谷歌DeepMind首席执行官Demis Hassabis：AI的下一次突破、通用人工智能时间表及谷歌AI眼镜赌注

谷歌DeepMind首席执行官德米斯·哈萨比斯加入我们，讨论从当前状态通往通用人工智能（AGI）的道路、谷歌AI眼镜的推出时间，以及AI进展的速度能否维持当前水平。

Google DeepMind CEO, Demis Asabas, joins us to talk about the path from here to AGI, when Google's AI glasses are coming, and whether the pace of AI progress can keep up at this rate.

这些内容将在稍后播出。

That's coming up right after this.

欢迎收听在达沃斯特别制作的《大科技》播客。

Welcome to a special edition of big technology Podcast from Davos.

我是亚历克斯·坎特罗维茨，今天我有幸邀请到谷歌DeepMind的首席执行官德米斯·哈萨比斯作为特别嘉宾。

I'm Alex Kantrowitz, and I'm joined today by a special guest, Demis Esteves, the CEO of Google DeepMind.

德米斯，欢迎再次做客我们的节目。

Demis, welcome back to the show.

很高兴来到这里。

Great to be here.

一年前，人们确实对AI进展是否正在放缓提出了质疑。

A year ago, there were real questions about whether AI progress was tailing off.

当时，询问大型语言模型是否会遇到瓶颈成了一种潮流。

It was in fashion to ask whether LLMs were going to hit a wall.

这些疑问似乎已经得到了解答。

And those questions seem like they've been settled.

过去一年里，取得了巨大的进展。

There's been a tremendous amount of progress over the past year.

您能否告诉我们，具体发生了什么，让人工智能行业从去年的那个充满疑问的时刻，发展到今天这个地步？

Could you tell us what specifically has happened that's gotten the AI industry from that moment of question last year to the point that it is today?

对我们内部来说，我们从未质疑过这一点，这一点要澄清清楚。

Well, for us internally, we were never questioning that just to be clear.

我认为我们一直都在看到显著的改进。

I think we've always been seeing great improvements.

所以我们对为什么外界会有这样的疑问感到有些困惑。

So we were a bit puzzled by why there was this question in the air.

其中一部分原因在于，人们担心数据会用尽，而确实，所有可用的数据都已经被用过了。

I mean, some of it was to do, people were worried about data running out and there is, you know, some truth in that is all the data had been used.

我们能否创造出有用的合成数据来用于学习？

Can we create synthetic data that's gonna be useful to learn from?

但事实上，你可以在现有的架构和数据中挖掘出更多的潜力。

But actually it turns out you can ring more juice out of the existing architectures and data.

所以我认为还有很多空间。

So there's plenty of room, I think.

我们仍在预训练、后训练和推理范式中看到这一点，以及它们如何相互配合。

And we're still seeing that in both the pre training, the post training and the thinking paradigms and also the way that they all kind of fit together.

因此，我认为仅靠我们已知的技术，通过调整和在此基础上创新，仍然有很大的提升空间。

So I think there's still plenty of headroom there just with the techniques we already know about and tweaking and kind of innovating on top of that.

好吧，以下是怀疑者会说的话。

Alright, here's what a skeptic would say.

对。

Yeah.

已经有很多技巧被叠加在大语言模型之上。

That there have been a lot of tricks that have been put on top of LLMs.

我经常听到关于框架、编排以及AI使用工具搜索网页的说法，但它不会记住自己学到的内容。

I hear often about scaffolding and orchestration and AI that can use a tool to search the web, but but it won't remember what it learns.

一旦你关闭会话，它就忘记了。

As soon as you close that session, it forgets.

是

Is

这仅仅是大语言模型范式的局限性吗？

that just a limitation of the large language model paradigm?

嗯，我认为确实存在这样的问题，我 definitely 认为在我们实现通用人工智能之前，可能还需要一到两个重大突破。

Well, look, I think there is, and I'm definitely subscriber to the idea that maybe we need one or two more big breakthroughs before we'll get to AGI.

我认为像持续学习、更好的记忆、更长的上下文窗口，或者说更高效的上下文窗口，才是更准确的表达方式。

And I think there are along the lines of things like continual learning, better memory, longer context windows, or perhaps more efficient context windows would be the right way to say it.

所以不要存储所有内容，只存储重要的部分。

So don't store everything, just store the important things.

那样会高效得多。

That would be a lot more efficient.

大脑就是这样做的，还有更好的长期推理和规划能力。

That's what the brain does and better long term reasoning and planning.

现在还有待观察，仅仅是扩大现有想法和技术是否足以实现这一点。

Now it remains to be seen whether just sort of scaling up existing ideas and technologies will be enough to do that.

或者我们需要一到两个真正重大的突破性创新。

Or we need one or two more really big insightful innovations.

如果你非要我选，我可能会站在后者这一边。

I'm probably if you were to push me, I would be in the latter camp.

但我觉得，无论你支持哪一派，我们都需要大型基础模型作为最终AGI系统的关键组成部分，这一点我确信无疑。

But I think no matter what camp you're in, we're gonna need large foundation models as the key component of the final AGI systems of that I'm sure.

因此，我不认同像扬·莱昆这样的人，他认为这些模型已经到了某种死胡同。

So I'm not a subscriber to someone like Yan Lecun who thinks, you know, that they're sort of at some kind of dead end.

