谷歌的人工智能历史与未来，由桑达尔·皮查伊讲述

是的。

Yeah.

这对我来说没有意义。

So, that doesn't make sense to me.

我认为其中一部分原因是，如果你是一家年轻公司，或者你更像一个纯粹的研究实验室，你可能总部设在旧金山。

I think some of it is, you know, if you're a younger company, you know, or you're more a pure research lab, you know, you're maybe headquartered in San Francisco.

是的。

Yeah.

是的。

Yeah.

是的。

Yeah.

有一些

There are a

许多细微的特征可能会影响，但我认为在根本层面上，对趋势的看法并没有差异。

lot of small attributes which can probably make But a I don't think at a foundational level, there is a difference in outlook on what the curve is.

是的。

Yeah.

对吧？

Right?

或者我们如何内化这项技术。

Or how we internalize the technology.

听我说，我认为即使在公司内部，也有一群人站在技术前沿，部署智能代理，观察这些工具能做什么，看代理如何习得技能、完成任务，同时回看三个月前它们能做什么，现在又进步了多少。

Look, I think even within the company, there's a set of us living on the bleeding edge, firing agents, seeing what these things can do, see the agents pick up skills, do stuff, and also look back three months ago what they could do now.

我们内部正亲历这种指数级增长。

And we are living that exponential internally.

嗯。

Mhmm.

对吧？

Right?

我觉得你们俩说的都对，我同意。

I think you're both right where I agree.

你可以大致指出谷歌的发展历程。

You can kind of point out the history of Google.

我认为埃拉德想表达的是一种感觉，我最近看到一条推文，他说要解释硅谷当前发生的事情，你必须意识到每个科技高管现在都患有严重的AI妄想症，花大量时间写代码、与AI交流等等。

I think what Elad's getting at is like a a feeling where I saw a tweet go by this, and he was saying, what you have to realize to explain what's currently going on in the Valley is that every tech executive has severe AI psychosis right now and is spending, you know, a huge amount of time writing code and talk to AI and things like that.

我觉得这个观点很有趣，而且并非毫无道理。

I thought that was a funny take and not without any truth to it.

我想知道，你在近期AI发展的过程中，有没有经历过什么AGI的瞬间？或者说，你现在有多严重的AI妄想症？

And I'm curious, what were your feeling the AGI moments along the way of the recent or, you know, to what extent do you have AI psychosis these days?

我第一次感受到AGI的时刻是2012年，当时杰夫·迪恩展示了Google Brain的最早版本。

My first feeling the AGI moment was 2012 when Jeff Dean demoed the earliest version of Google Brain.

就是神经网络识别出猫咪的那一刻，对吧？

This is when the neural networks recognized a cat, right?

所以那是2012年。

So that was 2012.

我和拉里一起去参加了DARPA挑战赛。

I went with Larry to the DARPA challenge.

可能是2014年，我觉得。

Might have been 2014, I think.

我需要准确回忆我们去那里的时间，当时看到那些车在行驶。

I need to be exact about when when we went there, seeing the cars drive there.

黛米正在演示这些模型最早期的版本，展现出我们所谓的想象力。

Demi is demoing the earliest versions of the models having what we would call as imagination.

所以类似这样的时刻有很多。

So there have been many moments like that.

因此，当技术已经真正融入生活，并让你产生直观感受时，我认为最接近的体验是：当你在编程时，给它一个复杂的任务，然后你根本不用打开IDE，而是身处某个智能代理管理系统中，看着它自行完成任务，感受它的强大之处。

So it was obvious the technology is In terms of living now and kinda having a visceral feel for it, I think the closest I would say is if you're coding and you give it a complex task and you never open the IDE, you're in some agent manager world and you see it kind of do it, you know, and how powerful it is.

所以，如果你愿意，可以称之为现场版的AGI。

So, you know, if you can, you know, call it field AGI.

确实存在这样的时刻。

So there are moments like that.

是的。

Yeah.

是的。

Yes.

我最近做了一个小的爱好项目，过了一段时间，我就想，咦，它用的是什么语言？

I've done a little hobby project recently, and after a while, I was like, oh, I wonder if that language it's using.

但那是个细节，我得在一切运行起来之后才问它。

But that was, like, a detail that I needed to ask it about after everything was up and running.

是的。

Yeah.

是的。

Yeah.

感觉像魔法一样。

Word feels like magic.

是的。

Yeah.

是的。

Yeah.

所以，确实，这样的时刻肯定有。

So, yeah, moments like that for sure.

对。

Yeah.

但曲线的斜率才真正让你惊讶。

And but the slope of the curve is what surprises you.

对。

Yeah.

对吧？

Right?

你在这么多范式上都在不断改进。

And you're improving it on so many paradigms.

很明显，未来还会取得进展。

It feels clear that there's gonna be progress ahead.

对吧？

Right?

所以

So

当你谈到那种直观的感受时，我觉得有一件事对科技公司至关重要，每位CEO对此都有不同的理解，那就是如何保持与产品体验和日常用户的联系，因为科技产品太过抽象，你不可能仅仅通过Teams的报告、幻灯片和电子表格来管理。

When you talk about the visceral feel, feel I like one thing that's important to tech companies, and every CEO thinks about this differently, is how you stay connected to the product experience and everyday users because tech products are so abstract that it's easy to you know, you cannot just manage through reports from Teams and slide decks and spreadsheets.

