未来还会继续这么做，这也没关系。
I think most labs around the world are still probably doing science the way they've done it before and probably will for continue to do so for a while, and that's okay.
你知道，在编程方面，很多人已经开始采用了，因为在生物学中，编程历来是一个巨大的瓶颈。
You know, one place, I think, with coding, a lot of people are already adopting it because in biology, historically, coding has been a big bottleneck.
现在，那些原本不会编程的生物学家，可以通过云代码、OpenAI的模型、Gemini等工具轻松进行大量编程，这是一次巨大的突破。
It's a huge unlock now that biologists who didn't know how to code can, like, do a lot of coding using Cloud Code, using OpenAI's models, Gemini, etcetera.
所以这是一个巨大的突破。
So that's a huge unlock.
我认为这将很快得到广泛采用。
Think that that's gonna see a lot of adoption quickly.
文献检索，对吧？能够解析科学文献的庞大数量，这是一次巨大的突破。
Literature search, right, like being able to parse the immensity of the scientific literature, that's a huge unlock.
这将被迅速采用。
That's gonna get adopted very quickly.
对吧？
Right?
我们正在开发的这类工具，还属于前沿领域。
The tools like what we're building are are, like, a little bit more frontier.
最终，当人们看到其他人使用这些工具并取得优异成果时，就会采纳它们。
Ultimately, people will adopt them when they see other people using them and getting great results.
萨姆，我们能和你玩个小快问快答游戏吗？
Sam, can we play a little lightning round game here with you?
好的。
Yeah.
我们把这个环节叫做‘过度炒作与被低估’。
We're calling this one overhyped, underhyped.
我们会告诉你一件事，你从你的科学观点出发，判断它是被高估了还是被低估了。
So we'll tell you something, and you tell us whether in your scientific opinion it is overhyped or underhyped.
很好。
Great.
准备好了吗？
You ready?
好了。
Yeah.
氛围证明。
Vibe proving.
这指的是AI系统外出撰写数学证明。
This is when AI systems go out and, like, write math proofs.
如果非得选一个，我可能觉得被高估了。
Probably just if I have a forced choice, probably overhyped.
这很好，我是说，它作为AI进展的推动力很棒。
It's great for I mean, it's great as, like, a progress driver in AI.
就像这样，我们可能并不会因为擅长它而产生其他方面的影响。
It's like and we'll probably have not you know, being good at it will probably have implications elsewhere.
但它本身真的有用吗？
But is it itself that useful?
我不确定。
I'm not sure.
机器人用于AI实验室自动化。
Robotics for AI lab automation.
用机器人来自动化AI实验室？
Robotics for automating AI labs?
或者
Or
是的。
Yes.
或者为了
Or for
用于自动化科学实验室。
for automating scientific labs.
机器人用于自动化科学实验室。
Robotics for automating scientific labs.
我觉得被适度夸大了。
I think appropriately hyped.
这将带来彻底的变革。
It is going to be totally transformative.
这项技术还远未成熟。
The technology is not at all there yet.
我们还有很多工作要做，但确实，可能被适度夸大了。
There's a lot that we need to do, but, like, yeah, probably appropriately hyped.
AlphaFold 3？
AlphaFold three?
这是一个有趣的话题。
That's an interesting one.
我的意思是，我认为可以说它们被低估了，因为虽然围绕这些蛋白质结构模型的炒作很多，但它们仍然可能具有极其变革性。
I mean, I think that I would say probably, like, underhyped in that I think, like, all of the protein structure models, there's a lot of hype around them, but they're still they're still probably like, they're gonna be extremely transformative.
所以也许我会说，它们被低估了。
So maybe I would I would I would say probably underhyped.
不过，围绕它的炒作确实很多，所以很难做出判断。
It's a hard there's a lot of hype around it, though, so it's a hard decision to make.
但是
But
虚拟细胞，比如今年夏天我们听帕特里克·科利森谈过阿尔克研究所如何构建虚拟细胞。
Virtual cells, like, we heard from Patrick Collison this summer about what the Arc Institute has done with making a virtual cell.
这被高估了，但有特定原因。
This is overhyped, but for a specific reason.
对吧？
Right?
阿尔克研究所开发的这些模型很棒，真的很棒。
Like, the models that they're building at Arc are awesome, the models.
像New Limit、陈-扎克伯格 Initiative 这些地方，还有许多其他伟大的公司和机构，也在做类似的事情。
And they're doing similar things at, like, New Limit, Chan Zuckerberg, right, like, many of these places many of these great companies and great organizations are doing it.