在我看来，唯一的争议在于：它们是关键组成部分，还是唯一组成部分？

I think the only debate in my mind is are they a key component or the only component?

所以我认为，问题就在这两种选择之间。

So I think it's between those two options.

对我而言，我们拥有的一个优势是拥有深厚而丰富的研究储备。

And for me, is one advantage we have of having such a deep and rich research bench.

我们可以全力以赴地同时推进这两方面，即扩大当前的范式和理念。

We can go after both of those things at maximum with maximum force, both, you know, scaling up the current paradigms and ideas.

我说的扩大规模，其实也包含创新，比如预训练，我认为我们在这一块非常强大。

When I say scaling up, also involves innovation, by the way, pre training, especially I think we're very strong on.

还有那些全新的、天马行空的创意，比如新架构之类的东西——过去十年里，谷歌和DeepMind发明的那些，当然也包括Transformer。

And then really new blue sky ideas for new architectures and things, you know, the kinds of things we've invented over the last ten years as Google and DeepMind, you know, of course, including transformers.

带有大量硬编码内容的系统能被称为通用人工智能吗？

Can something with a lot of hard coded stuff ever be considered AGI?

不，我认为这要看你如何定义‘大量’。

No, I think well, depends what you mean by a lot.

我对混合系统非常感兴趣，我称之为神经符号系统。

I think that I'm very interested in hybrid systems is what I would call them or neuro symbolic.

有时候人们会举AlphaFold、AlphaGo作为这类系统的例子。

Sometimes people call them, you know, AlphaFold, AlphaGo are examples of that.

因此，我们一些最重要的工作将神经网络和深度学习与蒙特卡洛树搜索等技术结合起来。

So some of our most important work combines neural networks and deep learning with things like Monte Carlo tree search.

所以我认为这是可能的。

So I think that could be possible.

我们正在做一些非常有趣的工作，利用大语言模型结合进化方法，比如AlphaEvolve，来真正发现新知识。

And there's some very interesting work we're doing building using the LLMs with things like evolutionary methods, AlphaEvolve to actually go and discover new knowledge.

你可能需要超越现有方法的东西。

You may need something beyond what the existing methods do.

但我认为学习是AGI的关键部分。

But I think learning is a critical part of AGI.

这几乎是其定义性特征。

It's actually almost the defining feature.

当我们说‘通用’时，我们指的是通用学习。

When we say general, we mean general learning.

它能否学习新知识？能否在任何领域中学习？

Can it learn new knowledge and can it learn across any domain?

这就是‘通用’的部分。

That's the general part.

所以对我来说，学习与智能是同义词，一直以来都是如此。

So for me, learning is synonymous with intelligence and always has been.

好吧，如果学习等同于智能，而这些模型仍然不具备持续学习的能力。

Okay, so if learning is synonymous with intelligence and these models still don't have the ability to continually learn.

就像我前面说的，它有金鱼的记忆。

Like I said earlier, it has goldfish brain.

是的。

Yeah.

它可以搜索互联网，然后好像找到了答案。

It can search the internet and it can be like, figured this out.

是的。

Yeah.

但它不会改变模型本身。

But it doesn't change the model.

它只是会在这之后就忘记

It's just, it will forget it after the

会话。

session.

对。

Right.

你有没有关于如何解决持续学习问题的理论？

You have a theory as to how continual, the continual learning problem can be solved?

你愿意跟我们大家分享一下吗？

And do you want to share it with us all?

我

I

我可以给你一些线索。

can give you some clues.

我们正在为此付出巨大努力。

We are working very hard on it.

我们过去在类似AlphaZero这样的项目上做过一些工作，你知道，它们从零开始学习AlphaGo，AlphaGo Zero也在已有知识的基础上继续学习。

We've done some work on, you know, I think the best work on this in the past with things like AlphaZero, you know, that learn from scratch versions of AlphaGo, AlphaGo Zero also learnt on top of the knowledge it already had.

因此，我们已经在更狭窄的领域中实现了这一点。

So we've done it in much narrower domains.

你知道，游戏显然比混乱的现实世界简单得多。

You know, games are obviously a lot easier than the messy real world.

所以，这些技术是否真的能扩展并泛化到现实世界和实际问题上，还有待观察。

So it remains to be seen if that those kinds of techniques will really scale and generalize to the real world and actual real world problems.

但至少，我们已知的方法已经能实现一些相当令人印象深刻的事情。

But at least the methods we know can do some pretty impressive things.

因此，现在的问题是，我们能否将这些方法——至少在我看来——与这些大型基础模型结合起来？

And so now the question is, can we blend that, at least in my mind with these big foundation models?

当然，基础模型在训练过程中是在学习的，但我们希望它们能在真实环境中持续学习，包括个性化等方面，我认为这一定会实现。

And so of course, the foundation models are learning during training, but we would love them to learn, you know, out in the wild and including things like personalization, I think that's gonna happen.

我认为，打造一个出色的助手的关键在于，它能理解你，并成为为你服务的技术。

And I feel like that's a critical part of building a great assistant is that it understands you and it works for you as technology that works for you.

我们就在上周发布了这一功能的首个版本。

And we've released our first versions of that just last week.

个人智能是朝这个方向迈出的第一步。

Personal intelligence is the sort of first baby steps towards that.