所以，托尼·徐提到他仍然在做DoorDash骑手，以保持对这种体验的紧密联系。

And so, you know, Tony Hsu is talking about how he still works as a DoorDasher, you know, to stay very connected to that experience.

在我们每周的全员会议上，有一个固定的环节叫‘逛商店’，我们会一起点击浏览仪表盘，质疑为什么这个模型会在这里？

We do at our, like, little weekly all hands we have a recurring segment of just walk the store, where we, like, click around in the dashboard together and we're tripping over, like, why is that model there?

这有点让人困惑之类的，只是为了让我们共同使用产品。

And that's a bit confusing or whatever, just so we're, like, collectively using the product.

我想知道这对你们来说是怎么运作的，你们在谷歌是如何确保自己始终与产品的使用体验保持联系的？

I'm curious how it works for you and how at Google you ensure this you're staying connected to the experience of using the products.

除了你们每天都在使用Gmail和其他所有产品之外。

Other than you use, like, Gmail and everything every day.

不是的

No.

是的

Yeah.

对

Yeah.

比如，你知道的，内部版本的狗粮测试。

Like, you know, dogfooting, like, literally internal versions.

嗯哼

Mhmm.

我会专门预留时间，集中精力深入使用产品，也就是专注使用的时间。

I do block time, like, to kinda use it intensely, so, like, kinda focus time to do it.

这样确实有帮助。

And so that helps.

就在两周前，我在健身房拉伸时，手里还拿着手机，开着Gemini Live。

Like, even just two weeks ago, I was stretching in the gym, and I had the phone with Gemini Live.

所以我就想，花三十分钟就专注和它聊一个话题。

And so I'm like, wanna talk to it for, like, at the end of thirty minutes on, like, one topic.

嗯。

Mhmm.

所以你会做这些事，其中一些有效，一些让人沮丧。

So you you do those things, and it's it's some of it works, some of it is frustrating.

我们学到了很多。

We kinda learn a lot.

对吧？

Right?

比如，我会强迫自己以高级用户的方式使用它，保持这种联系。

Like, so I force myself to use it in those power user mode ways and and stay in touch that way.

X 有帮助，因为有时你能获得直接的反馈。

X helps because sometimes you get the raw feedback.

谢谢你解决了谷歌日历的问题。

Thank you for fixing the Google Calendar thing.

那真是太好了。

That was so good.

哦，哇。

Oh, wow.

我们还得修复几个，

The few more we

但

have to fix, but

不。

No.

太棒了。

That's awesome.

谢谢你提出来。

Thanks for flagging it.

是的。

Yeah.

所以，是的，X 有帮助，因为你能直接看到原始评论，我试着直接跟进。

So, yeah, X helps because you you kinda get the raw raw comments, and I tried to follow it directly.

嗯。

Mhmm.

但我告诉你，真正有帮助的是什么。

But I'll tell you what what has helped.

在内部，我会去查看我们早期的部分。

Internally, like, I would go fire to our earlier part.

比如，我会在反重力系统里查询，就是我们内部版本的反重力系统。

Like, I would query in anti gravity, just our internal version of anti gravity.

嘿，我们上线了这个功能。

Hey, we launched this thing.

嗯。

Mhmm.

大家对这个有什么看法？

Like, what did people think about this?

告诉我，人们讨论得最差的五件事和最好的五件事是什么，我就把这些打下来了。

Tell me that, like, the worst five things people are talking about, the best five things people are talking about, and I typed that.

嗯。

Mhmm.

所以现在又回到了这一点。

So now that brings it back.

那么，我的生活变轻松了吗？

So has my life gotten easier?

是的。

Yes.

过去，我会花更多时间去体会它。

So in the past, I would have spent a lot more time trying to get a sense for it.

现在，一个AI代理正在帮助我完成这一过程。

Now an AI agent is helping me in that journey.

所以，你可能会想，究竟该花多少时间亲自去感受，而不是直接利用这些工具？

So you can get you know, well, how much should I be spending firsthand to get that feel versus actually leveraging these tools?

所以，就连我本人也正在经历这样一个过程，对吧？

So even I'm going through a journey there, right?

所以我正在努力适应这个未来。

So I'm trying to adapt to this future.

我想你提到过，A，这并不是零和游戏。

I guess there's you mentioned, A, that it's not zero sum.

B，人们正在看到各种生产力的提升。

B, there's all these productivity gains people are seeing.

如果你回顾以往的技术周期，互联网、移动技术或SaaS花了一段时间才反映在实际的GDP数据中。

And if you look at a lot of prior technology cycles, it took a while for the internet or for mobile or for SaaS to show up in actual GDP numbers.

在AI的背景下，我们正从数据中心建设的角度看到这一点。

In the context of AI, we're seeing it from a data center build out perspective.

这推动了GDP的部分增长。

That's driving part of GDP growth.

你如何看待三到五年后的前景？

How do you think ahead in terms of three, four, five years?

你认为美国经济因为AI而变得更庞大了吗？

Do you think The US economy is bigger because of AI?

如果是的话，大多少呢？

And if so, how much bigger?

听我说，为了让这些回报有意义，你得想想，之前花了多久？

Look, for these returns to make sense somewhere, it has to you know, how long was it before?