我认为，虚拟细胞这个概念有点被高估了。
I think that, like, it a virtual cell, like, is a little bit that's, like, a little bit over is is that's that's overhyped.
对吧？
Right?
这些模型最终只是模拟了某种非常具体的东西。
Like, ultimately, those model that kind of model models something, like, very specific.
真正构建一个虚拟细胞，也就是在计算机中模拟一个细胞，是一个了不起的目标。
Like, actually building, like, a true virtual cell, like, being able to simulate a cell in a computer is an amazing goal.
我们离那个目标还很遥远。
We are very far away from that.
量子计算。
Quantum computing.
被高估了。
Overhyped.
脑机接口。
Brain computer interfaces.
我也是，天啊。
I'm also oh, man.
这个真的很难。
This one's really hard.
我会说，被过度炒作了。
I will I'm gonna say overhyped.
我非常相信脑机接口。
I'm a huge believer in in BCIs.
我认为，像科幻作品中所想象的那种高效脑机接口，比人们想象的要遥远得多。
I think, like, effective BCIs or the way that we imagine them in sci fi are further out than people imagine.
即使是像Neuralink这样的公司也在取得惊人的进展。
Even like, Neuralink is making amazing progress.
是的。
Yeah.
凯西现在脑子里就有一个。
Casey's got one in his head right now.
是的。
Yeah.
它出问题了。
It's it's on the fritz.
是的。
Yeah.
有很多很棒的
There are a lot of great
有很多了不起的人在那里取得了进展，但我认为，这比人们想象的要更遥远。
there are a lot of great people who are making progress there, but it's further out, I think, than people think.
我们快要接近年底了。
So we're we're nearing the end of the year.
如果你能稍微反思一下，你认为今年最重要的三项由人工智能推动的科学进展是什么？
If we can put you in a bit of a reflective mode, what do you think were the top three AI driven scientific advancements this year?
是的。
Yeah.
我认为，老实说，今年是代理之年。
I think that, honestly, like, this year is the year of has been the year of agents.
今年是人们发现代理的一年。
This was the year when people discovered agents.
所以，我真心地必须把自己和我们团队列入这个名单。
And so I I do, like, you know, in good faith, have to put myself have to put us on the that list.
同时，与谷歌的科学家们一起，我的意思是，我们并不是唯一在做这件事的人。
When also with Google coscientists, I mean, we're not the only people who are working on this.
你知道，谷歌一直做得很好。
You know, Google has been doing a great job.
还有很多其他人在做。
There are a bunch of other people.
所以，AI代理在科学领域，绝对是。
So AI agents for science, definitely.
然后，生成式设计现在正迎来巨大热潮。
And then, like, generative design is just having a huge moment.
对吧？
Right?
其他方面可能是Chai所做的工作、Nabla所做的工作，以及许多其他人在从头设计抗体方面的研究。
So the other ones would probably be the work that Chai has been doing, the work that Nabla has been doing, and many others on de novo antibody design.
顺便说一句，我很高兴你之前在节目中定义了‘从头设计’这个词。
I'm really glad you defined de novo earlier in the broadcast, by the way.
这个词已经多次出现了。
It's come up a lot.
是的。
Yes.
抱歉。
Sorry.
当我提到‘从头设计’时，我只是指纯粹从零开始生成。
When I say de novo, I just mean, like, literally, you just, it generates it from scratch.
你什么也不给它。
You don't give it anything.
对吧？
Right?
你只需要给它一个你想让它结合的目标，它就能从零开始生成。
You just, like or you give it a target that you want to bind to, and it generates it from scratch.
这太重要了，因为像Chai、Nabla等公司所追求的愿景是：你可以直接说，我们要治愈这种疾病，就必须靶向这个蛋白质。
This is huge because, like, basically, the the promise that companies like Chai, Novela, and so on are going after is a world in which you can say, like, we know to cure this disease, we have to target that protein.
你点一下按钮，就能得到一个明天就能用于人体的抗体。
You click a button, and you have an antibody that you can go and put in humans tomorrow.
这太重要了。
It's huge.
它省去了人们在疾病治疗前必须做的大量工作。
It cuts out a enormous amount of what people had to do pre disease.
所以这是一个巨大的突破。
So that's a huge one.