但我认为，要实现这一点，不能仅仅把你的数据放在上下文窗口里。

But I think to have it, you wanna do it more than just having your data in the context window.

你需要更深入一些的东西，就像你所说的，随着时间推移改变模型。

You wanna have something a bit deeper than that, which is as you say, changes the model over time.

这正是你理想中应该拥有的。

That's what ideally you would have.

但这种技术目前还没有被突破。

And that technique has not been cracked yet.

我们已经提到了几次通用人工智能。

We've brought up AGI a couple times.

所以让我问你一个问题，因为我在年底时和萨姆·阿尔特曼聊过。

So let me put this to you because I was speaking with Sam Altman towards the end of the year.

我问他，我觉得你似乎在表达两层意思。

And I asked him, I was like, you know, you seem to be saying two things.

我们还没有达到AGI，但每次他谈论GPT模型能做什么时，似乎都符合他的定义。

We're not at AGI yet, but every time he talks about what GPT models can do, it seems like it fits his definition.

他说AGI的定义过于模糊。

And he said that AGI is under defined.

他希望每个人都能认同的是，我们已经不知不觉地越过了AGI，正朝着超级智能前进。

What he wishes everybody could agree to was that we've sort of wooshed by AGI and we move towards super intelligence.

你同意吗

Do you agree with

我肯定他希望如此，但不，绝对不同意。

I'm sure he does wish that, but it's no, absolutely not.

我不认为AGI应该被当作一个营销术语或用于商业利益。

I don't think AGI should be sort of turned into a marketing term or for commercial gain.

我一直认为它有一个明确的科学定义。

I think there is always been a scientific definition of that.

我的定义是：一个能够展现人类所有认知能力的系统。

My definition of that is a system that can exhibit all the cognitive capabilities humans can.

而且我说的是全部。

And I mean all.

这意味着，你知道的，我们一直推崇的那些最高层次的人类创造力，比如我们敬仰的科学家和艺术家。

So that means, you know, the kind of highest levels of human creativity that we always celebrate the scientists and the artists that we admire.

也就是说，不仅仅是解一个数学方程或猜想，而是提出一个突破性的猜想。

So it means, you know, not just solving a maths equation or a conjecture, but coming up with a breakthrough conjecture.

这要难得多，你知道的，不是在物理或化学中解决某个问题，即使是像AlphaFold那样解决蛋白质折叠问题，而是真正提出一种新的物理理论，就像爱因斯坦提出广义相对论那样。

That's much harder, you know, not solving something in physics or some bit of chemistry, some problem, even like alpha folds, you know, protein folding, but actually coming up with a new theory of physics, something like, you know, like Einstein did with general relativity.

对吧？

Right?

一个系统能提出这样的理论吗？

Can a system come up with that?

因为，当然，我们人类可以做到。

Because, of course, we can do that.

历史上，最聪明的人类凭借他们的大脑结构，已经做到了这一点。

The smartest humans with their brain architecture, our human brain architectures have been able to do that in history.

艺术方面也是如此，不仅仅是模仿已知的东西，而是要像毕加索或莫扎特那样，创造出我们从未见过的全新艺术流派。

And the same on the art side, you know, not just create a pastiche of what's known, but actually be Picasso or Mozart and create a completely new genre of art that we'd never seen before.

对吧？

Right?

以我的观点来看，如今的系统还远未达到那种水平。

And today's systems, in my opinion, are nowhere near that.

不管你解决了多少看似复杂的问题——说实话，我们做这些事固然不错，但我认为这离真正的发明，离像拉马努金那样的人所能做到的成就还差得远。你需要一个系统，能够在所有这些领域都有潜力做到这一点。

Doesn't matter how many, you know, errdous problems you solve, which for some reason, you know, I mean, you know, that's it's good that we're doing those things, but I think it's far, far from what, you know, a true invention or someone like Ramanujan would have been And able to you need to have a system that can potentially do that across all these domains.

除此之外，我还想加入物理智能，因为当然，我们能运动、能控制身体，达到惊人的水平，就像今天在达沃斯现场那些顶尖运动员一样。

And then on top of that, I had add in physical intelligence, because of course, you know, we can play sports and control our bodies and to amazing levels, the elite sports people that are walking around, you know, here today in Davos.

在机器人领域，我们离这个目标还差得很远。

And we're still way off of that on robotics as another example.

所以我认为，一个通用人工智能系统必须能够做到所有这些事情，才能真正实现人工智能领域的原始目标。

So I think an AGI system would have to be able to do all of those things to really fulfill the original sort of goal of the AI field.

我觉得，我们离这个目标还有五到十年的时间。

And I think, you know, we're five to ten years away from that.

我认为这个论点是，如果某样东西能做所有这些事，它就会被认为是超级智能，但你觉得AGI是

I think the argument would be that if something can do all those things, it would be considered super intelligence, but you think AGI is a

好的，不，当然不是，因为这些个体人类确实能做到，比如提出爱因斯坦、费曼那样的新理论，那些伟大的科学家，我的科学偶像们，他们就能够做到。

good No, course not because those individual humans could, we can come up with new theories that Einstein did, Feynman did, the greats that more my scientific heroes, they were able to do that.

这很罕见，但凭借人类的大脑架构，这是可能的。

It's rare, but it's possible with the human brain architecture.

所以超级智能是另一个值得讨论的概念，但它指的是能够真正超越人类智能的事物。

So super intelligence is another concept that's worth talking about, but that would be things that can really go beyond what human intelligence can do.