我记得好像是红杉资本有人写过，说嗯。

I think it was maybe from Sequoia someone wrote and saying Mhmm.

人们正在投入这么多资金。

People are investing this much.

是的。

Yeah.

他们正在将资本支出与

They're comparing the CapEx to

这个嗯。

the Yeah.

这可能已经是第二

This might have been two

和一个

and a

半年前，有人在一次谈话中提到，这说不通，因为你需要在这个水平上获得回报。

half years ago, it was a talk and, like, saying, it doesn't make sense because you would need to return at that level.

你接下来可能还要做十件事。

You're probably 10 next things.

是的。

Yeah.

对。

Yeah.

从那时起，我需要重新看一下数据。

Since that moment, I need to go look at the numbers again.

对吧？

Right?

所以，在某个时候，它必须得到平衡。

So at some point, you know, it has to reconcile.

为了说得清楚一点

To be very clear

嗯。

Mhmm.

你知道，我们的供应是受限的。

You know, we are we are supply constrained.

嗯。

Mhmm.

我们在所有我们涉及的领域都看到了需求。

We are seeing the demand across all the surface areas we all

我其实

I actually

我丝毫不怀疑这是一个巨大的市场和前景。

don't have any doubt that this is a massive market and outcome.

所以我的问题是，我认为很多人对此存在误解。

So my my question and I I think there's a lot of things that people misunderstand.

例如，人们经常谈论软件工程预算，以及其中有多少比例用于代币而非薪资。

So for example, people often talk about software engineering budgets and then what proportion of that is token versus salary.

在某种程度上，我认为由于优秀软件工程师的供给严重不足，一旦增加供给，这个市场可能会扩大十倍。

And to some extent, I think that market has been so demand constrained for great software engineers that suddenly adding supply can 10x that market.

对吧？

Right?

换句话说，我认为软件工程和编程市场的规模远比任何人想象的都要大，用代币预算与工程师数量来衡量是错误的指标。

In other words, I think the market for software engineering and coding is dramatically bigger than anybody thinks, and it's the wrong metric to say, you know, token budget versus engineers.

所以我实际上认为它应该推动很多方面的发展。

So I actually think it should grow a lot of things.

是的。

Yeah.

我只是好奇，你认为这个领域实际可能带来多大的增长？

I was just sort of curious of your view of, like, how much growth do we think is likely actually to come of this?

我根本没有任何怀疑，关于资本支出与成果之间的关系，对吧？

I actually wasn't doubting at all sort of CapEx versus outcomes or you know?

我明白了。

I see.

是的。

Yeah.

听好了。

Look.

我的意思是，回过头来看互联网和GDP增长，你知道，GDP并没有完全反映出我们所有人对互联网的真实感受。

I mean, going back to the Internet and looking at GDP growth, you know, it doesn't quite capture what we all feel with the Internet.

对吧？

Right?

所以，也许如果没有互联网，我们的GDP增长会是负的。

And so maybe we would have had negative GDP growth without the Internet.

消费者剩余。

Consumer surplus.

是的。

Yeah.

所以，你知道，展望未来是很难的。

So, you know, it's it's tough to look ahead.

我认为社会各个层面都存在一些自然的抑制机制。

I do think there are natural dampening mechanisms in society at various levels.

很明显的是，计算能力的扩张曲线与我们提升模型的速度并不一致。

And the obvious ones being, you know, the compute build out is a different curve than the rate at which we can improve the models.

嗯，是的。

Uh-huh.

对吧？

Right?

所以你已经在应对一个更受限的曲线了。

So you're already dealing with a more constrained curve there.

那么，你如何将其普及到社会中呢？

Then how do you diffuse it into society?

对吧？

Right?

我们正在与Waymo合作做这件事。

We are doing this with Waymo.

嗯。

Mhmm.

对吧？

Right?

你可以让Waymo比人类司机更安全，但你必须谨慎控制我们推广的速度等等。

And, you know, you can make Waymo safer than human drivers, but, you know, but you have to be careful at, like, the pace at which we are rolling out, etcetera.

所以有时候，你如何负责任地将它融入社会呢？

So sometimes, you know, how do you diffuse it through society responsibly?

这些层面都存在制约因素。

There are constraints in all these layers.

对吧？

Right?

但我认为，美国经济比十年前大得多。

So but I think The US economy is so much larger than it was ten years ago.

所以，即使只提高百分之零点五，那也是一个巨大的贡献。

So to grow that even at a half a percentage point higher, then, you know, that that's a massive contribution.

所以我预计会以这种方式发展。

So I expect it to play out that way.

听桑达尔讲话，让人深刻体会到什么是真正的互联网规模运营。

Listening to Sundar is a powerful reminder of what it means to operate at true Internet scale.

然而，在商业领域，大多数企业不得不在信息孤岛中做出关键决策。

When it comes to commerce, though, most businesses are forced to make critical decisions in a vacuum.

他们只从自己的视角看待互联网经济，仅使用企业内部的数据。

They view the Internet economy through only their own lens using only the data that exists within their four walls.

这就是Stripe存在的意义。

This is where Stripe comes in.

因为我们每年处理1.9万亿美元的全球支付，所以我们拥有全景视野。

Because we process $1,900,000,000,000 a year in global payments, we have a panoramic view.