第三个是，我觉得Brian He、Patrick Shu等人在Arc研究所所做的，关于生成生物体的工作，抱歉。
And the third one, I just think, like, what Brian He, Patrick Shu, and so on at the Arc Institute have done with, like, generating organisms sorry.
生成生物体，我们可以说
Generating organisms We can say
我们现在明白这意味着什么了。
we know what it means now.
这才是关键。
That's the important thing.
这是我们这周的‘Pee Wee's Playhouse’本周词汇。
This is our, like, Pee Wee's Playhouse word of the week
本周。
this week.
从头设计生物体，有用吗？
The de novo design of organisms, is it useful?
我不知道。
I don't know.
这很厉害吗？
Is it awesome?
绝对是的。
Like, absolutely.
这真是一个巨大的突破。
It's so it's such a big breakthrough.
萨姆，明年我们应该关注什么？
And, Sam, what should we be watching for next year?
你对2026年可能到来的哪些进展感到兴奋？
What are you excited about that may be coming down the pipe for 2026?
老实说，再次强调，代理系统将迎来爆发式增长。
Honestly, it is, again, going to be the agents that see an explosion.
我们现在正处于这条S型曲线的起点，而且这种趋势将持续下去。
We are right now at, like, the beginning of that s curve, and that is going to continue.
也许一年前，我会告诉人们，我认为到2026年或2027年，科学界产生的绝大多数高质量假设都将由我们或我们所构建的代理系统生成。
Maybe a year ago, I would tell people that I thought in 2026 or maybe 2027 that, like, the majority of the high quality hypotheses that are generated by the scientific community would be generated, like, by us or by, like, agents that are, the ones that we're building.
当我2024年这么说的时候，我以为自己是在过度炒作。
And when I said it in 2024, I thought I was overhyping.
对吧？
Right?
我的意思是，我只是觉得我需要一些炒作。
I mean, but I was just like, I need some hype.
到目前为止，这可能是真的。
At this point, it may be real.
我的意思是，我认为2026年实现这一点已经很激进了。
I mean, I think 2026 would be ambitious for that.
我的意思是，这很了不起吧？
I mean, that's a huge right?
大部分优质假设由代理生成，这是一次巨大的飞跃。
For the majority of the good hypotheses that come out to be made by agents, that's a huge leap.
但2027年，没错，老兄。
But, like, 2027, yeah, man.
我的意思是，2026年将是我们看到这些代理开始渗透到方方面面的一年。
I mean, 2026 is gonna be the year when we just see these agents start to, like, infiltrate everything.
对吧？
Right?
渗透到实验室，渗透到人们的日常生活。
Infiltrate labs, infiltrate people's normal life.
我的意思是，这已经发生了。
I mean, it's already happening.
酷。
Cool.
是啊。
Yeah.
嗯，我期待着这一天。
Well, I look forward to it.
萨姆，非常感谢你给我们上了这堂我们学校显然没教过的科学课。
Sam, thank you so much for giving us the science education that we clearly didn't get in school.
是的。
Yeah.
你确实给我们带来了一些全新的思考方向。
You've really given us some de novo things to think about.
我很感激
I appreciate
这个。
that.
很好。
Good.
谢谢大家。
Thank you, guys.
你们。
You.
《Heartfork》由蕾切尔·科恩和惠特尼·琼斯制作。
Heartfork is produced by Rachel Cohen and Whitney Jones.
我们由珍·波扬剪辑。
We're edited by Jen Poyant.
本期节目由威尔·皮施尔核对事实，由克里斯·伍德制作。
Today's show was fact checked by Will Pischel and engineered by Chris Wood.
原创音乐由黛安·王、罗汉·内马斯托、艾莉莎·莫克利和丹·鲍威尔创作。
Original music by Diane Wong, Rohan Nemastow, Alyssa Moxley, and Dan Powell.
视频制作由萨沃·罗凯、帕特·冈瑟、杰克·尼科尔和克里斯·肖特完成。
Video production by Sawyer Roquet, Pat Gunther, Jake Nichol, and Chris Schott.
你可以在YouTube上观看本集完整内容，网址为 youtube.com/hardfork。
You can watch this whole episode on YouTube at youtube.com/hardfork.
特别感谢保拉·舒曼、谭培盈和达莉亚·哈达德。
Special thanks to Paula Schuman, Puiwing Tam, and Dahlia Hadad.
你可以发送邮件至 HardFork@NYTimes.com 与我们联系。
You can email us at HardFork@NYTimes.com.
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。