我们还不能在14个维度上思考，或者把气象卫星直接接入大脑——至少目前还不能。

We can't think in 14 dimensions or, you know, plug in weather satellites into our brains, Not yet anyway.

而那些才是真正超越人类或达到超人类水平的能力。

But and so that that those are truly beyond human or superhuman.

而这，你知道，是另一个需要展开的议题。

And that, you know, that's a whole another debate to have.

但一旦我们实现了AGI。

But once we get to AGI.

我最近在听你讲话时，你的一番话让我非常惊讶。

I was listening to you recently and something you said really surprised me.

你在Google DeepMind播客中被问到，那是个很棒的节目。

You were asked on the Google DeepMind Podcast, which is a great listen.

如果你认为如今有一个系统已经接近AGI，我原本以为可能是Gemini 3。

If you have a system today that is close to AGI, I thought it might be Gemini three.

你却提到了Nano Banana。

You named Nano Banana.

是的。

Yes.

那个图像生成器。

The image generator.

是的。

Yes.

什么？

What?

嗯，你知道，有时候你得给这些名字加点趣味，玩得开心点，但是

Well, you know, sometimes you have to have these fun names and have fun with those and, you know, But

图像生成器怎么就算接近通用人工智能了呢？

how is an image generator close to AGI?

哦，当然，我们以图像生成器为例，但也要谈谈我们的视频生成器VIO，它是目前视频生成领域的顶尖技术。

Oh, well, of course, look, let's take image generators, but also let's talk about our video generator, VIO, which is the state of the art in video generation.

我认为从通用人工智能的角度来看，这甚至更有趣。

I think that's even more interesting in from an AGI perspective.

你可以想象一个视频模型，能生成十秒、二十秒的逼真场景。

You know, you can think of a video model that can generate you ten seconds, twenty seconds of a realistic scene.

这某种程度上是对物理世界的建模。

It's sort of a model of the physical world.

在物理学领域，我们有时称之为直觉物理。

Intuitive physics, we'd sometimes call it in physics land.

它能直观地理解液体和物体在现实世界中的行为方式。

And it sort of intuitively understood how liquids and and and and objects behave in the world.

而展示理解的一种方式，就是能够生成足够准确、足以让人类感到满意的内容。

And that's and, obviously, one way to exhibit understanding is to be able to generate it, at least to the human eye being accurate enough to be satisfying to the human eye.

显然，从物理学角度来看，它还不完全准确，但我们正在取得进展，而且会不断改进。

Obviously, it's not completely accurate from a physics point of view, and we're getting it and we're gonna improve that.

但这确实是朝着构建世界模型迈出的一步——一种能够理解世界、其机制和因果关系的系统。

But it's steps towards having this idea of a world model, a system that can understand the world and the mechanics and the causality of the world.

当然，我认为这对通用智能至关重要，因为这将使这些系统能够在现实世界中进行长期规划，甚至跨越非常长的时间跨度，而我们人类恰恰能够做到这一点，比如我会花四年时间攻读学位，以获得更多资质，从而在十年后找到更好的工作。

And then of course, that would be, I think, essential for a GI because that would allow these systems to plan, long term plan in the real world over perhaps very long time horizons, which of course we as humans can do, you know, I'll spend four years getting a degree so that I have more qualifications so that in ten years I'll have a better job.

你知道，这些我们每个人都能轻松完成的长期规划。

You know, these are very long term plans that we all do quite effortlessly.

但目前，没有这些系统，我们还不知道该如何实现。

And at the moment, without these systems, we still don't know how to do.

我们只能进行短时间尺度的规划。

We can do short term plans over one timescale.

但我认为，你需要这种类型的世界模型。

But I think you need these kind of world models.

我认为在机器人领域，这正是你所期望的——机器人能够在现实世界中进行规划，从当前所处的状态出发，想象出多种可能的轨迹，以完成某项任务。

And I think you imagine robotics, that's exactly what you want for robotics is robots planning in the real world, being able to imagine many trajectories from the current situation they're in, in order to complete some task.

这正是你想要的。

That's exactly what you'd want.

从我们的角度来看，这正是我们从一开始就选择与Gemini合作的原因——它从设计之初就是多模态的，能够处理视频、图像，并最终将所有这些整合到一个统一的模型中。

And then finally, from our point of view, why this is why we worked with Gemini as being multimodal from the beginning, able to deal with, you know, video image and eventually converge that all into one model.

我们的计划是，它也将对通用助手大有裨益。

That's our plan is that it'll be very useful for a universal assistant as well.

让我们来聊聊产品吧。

So let's talk product a little bit.

我和三亿其他人一起观看了纪录片《思考的游戏》。

I watched the documentary, The Thinking Game, along with 300,000,000 other people.

那部纪录片中发生了一些有趣的事情。

There was something kind of interesting that happened there.

在整部纪录片中，你和一些同事不断用手机对准各种事物，向助手Alpha提问发生了什么。

Throughout the documentary, yourself and some colleagues kept pointing your phone at things asking an assistant Alpha, what was going on.

我像往常一样对着电脑大喊：这人该戴眼镜了。

And I was yelling at the computer as I usually do and said, this guy needs glasses.

他需要智能眼镜才能做到这一点。

Like he needs smart glasses to be able to do it.