当你基于Stripe构建时，你接入的就是这套网络智能，从验证用户身份、预防欺诈，到优化企业间的连接方式，以及了解本地最有效的支付方式。

When you build on Stripe, you're plugging in to that network intelligence from verifying a user's identity to stopping fraud before it starts to streamlining how businesses connect and knowing what payment methods work best locally.

我们建立了一个系统，将全球GDP的1.6%转化为保护和增长您收入的动力。

We've built a system that puts 1.6% of global GDP to work to protect and grow your revenue.

所以，如果您想利用Stripe网络智能的力量，就来看看您能在这里构建什么吧。

So if you want to use the power of Stripe's network intelligence, come see what you can build here.

您提到了供应限制，我认为这是2026年一个非常有趣的决定性特征，您之前提到资本支出为1500亿美元，是1980年1月吗？

You referenced the supply constraints, and I think that's a really interesting defining aspect of 2026, basically, where you said 150,000,000,000 in CapEx, 01/1980?

我们说过，它将在1750亿到1850亿美元之间。

We have said it's it'll be between 175 and 185.

好的。

Okay.

所以资本支出大约是1800亿美元。

So 180 ish billion of CapEx.

让我觉得有趣的是，即使谷歌想花4000亿美元做资本支出，它也做不到，因为内存不够，电力也不足，还有这些其他组件。

And what's interesting to me is that Google could not spend $400,000,000,000 in CapEx if it wanted to because the memory isn't there and, you know, the power isn't there and all these components.

所以您能逐一说明一下吗

So can you just tick through

我们可以找到我们需要的电工数量

We can find a number of electricians we would need

没错。

to Exactly.

我很想听听你对各种瓶颈的总体看法。

So I'd love to hear just your overview of the various bottlenecks.

说到底，我们得回溯到实际的晶圆产能之类的问题。

Look, at some level, have to work back to actual wafer capacity or something like that.

对吧？

Right?

所以存在一些更深层的现实基础。

So there are, like, deeper ground truths.

对吧？

Right?

我觉得是晶圆开工量。

Think so, wafer starts.

这是一种根本性的限制。

It's kind of a fundamental constraint.

我认为电力和能源问题更容易解决。

I think power and energy are more solvable.

许可和实际应对监管环境可能是一个限制。

Permitting and actually working through a regulatory environment might might be a constraint.

对吧？

Right?

所以你做事的速度。

So the the the pace at which you can do things.

尽管在促进增长的地区，比如德克萨斯、内华达或蒙大拿，土地很多，但也许还是不够？

Even though there's lots of land in pro growth, you know, Texas or Nevada or Montana, just maybe not enough?

我认为我们正在取得巨大进展。

I think we're we're making tremendous progress.

我认为对于美国来说，这尤其重要。

I think for The US, I think it's a particularly important thing.

你知道吗，你会对中国的建设速度感到惊叹，他们做事真的太快了。

You know, it's you're in awe of, like, the pace in, you know, in China, how fast they can build things.

所以我真的认为我们需要学会更快地建设东西。

So I really think we need to learn to build things much faster.

你几乎必须转变思维，去思考在现实世界中，如何把事情的速度提升十倍，对吧。

Like, you you almost have to shift your mentality to think about what would it take to do things 10 x faster, right, in the physical world.

是的，先生。

Yes, sir.

把建设速度提升十倍。

Construct 10 x faster.

但我担心这会成为一个限制因素。

But I would worry about that as a constraint.

嗯。

Mhmm.

对吧？

Right?

我认为，可能会出现越来越多的阻力。

I think, you know, there could be growing resistance.

所以，这并不像少数人决定要加快建造速度那么简单。

So, you know, it's not as simple as, like, a few people deciding you wanna build faster.

数据中心的暂停建设之类的事情。

The data center moratorium and stuff.

是的。

Yeah.

对。

Yeah.

所以我会说，晶圆启动、获得许可和推进项目的能力。

So I would say wafer starts, the ability to permit and do things.

而且我认为政府正在做很多有益的工作。

And I do think there's a lot of good work being done from the government on.

我认为人们已经意识到，我们需要更好地完成这些事情。

I think people realize you need to do these things better.

接下来是供应链中的关键组件。

Then comes critical components in the supply chain.

内存就是一个很好的例子。

Memory is a good one.

我们在短期内在这些方面受到限制。

We are we are constrained in those things in the short term.

是的。

Yeah.

每个人都会对此做出反应。

Everyone will respond to it.

没错。

Yes.

但我认为，我们所有经营公司的人，无论你对AGI有多乐观，嗯。

But I think all of us running companies, regardless of how AGI pilled you are Mhmm.

然后就是关于你有多乐观的误差范围，你知道的。

Then comes this error bands of, like, you know, how bullish can you be?

你能承受的利润率是多少？

What's the margins you can afford?

因为世界上有一些外部的、你无法控制的偶然因素可能会出问题，对吧？

Because there are extraneous factors which can go wrong in the world, right, which are outside of your control.

所以大家都在做这些调整。

So everyone is making those adjustments.

这些都是限制因素。

Those are all constraints.

对吧？

Right?

还有各种限制。

And constraints.

所以我认为，这就是我看到的限制所在。

So I think but that's where I see the constraints.

你认为内存是你最关注的最大组件吗？

Is memory the biggest component that you think about?

内存现在绝对是最重要的组件之一。

Memory is definitely one of the most critical components now.

是的。

Yes.