手机的形态根本不合适。

The phone is the wrong form factor.

你对AI眼镜的愿景是什么？

What is your vision for AI glasses and

什么时候会推出？

when is the rollout happening?

是的，我觉得你说得完全对。

Yeah, I think you're exactly right.

这正是我们的结论。

And that was our conclusion.

当你真正内部使用这些技术时，就会非常清楚——就像你在影片中看到的，我们当时举着手机，想让它告诉我们现实世界的情况。

It's very obvious when you sort of dog food these things and internally that as you saw from the film, we were holding up, you know, you're holding up your phone to get it to tell you about the real world.

这功能确实很神奇，但显然它并不适合很多你想要做的事情，比如做饭、在城市里漫步时询问方向或推荐，甚至帮助视力部分受损的人群，我认为这些场景都有巨大的应用潜力。

And it's amazing it works, but it's not that it's clearly not the right form factor for a lot of things you want to do, you know, cooking or you want roaming around the city and asking for directions or recommendations, or even helping the, you know, partially sighted, there's a huge, I think, use case there to help with those types of situations.

为此，我认为你需要一种可以解放双手的设备。

And for that, I think you need something that's hands free.

对我们这些像我一样戴眼镜的人来说，最明显的选择就是把它集成到眼镜上。

And the obvious thing is for those of us anyway that wear glasses like me, is to put it on glasses.

但可能还有其他类型的设备。

But there may well be other devices too.

我不确定眼镜就是最终形态，但它无疑是一个清晰的下一步形态。

I'm not sure that glasses is the final form factor, but it's definitely it's obviously a clear next form factor.

当然，在谷歌和字母表公司，我们长期以来一直致力于眼镜领域。

And of course, at Google and at Alphabet, we have a long history with glasses.

也许我们过去有些过于超前了。

And maybe we're a bit too early in the past.

但我对这个项目的分析，以及与参与该项目人员的交流，得出了几点看法。

But I think my analysis of it and talking to the people working on that project was a couple of things.

当时的形态设计过于笨重，电池续航等方面也存在问题，但这些问题如今大多已经解决了。

The form factor was a bit too chunky and clunky in the battery life and these kind of things, which are now more or less solved.

但我认为当时缺少的是一个杀手级应用。

But I think the thing it was missing was a killer app.

我认为这个杀手级应用是一个无处不在的数字助手，它能伴随你，帮助你应对日常生活中的各种需求。

And I think the killer app is universal digital assistant that's with you, helping you in your everyday life.

它可以在任何设备上为你服务——你的电脑、浏览器、手机，甚至在你漫步城市时，通过眼镜这样的设备提供支持。

And they're available to you on any surface, on your computer, on your browser, on your phone, but also on, you know, devices like glasses when you're, you know, walking around the city.

我认为它需要做到无缝衔接，能够感知并理解你所处的每一个具体情境。

And I think it needs to be kind of seamless and kind of knows each of those contexts and understands each of those contexts around you.

我认为我们现在已经很接近了，尤其是有了Gemini 3，我感觉我们终于拥有了足以让这一愿景成为现实的AI能力。

And I think we're close now, especially with Gemini three, I feel we finally got AI that is maybe powerful enough to make that a reality.

这可以说是我们正在推进的最令人兴奋的项目之一，而我个人也在致力于让智能眼镜真正发挥作用。

And we're, you know, it's one of the most exciting projects we're working on, I would say, and it's one of the things I'm personally working on is making smart glasses really work.

我们已经与Warby Parker、Gentle Monster和三星等公司建立了出色的合作关系，共同打造下一代眼镜产品。

And we hope to, we've done some great partnerships with Warby Parker and Gentle Monster and Samsung to build these next generation glasses.

你应该会在夏天前后开始看到这些进展。

And you should start seeing that, you know, maybe by the summer.

是的，Warby Parker 提交的文件显示，这些眼镜很快就会推出。

Yeah, Warby Parker did have a filing that said that these glasses are coming out pretty soon this Yeah,

原型设计取决于原型阶段的进展速度，但我认为这很快就会实现，并且将成为一种定义新类别的技术。

and the prototype design, depends how, you know, when prototype phase, it depends how quickly that advances, but I think it's gonna happen very soon and I think it will be, you know, a new category defining technology.

鉴于你个人的参与，可以说这是对你们来说非常重要的一个项目吗？

Given your personal involvement, is it safe to say that this is a pretty important initiative for

是的，这确实很重要，但我想说，它的重要性不止于此。

Yeah, it's one, well, yes, but it's, I mean, I like to, it's not just as important.

显然，我喜欢把时间花在重要的事情上，但我更倾向于推动最前沿的技术，而这往往是最困难的——设定阶段性目标，给团队信心，同时也要判断时机是否成熟。

Obviously, I like spending my own time on important things, but I like to be at the push the most cutting edge thing and that's often the hardest thing and picking interim goals and giving confidence to the team and also just sort of understanding if the timing's right.

多年来，我已经做了这么多年，现在我对这些事已经相当在行了。

And over the years I've been doing this, the many, you know, the decades now, you know, I've got quite good at doing that.

所以，我努力站在最前沿，因为我觉得在那里我能产生最大的影响。

So, I try to be at the most cutting edge parts of I feel I can make the most difference there.

所以，像眼镜、机器人和世界模型这样的事情，我都花时间在上面。

So, things like glasses, robotics I'm spending time on, and world models.