你提到在短期内，人们会增加供应，从而让高价格自行解决问题，是这样吗？

And you said in the short term, do you think just people will ramp up supply, and so high prices will take care of us.

领先的内存公司不可能大幅提高产能。

There is no way that the leading memory companies are going to dramatically improve their capacity.

因此在短期内你面临这些限制，但随着时日推移，这些限制会逐渐缓解。

So you have those constraints in the short term, but they they get more relaxed as you go

出去。

out.

对。

Yes.

但我确实预计这一切都会构成制约。

But I do expect all of this to constrain.

顺便说一下，我认为这将推动许多关于Mhmm的创新。

By the way, I think it'll push a lot of innovations on Mhmm.

我们会让这些技术的效率提升30倍。

We will make these things 30x more efficient.

是的。

Yes.

所有这些其实也在同时发生。

Like, so all that is happening simultaneously as well.

有效。

Works.

确实如此。

Does

这会形成寡头垄断市场吗？

that enforce a oligopoly market?

所以如果你从模型的角度来看，看看许多关于模型如何改进的观点，很多都将是自我提升。

So if you actually look on the model side because if you look at a lot of the views of models and how they're gonna improve, a lot of it is gonna be both self improvement.

所以模型会开始越来越多地编写自己的部分，为自己做更多的数据标注等等。

So the models will start writing more and more pieces of themselves, do more data labeling for themselves, etcetera.

所以现在本质上是一场谁拥有计算资源的音乐椅游戏，

So there's a musical chairs game of who has compute right now, basically,

你的意思是？

you're saying?

没错。

Exactly.

现在谁拥有计算资源，以及相对于整个行业的产能，你能实际扩展到什么程度？

Who has who has compute right now and how much can you actually scale relative to overall industry capacity?

如果每个人大致按比例增长到某个上限，你就实际上设定了一个人能比其他人领先多少的天花板。

And if everybody is roughly pro rata up to some number, you've effectively put a ceiling on how much far ahead somebody can pull versus everybody else.

你觉得这个说法是正确的还是错误的？

Do you think that's a correct statement or an incorrect statement?

我认为用这种方式来思考是合理的，但还有一些事情，你知道的，我来这里是因为我们刚刚发布了Gemma 4。

I think it's a reasonable framework to think about it that way, but there are things which are you know, I'm coming here as we just ship Gemma four.

是的

Mhmm.

对吧？

Right?

这是一个非常优秀的开源模型。

And it's a really good open source model.

我的意思是，

I mean, the

中国模型非常好，但我觉得在中国以外，它也是一个非常优秀的开源模型。

Chinese models are very good, you know, but I think outside of China, you know, it's a very good open source model.

从前沿模型到Gemma四，确实是这样。

You know, the frontier to Gemma four is both Mhmm.

在时间上，差距巨大但也不算巨大。

Huge and not so huge in terms of time.

比如，如果Gemma四基于Gemini三的架构。

Like, if Gemma four is based on Gemini three architecture.

对吧？

Right?

你知道，这真的很奇怪。

You know, it's a very weird thing.

对吧？

Right?

你在谈论一组可以装在U盘里的WAV文件。

You're talking about a set of WAVs, which can fit on a USB stick.

是的。

Yeah.

是的。

Yeah.

太惊人了。

It's amazing.

是的。

Yeah.

所以这就像一种非常疯狂的情况，它不像SpaceX的火箭。

So it's like a really, you know, crazy it's not like a SpaceX rocket.

你知道我想说什么吗？

You know what I guess?

我总是很惊讶，你们能运行一个数据中心好几个月，但最终的输出却只是一个普通文件。

Like I'm always shocked that you run a data center for months and months and months, and then your output is a flat file.

是的。

Yeah.

真的就是如此。

Literally.

这就像是有一个Word文档，而那就是你的模型。

It's like having a Word doc or something, and that's your model.

这太神奇了。

It's amazing.

所以这些特性非常独特，这让我质疑那些框架，想一想我们该如何看待这个问题？

So there are these unique attributes about this, so which makes me challenge those frameworks and say, you know, how should we think about this?

嗯。

Mhmm.

但我觉得，至少在推理层面，你所说的是一种非常合理的思考方式。

But I think it's a reasonable at least on the inference side, what you're saying is a very reasonable way to think about it.

嗯。

Mhmm.

想想看。

Think about it.

但我确实觉得，大家都在试图突破资本主义的激励机制，嗯。

But I do think I do think everyone is trying to figure out how to blow through the capitalist incentive Mhmm.

突破这些限制。

To break through these constraints.

是的。

Yeah.

你知道，这规模巨大。

You know, it's immense.

但正如你所说，世界上内存资源是有限的。

But as you say, there's only so much memory in the world.

所以，任何资本主义的激励都无法真正解决26或27的内存供应问题。

So, like, no capitalist incentive will really solve 26 or 27 memory supply.

这可能是你看到更多分化出现的时代。

That may be the era where you see more divergence

在模型上。

in models.

你知道的。

You know?

而且别忘了，这必须与晶圆产能的提升以及你能够批准这些数据中心相平衡。

And and remember, that has to balance with wafer capacity increasing, you being able to permit those data centers.

因此，这种限制可能没有看起来那么严重。

So this constraint may be less severe than it appears.

对吧？

Right?