对。

Right.

好的。

Okay.

所以，现在是推出眼镜的好时机。

So, timing's right for glasses.

我们来谈谈广告吧。

Let's talk about ads.

当然。

Sure.

现在是时候让我来铺垫一下了。

Is the timing right Let for me set it up.

是的。

Yes.

好的。

Okay.

最近有一些消息说，Gemini 可能会包含广告。

There's been some news that Gemini might include ads.

最近也有一些消息说，你的某些竞争对手可能会加入广告。

There's been some news that some of your competitors might include ads.

我在社交媒体上看到最有趣的一件事是，有人说，这些人离通用人工智能还差得远。

The funniest thing I saw about that on social media was someone who said, these people are nowhere close to AGI.

如果商业模式是广告，那这将不会是一项颠覆世界的技术。

It's not going to be this world disrupting technology if the business model is advertising.

是的，这确实挺有意思。

Yeah, well, it's interesting.

我认为这些迹象表明，行动胜于言辞，这让我们回到之前和萨姆等人讨论的话题——他们声称通用人工智能近在咫尺。

I think those are tells on, I think actions speak louder than words going back to the original conversation we're having with, Sam and others claiming AGI is around the corner.

那你们为什么还要搞广告呢？

Why would you bother with ads then?

所以，我认为这是一个合理的问题。

So that is, I think a reasonable question to ask.

但从我们的角度来看，目前我们没有计划做广告。

But I think, look, from our point of view, we have no plans at the moment to do ads.

如果你说的是Gemini应用，那么显然我们会密切关注Chachi PT声称他们将要采取的举措。

If you're talking about the Gemini app, right, specifically, I think are going obviously, we're going to watch very carefully what the outcome of what Chachi PT is saying they're going to do.

我认为必须谨慎处理，因为我看到的二元对立是：如果你想要一个为你服务的助手，最重要的事情是什么？

I think it has to be handled very carefully because the dichotomy I see is that if you want an assistant that works for you, what is the most important thing?

信任。

Trust.

好的，所以是信任、安全和隐私，因为你希望与这个助手分享你生活的方方面面，那么你希望确信它是在为你着想，并以你的最佳利益为出发点。

Okay, so trust and security and privacy, because you want to share potentially your life with that assistant, then you want to be confident that it's working on your behalf and with your best interests.

因此，你必须小心。

And so, you you've got to be careful.

我认为有一些方式可以做到，但你必须谨慎，确保广告模式不会渗透进来，让用户混淆这个助手推荐的内容究竟是什么。

I think there are ways one could do it, but you'll be careful that the advertising model doesn't bleed into that and confuse the user as to what is this assistant recommending you?

我认为，这将是该领域的一个有趣挑战。

And I think, that's gonna be interesting challenge in that space.

这就是不该做的事。

And that's what not to do.

孙达尔在最近的财报电话会议上提到，谷歌内部有一些关于如何正确处理这个问题的想法。

And Sundar in a recent earnings call said, there are some ideas within Google of the right way to approach this.

当然。

Sure.

你如何对待广告？

How do you approach advertising?

嗯，

Well,

你知道，我们还在头脑风暴这个问题，但我认为，如果你考虑眼镜类设备，还有其他一些非常有趣的盈利模式。

you know, we're still brainstorming that, but I think it's, there are also, you know, very interesting ways when if you think about glasses devices, there are other revenue models out there.

所以，这将会很有趣，让我们拭目以待。

So, you know, it's going be interesting to see.

我认为我们还没有在这方面得出明确的结论，但这是一个需要非常谨慎思考的领域。

I don't think we've made any strong conclusions on that, but it's an area that needs very careful thought.

为了让你给出一个明确的答案，我觉得你已经说了，但我还是想再确认一次。

Just to get a definitive answer from you, I think you've given it, but I'm just going do it one more time.

在我们见面之前，我读到过，谷歌在过去一年里曾告诉广告商，计划在2026年将广告引入其AI聊天机器人Gemini。

I read before we met, Google has told advertisers in recent days from last year that it plans to bring ads to its AI chatbot Gemini in 2026.

没有。

Nope.

我们目前没有任何计划。

We have no current plans.

我只能说到这里。

That's all I can say.

好吧，这已经非常明确了。

All right, pretty definitive.

好吧，我们继续聊聊你们的一些竞争对手，Anthropic。

All right, let's just keep going through some of your competitors, Anthropic.

Claude Code 和 Claude Cowork 引发了巨大的关注。

Claude Code and Claude Cowork have caused a tremendous amount of buzz.

看到有些人所做到的成就，真是令人惊叹。

It is amazing to see what some people have done.

我看到一篇前亚马逊高管的帖子，他说他在一个周末，或者实际上是一天半内构建了一个定制的客户关系管理系统。

I saw a post from an ex Amazon executive who said that he built a custom CRM in a weekend or actually a day and a half.

我们就说是周末吧。

Let's call it a weekend.

你怎么看？

What do you think about it?

是的。

Yeah.

你们有计划推出相应的应对方案吗？

And do plan to have an answer to

吗？

it?

这非常令人兴奋。

It's very exciting.

我认为，要给Anthropic点赞，他们用ClawCode打造了一个非常优秀的模型。

And I think, you know, kudos to Anthropic, I think they built a very good model there with ClawCode.