所以你得想象一下，你需要的所有东西的完整集合，然后仔细思考一下。

So you have to kinda you have to envision the total square set of, like, all the things that you need and think it through.

对吧？

Right?

你是持续性资本吗？

Are you Recurring capital.

是的。

Yes.

对。

Yes.

但再说一遍，让我感兴趣的是，人们可能会在当前资本支出之外继续投资，但我们现在正面临26年和27年现实世界的限制。

But but, again, what's interesting to me is that, plausibly, people would invest beyond the current CapEx, but we're now just running against twenty six and twenty seven real world constraints.

这完全关乎霍尔木兹海峡。

It's all about the Strait Of Hormuz.

你想要什么样的油价都可以。

You can have whatever price of oil you want.

最终，如果你从系统中移除每天两千万桶石油，你就必须摧毁每天两千万桶的需求。

Ultimately, if you take 20,000,000 barrels a day out of the system, you need to, like, destroy 20,000,000 barrels a day of demand.

这和内存的情况有点类似，最终总得有人得不到他们想要的内存。

And it's kinda similar with memory where, like, ultimately, some people have to not get the memory they want.

但还有其他限制，比如把安全作为一个限制因素。

But there are other constraints, right, like, which, you you know, take security as a constraint.

嗯。

Mhmm.

这些模型肯定会彻底破坏几乎所有的现有软件。

And these models are definitely, like, really gonna break pretty much all software out there.

也许现在就已经开始了，只是我们还不知道。

Maybe already we don't know.

我们坐下来好好聊聊吧。

Let's sit here and speak.

你真的需要所有那些软件吗？

Do you really need all software there?

因为像SSH这样的东西，人们已经尝试破解很久了。

Because, like, SSH, people have been trying to break for a long time.

你觉得这有什么不同吗？

Do you think that's different?

我说的是直接让你的软件、大型平台崩溃。

I'm talking about just think break your software, large platforms.

是的。

Yeah.

对吧？

Right?

有多少个零日漏洞？

How many zero days?

对。

Yeah.

你知道的，所以系统里存在一些限制。

You know, so there are constraints here in the system.

嗯。

Mhmm.

对吧？

Right?

你不能光靠愿望就让它消失。

You just can't wish away.

对。

Right.

有人告诉我，由于人工智能导致零日漏洞的供应增加，黑市上的零日漏洞价格正在下降，我觉得这真的很有趣。

Somebody was telling me the black market price of zero days is dropping because supply is growing due to AI, which I thought was a really interesting Yeah.

市场。

Market

一点也不意外。

Not at all surprised.

对吧？

Right?

一点也不惊讶。

And not at all surprised.

但它是如何在社会中实际传播的呢？

So but when it how does it practically diffuse through society?

这会带来什么影响？

What are the implications of it?

是的。

Yes.

对吧？

Right?

就像你知道的，我认为存在一些相似之处。

Like, you know and so there are parallels, I think.

所以我认为可能存在一些隐藏的限制。

So I think there could be hidden constraints.

是的。

Yes.

而且系统可能会出现冲击，这么说吧。

And there could be shocks to the system, if you will.

但话又说回来，我真的认为前方有很多积极面。

But having said that, like, you know, I I genuinely think there's a lot of upside ahead.

一些限制可能反而有帮助。

Some of the constraints maybe are helpful.

是的。

Yes.

对吧？

Right?

我认为限制能激发创造力。

I think constraint inspires creativity.

迫使我们进入一个精简的循环，其中

Forces a compaction cycle where we

我们能获得更高的效率。

get more efficiency.

这些压力可能会促使一些原本不会发生的重大对话进行。

Forces may be important conversations to be had, which otherwise wouldn't wouldn't happen.

是的。

Mhmm.

对吧？

Right?

我觉得，就我提到的安全问题而言，我已经意识到我们需要更多的协调，是的。

I think, you know, just on my security point alone, like, I've thought about we are going to need more coordination Yes.

但今天并没有发生。

Which is not happening today.

将会有一个时刻，你知道的，可能是一个急剧的时刻。

There will be a moment of you know, it could be a sharp moment.

对吧？

Right?

而且，你知道的？

And, like you know?

所以所有这些事情，我认为你无法指望它们自行消失。

And so all those things, I I don't think you can wish them away.

是的。

Yes.

是的。

Yes.

对。

Right.

没错。

Yeah.

实际上，与此相关的是，谷歌拥有一个非常出色的资产组合，这些资产既有自主研发的，也有收购而来的。

Actually, related to that, Google does have amazing portfolio of things that is both built and bought into.

从所有权的角度来看，你持有太空探索技术公司相当可观的股份。

From a ownership perspective, you own a reasonable amount of SpaceX.

我不清楚具体数字，但记得当初大概是10%左右。

I don't know the exact amount, but I think it was 10 ish percent way back when.

Anthropic，大约10%。

Anthropic, 10 ish percent.

Waymo的大部分，这是一件了不起的事情。

The majority of Waymo, which is an amazing thing.

然后，内部显然还有大量令人惊叹的技术被开发出来。

And then, internally, obviously, there's this enormous swath of amazing technology that's been developed.

我们谈到了人工智能和变换器。

We talked about AI and transformers.

还有TPU。

There's TPUs.

显然，Waymo也是其中之一。

Obviously, Waymo was another one of these things.

还有量子计算。

There's quantum.