我们对Gemini三号当前的编程能力非常满意。

We're very happy with the current coding capabilities of Gemini three.

它在前端工作等某些方面表现非常出色。

It's very good at certain things like front end work.

我一直在圣诞节期间用它来原型化游戏。

I've been using it over the Christmas to prototype games.

这太棒了。

So it's amazing.

它让我重新回到了编程中。

It's getting me back into programming.

我喜欢当前正在发生的整个编码浪潮。

I love the whole vibe coding wave that's happening.

我认为这将为设计师、创意人员和艺术家打开整个生产力领域，他们以前可能需要依赖编程团队才能完成工作。

I think it will open up the whole productivity space to designers, creatives, artists that maybe would have had to work with teams about access to teams of programmers.

现在他们很可能可以独自完成更多事情。

Now they can probably do, you know, a lot more just on their own.

我认为，一旦这种能力更广泛地普及开来，将创造出大量新的创意机会，这将非常了不起。

I think that's going to be amazing once that's sort of out in the world in a more general way to, you know, create lots of new creative opportunities.

我们在代码方面的工作进展顺利，对此我们非常满意。

We're working on, we're very happy with our work on code.

我们还有很多工作要做，你知道的。

We've a lot, you know, we've got more to do there.

我们刚刚发布了反重力——我们自己的集成开发环境，它非常受欢迎。

We've just released anti gravity, our own IDE, which is very, very popular.

我们目前无法满足所看到的所有需求。

We can't actually serve all the demand that we're seeing there.

我们正在全力提升Gemini在编码和工具使用方面的性能。

And we're pushing very hard on coding and tool use performance of Gemini.

但有一点我认为Anthropic完全专注于此，他们不做图像模型、多模态模型或世界模型，他们只专注于编码和语言模型。

But it's one thing that I think Anthropic have fully focused on, you know, they don't make image models, multimodal models, world models, they just do, you know, coding and language models.

而且他们在这一点上做得非常非常出色。

And they're very, very good at that.

另一方面，我们很高兴能与他们在这方面合作。

And, you know, we're pleased to be partnering on that on the one hand.

同时，这也为我们提供了动力，推动我们改进自己的模型。

And also it gives us something to push for to improve with our own models.

让我们 broadly 谈谈人工智能行业的商业情况。

Let's just talk broadly about the AI industry business.

我有一个关于这一切可能如何崩溃的理论，想跟你聊聊。

I have a theory for how this could all fall apart and I want to run it by you.

这是一个三步过程。

It's three step, a three step process.

第一步是，大型语言模型的训练投入回报有限。

The first is that large language model training runs produce limited returns.

第二点是，像Gemini Flash这样的闪存模型，其AI计算成本低至与搜索相当。

The second is that there are flash models like Gemini Flash that run AI computing as cheap as search.

第三步是，鉴于上述两个因素，此前所做出的庞大基础设施投入变得有些无用。

And then step three is that the massive infrastructure commitments that have been made become somewhat useless given those two factors.

随后就会发生连锁崩溃。

And there is a cascading collapse that happens.

这是一个合理的担忧吗？

Is that a legitimate worry?

我认为这是一个可能成立的情景。

I think it's a plausible possible scenario.

但在我看来，这种情况不太可能发生。

I don't it's likely one in my opinion.

我的意思是，在我看来，AI已经充分证明了自己的价值，我认为我们在科学、AlphaFold和药物发现等领域的工作表明，AI已经站稳脚跟。

I mean, in my mind, there's no doubt AI has gone already proved out enough, I would say and our work, I think in things like science and alpha fold and drug discovery that it's here to stay.

这并不是说明天我们会发现，哦，原来AI根本不管用。

It's not like tomorrow, oh, like, oh, we found that AI doesn't work.

我们已经远远超越了那个阶段。

We've gone way, we've blasted way past that.

所以我认为这无疑将成为人类历史上最具变革性的技术。

So I think that's it's clearly going to be the most transformative technology in human history.

关于时间表可能还存在一些疑问。

There's maybe a question mark about timelines.

是两年还是五年？

Is it two years or five years?

我的意思是，对于如此具有变革性的技术来说，它已经非常近了。

I mean, way, it's very soon for something this transformative.

我认为我们仍处于探索如何利用和部署它的初期阶段，因为技术进步得太快了。

And I think we're still in the nascent era of actually figuring out how to make use of it and deploy it because the technology is improving so fast.

我认为实际上，即使是我们这些构建这些模型的人，也未必完全了解当今模型所具备的巨大能力。

I think there's a huge capability overhang actually of what even today's models can do that maybe even us as building those things don't fully know.

所以我认为我们看到了大量前所未有的产品机会。

So I think there's just a vast amount of product opportunities that we see.

我认为，作为谷歌，我们才刚刚开始触及表面，将这些技术原生地整合到我们现有的优秀产品中，更不用说开发全新的产品了，比如AI邮箱，我们才刚刚开始试点。

And I think we're as Google only just started to scratch the surface now of actually natively sort of plugging these things in to our amazing existing products, let alone building the new ones, you know, AI inbox, we've just started trialing.

谁愿意做邮件管理呢？

I mean, who wants to do email admin?

难道我们不都希望这些繁琐的事情能彻底消失吗？

I mean, wouldn't we all love that to just go away?