你知道，你们刚刚发布了一个非常有趣的结果

You know, you just released a very interesting result

在那里。

there.

嗯哼。

Mhmm.

还有哪些其他人应该了解的隐藏亮点，或者特别有趣、可能产生重大影响的？

Are there other hidden gems that people should know about or that are especially interesting or that may have very big impact in

或者人们可能低估了。

the Or maybe people maybe underestimates.

是的。

Yeah.

听好了。

Look.

我们一直在努力推进这些长期项目，当你最初公布它们时，看起来似乎有点荒谬。

We are constantly trying to take these long term projects, which when you first announce them slightly marginally looks ridiculous.

嗯哼。

Mhmm.

你知道吗，我们目前还处于思考太空数据中心的最早阶段。

You know, like, we're in the earliest stages of thinking about data centers in space.

对吧？

Right?

但回到你之前关于限制激发创造力的讨论。

But to your earlier discussion around constraint inspires creativity.

但如果你放眼二十年后，这些数据中心大多会建在哪里呢？

But if you take a twenty year outlook, right, where are you going to put most of these data centers?

嗯。

Mhmm.

这些都是非常难解决的问题。

Really hard problems to solve.

但这些正是我们现在正在思考的项目，就像2000年或2010年的Waymo一样。

But those are examples of projects we think about today, which are Waymo in 2000, like, '10.

嗯。

Mhmm.

量子计算本身就是这样一个项目。

Quantum itself is one of the one of these projects.

我们正在以非常深入和专注的方式推进这方面的进展。

We are, like, in a deeply committed way making progress there.

是的。

Mhmm.

我对它感到非常兴奋。

And I'm excited about it.

你认为量子计算会在哪里产生最大的影响？

Where do you think quantum will have the biggest impact?

因为人们主要谈论分子建模，还有密码学。

Because mainly people talk about molecular modeling, they talk about cryptography.

人们一直在开发抗量子攻击的密码学技术。

There's quantum proof sort of cryptography that people have been developing over time.

在分子建模方面，实际上在某些情况下，深度学习模型表现得非常出色。

On the molecular modeling side, it actually looks like the the deep learning models tend to be very good at that in certain circumstances.

我的意思是，你们用AlphaFold率先实现了这一点。

I mean, you all pioneered that with AlphaFold.

你认为量子计算真的会很重要吗？

Do you think quantum will actually matter?

如果会，你认为它会在哪里产生最大的影响？

And if so, where where do where do you think it'll have the biggest impact?

听好了。

Look.

从抽象层面来说，我觉得我们的目标是越来越深入地模拟自然。

At abstract level, to me, it feels like to simulate nature more and more.

你知道，既然自然本质上是量子的，要更好地模拟它，就需要量子系统。

Like, you know, like, given it's inherently quantum, you would need quantum systems to better simulate it.

我们也许能通过经典计算技术以一种出人意料的方式实现，或者通过足够的压缩、足够的抽象来达到目标。

We may get there with classical computing techniques in a surprising way or get at it with enough compression, enough, you know, abstraction.

这也许行得通。

It may work.

但我从根本上觉得量子计算在这一点上会有优势。

But I fundamentally felt like quantum would have an edge there.

是的。

Mhmm.

我不知道。

And I don't know.

我们仍然不了解化肥的作用机制。

We still don't understand the behavior process for fertilizer.

我的意思是，这可能涉及很多复杂因素，我想这跟你大学时的研究背景有关。

Like, there are many complex I mean, you know, it's probably your background, going back to what you did in college more.

所以我的直觉告诉我，可能会有，嗯，

So my, you know, my instinct tells me there'll be, you know Mhmm.

模拟天气，模拟现实。

Simulating weather, simulating, you know, reality.

是的。

Mhmm.

我认为，所有这些都将具有优势。

All that, I think, will have an advantage.

我认为技术发展的历史是，当某项技术达到可规模应用的水平时，它就开始发挥作用。

I think the way the history of technology is you get something to a scale where it works Mhmm.

然后人们开始使用它，而且。

And then you use it, and Mhmm.

人们在上面发挥创造力，找到各种应用。

People's creativity on the top finds the applications.

所以，你知道，我总是举这个例子：手机加上GPS，催生了Uber。

So, you know, I mean, I always give this example of mobile phones plus GPS enabled Uber.

是的。

Yeah.

就像，是的。

Like like Yeah.

是的。

Yeah.

没有人从事手机工作的人会预测到这种平台转变会带来这样的结果。

There's nobody who is working on phones who would predict that as an outcome of this platform shift.

所以，我相信量子计算将会有很多很多的应用，前提是它真能实现。

So, you know, confident quantum will have many, many, many applications Mhmm.

如果你真能让它运行起来的话。

If you can actually make it work.

是的。

Yeah.

这就是我对它的看法。

So that's how I I think about it.

抱歉。

So sorry.

我们打断一下你。

We interrupt you.

你正在谈你最喜欢的谷歌的远期项目。

You're talking about kind of your favorite of the Google further afield.

我认为，我们正在让通用汽车团队深入思考机器人技术。

I think we're making you know, the GM team is deeply thinking through robotics.

对。

Right.

是吧？

Right?

你知道，机器人技术是我们公司以前过于超前的一个领域。

And, you know, robotics is an area where we were too early as a company before.