这对我来说是工作中最大的痛点，金日。

That's my number one pain point for my work, King Day.

还有很多类似的例子正等着被解决。

And there's so many examples like that just waiting to be addressed.

想想浏览器中的智能代理，帮助处理YouTube，显然我们现在已经在用它来增强搜索功能。

Think, you know, agents in browsers, helping out with YouTube, obviously we're now powering search with it.

我认为这背后有着巨大的机遇。

So I think there's enormous opportunities.

如果你在谈论AI泡沫，这就是你要问的问题。

And if you're talking about the AI bubble, if that's the question.

我本来想不问AI的问题

I was trying to not ask the AI

泡沫问题。

bubble question.

我觉得没问题。

I think it's fine.

我的意思是，我觉得这样问挺好的，我很乐意回答，因为在我看来，判断是否处于泡沫中并不是非黑即白的。

Mean, it seems like that's the I'm very happy to answer it because I think, look, my view is it's not binary when are we in a bubble, not in a bubble.

我认为AI行业的一部分可能确实处于泡沫中，而其他部分还有待观察。

I think parts of the AI industry probably are and other parts I think it remains to be seen.

所以我觉得，当你看到一些初创公司，种子轮融资高达数十亿美元，却根本没有产品或研究，只是几个人凑在一起，

So I think some of the things are, when you see seed rounds of tens of billions of dollars of companies that basically have no product or research, it's just some people coming together.

这在我看来在正常市场中有点不可持续，有点过热。

That seems a bit unsustainable to me in a normal market, a bit frothy.

但另一方面，像我们这样的企业，拥有庞大的现有业务和产品，AI如何提升这些产品的使用效率或生产力，这一点非常明确。

On the other hand, you know, we're businesses like us, we have massive underlying businesses that and products that it's very obvious how AI would increase the efficiency or the productivity of using those products.

然后，这些新的AI原生产品，比如聊天机器人、智能眼镜等，它们的商业化会有多受欢迎，还有待观察。

And then it remains to be seen how popular the monetization of these new AI native products like chatbots, glasses, all of these things.

我们得拭目以待。

We'll have to see.

我认为会存在巨大的市场，但这些市场尚未得到验证。

I think there will be enormous markets, but they're yet to be proven out.

但从我的角度来看，负责运营谷歌DeepMind是我的职责，我要确保无论AI泡沫是否出现、是破裂还是持续存在，我们都能获胜。

But from my perspective, you know, running Google DeepMind is my job is to make sure that whatever happens with an AI bubble, it bursts or if there isn't one and it continues, we win either way.

我认为，无论哪种情况，Alphabet都处于极其有利的位置：要么在现有业务上加倍投入，要么在牛市中站在前沿和尖端。

And I think we're incredibly well positioned as Alphabet in either case, as, you know, doubling down on existing businesses in the one case or being at the forefront and the frontier in the bull case.

回到思考游戏，说到这将如何影响经济，我开始为你们技术的对手感到难过。

Going back to thinking game, speaking of the way that this will impact the economy, I started to feel bad for the opponents of your technology.

李世石。

Lee Sedol.

好的。

Okay.

士气低落。

Demoralized.

当然。

Sure.

这位玩《星际争霸》的选手Mana击败了你的机器人，但他意识到人类对机器的较量基本上已经结束了。

This guy Mana who played Starcraft beat your bot, but realized that it's basically over for humans versus machines.

现在，随着这些技术逐渐渗透到知识型工作中，我们所有人都在某种程度上面临这一挑战。

Now we're all up against this in some way as this stuff makes its way into knowledge work.

我以为你指的是我们的AI竞争对手。

I thought you were meaning our AI competitors.

对于他们，我无所谓。

Them I'm okay with.

我对此并不感到难过。

I don't feel sad about that.

不，抱歉，我太执着了。

No, relentless sorry.

AI的进展。

Progress of AI.

你是说那些游戏玩家吗？

You mean the gamers.

针对游戏玩家。

For gamers.

是的，当然。

Yeah, sure.

让我为游戏玩家感到难过。

Made me feel bad for gamers.

嗯，你知道的，但我还想问问这个。

Well, you know But I wanna ask about this.

在知识工作领域，我们也将面临同样的情况：这些曾经出色击败世界顶尖《星际争霸》和围棋选手的模型，现在开始接手我们的工作。

We're gonna have the same situation with knowledge work that these models that performed admirably against the world's best Starcraft and go players are now starting to do our work.

我们会最终陷入

And are we gonna end up in the

同样的处境吗？

same position?

好吧，让我们以游戏为例来说明。

Well, let let's give you brought up games as an example.

让我们看看游戏领域发生了什么。

Let's look at what's happened in games.

国际象棋方面，自从我青少年时期以来，电脑的水平就已经超过了90年代的加里·卡斯帕罗夫，对吧？

So chess, we've had chess computers that are better since I was a teenager than Gary Kasparov in the 90s, right?

它们并不是通用人工智能系统，但它们是深蓝。

They weren't general AI systems, but they were Deep Blue.

国际象棋如今比以往任何时候都更受欢迎。

Chess is more popular than ever.

人们感兴趣的是看到电脑之间对弈。

One's interested in seeing computers playing computers.

我们更感兴趣的是马格努斯·卡尔森与世界上其他顶尖棋手的对局。

We're interested in Magnus Carlsen playing, top, the other top chess players in the world.