结果发现，人工智能是十五年前或十年前许多想法所缺失的关键要素。

It turned out AI was the missing ingredient for a lot of ideas maybe fifteen years ago or ten years ago.

但你知道，Gemini 机器人模型主要关注空间推理等方面。

But, you know, the Gemini robotics models are sorta on spatial reasoning, etcetera.

所以我们在这一领域确实拥有最先进的模型。

So we definitely have state of the art models there.

而且，以一种颇具讽刺意味的方式，我们正与波士顿动力、Agile 以及其他几家公司展开合作，并坚定地取得进展。

And we are partnering back in an ironic way with Boston Dynamics and and Agile and a few other companies and and, in a determined way, making progress.

而且那里还有一些非凡的初创公司。

And there are extraordinary startups out there as well.

嗯哼。

Mhmm.

所以我们正在投资，比如我提到过的量子计算、太空数据中心，以及与Wing合作的无人机配送。

But so we are investing in you know, I spoke about quantum, data centers in space, drone delivery with Wing.

嗯哼。

Mhmm.

我觉得我们正在扩大Wing的规模，在不久的将来，大约四千万美国人将能使用Wing的配送服务。

You know, I think we are scaling up Wing where, some reasonable time period, like, 40,000,000 Americans will have access to a Wing Wow.

配送服务。

Delivery service.

对吧？

Right?

我说的不是几年之后的事，也不是那种遥远的事情。

And I'm not talking years out or something like that.

但再次强调，当你着眼于这些长期项目时，这些都是系统性的复利积累。

But, again, these are all, like, methodical compounding when you take these long term projects.

所以，我们是认真的。

So, you know, we we are committed.

同构的？

Isomorphic?

嗯。

Mhmm.

是的。

Yeah.

同构非常令人兴奋。

Isomorphic is very exciting.

感觉没错。

Feel yeah.

想想看，专注于这些模型，有针对性地改进药物发现中的每一个可能步骤。

Think about you know, think about being focused on these models in a targeted way, improving all the possible steps in drug discovery.

即使你有像三期临床试验这样的长期流程，但成功的机会要高得多。

And even though you have long polls like phase three trials, etcetera, getting there with a much higher probability of success.

是的。

Yeah.

我认为这绝对是我在各种生物模型中见过的最聪明的方法，真正超越了分子设计本身，去思考更广泛的转化，而我认为大多数公司都卡在分子设计这一环。

I think it's definitely the smartest approach I've seen in terms of the different biomodels and really thinking about the broader swap beyond just the molecular design, which is I think where most of them are stuck.

所以，对。

So Yeah.

这看起来非常聪明。

It seems very smart.

我可以问一下吗？我很想知道，谷歌的资本配置实际上是怎样的？

Can I ask, I'm curious, how capital allocation actually works at Google?

我的意思是，好的资本配置在于内化资本的机会成本，将企业产生的现金流投入到最高价值的用途中。

And what I mean by that is, you know, the idea good capital allocation is about internalizing that the opportunity cost for capital and putting the cash that a business generates towards its highest and best use.

在商学院教科书里的一个简单例子中，你可能是波音公司。

And in the toy example in a business school book, you know, maybe you're Boeing.

我们可以选择，我们公司产生的现金，要么去竞标下一个国防合同，投入这么多研发资金，并预测合同带来多少收入；要么从零开始开发一款全新的商用客机，投入这笔资金，并预测这种结果。

And we can either you know, we have this cash that our business generates, and we can either go bid on the next defense contract, and we'll invest this much in R and D dollars, and we model this much revenue from the contract, or we go develop a clean sheet commercial airliner, and we'll put in this money, and we model this kind of thing.

这相当于16%的内部收益率对19%的内部收益率。

It's like a 16% IRR versus a 19% IRR.

好的。

Okay.

我更倾向于19%的那个。

I prefer the 19%.

在谷歌的情况下，这些项目极其多样化，比如，我们可以给YouTube团队更多资金，让他们改进推荐算法，从而增加用户停留时间并提升变现能力。

In Google's case, the projects are extremely heterogeneous, where it's like, okay, we can give the YouTube team more funding so they can go, you know, improve the recommender algorithm and therefore time on-site increases and so does monetization.

或者我们可以给Waymo团队更多资金，让他们更快进入市场或更快扩大规模。

Or we can give the Waymo team more funding so that they can actually get to market faster or scale up faster.

或者我们可以投资于一种新的AI方法，这种技术可能在五年后才见到回报。

Or we can invest in this new AI approach that might, you know, pay off in five years' time.

因此我很想知道，当你试图将资本配置到最高效益的用途时，最终需要进行比较，你们是如何比较这些性质迥异、回报曲线差异巨大的项目呢？

And so I'm curious, if you are trying to put capital towards the highest and best use and you're ultimately comparing, how do you compare initiatives that are so different in nature and so different in payoff curve shape?

这是

This is

史上最约翰的问题了。

the most John question ever.

我需要知道，你得提出一个OIC

I need to know You need to throw an OIC

然后这就像是

and then it's like

这是个好问题。

It's a good question.

听我说，今天我比以往任何时候都更感受到这一点，讽刺的是，这跟TPU分配有关。

Look, I feel it today more than ever, ironically, because of TPU allocation.

所以在某些方面，我觉得Waymo也需要TPU。

So in some ways, I feel even Waymo needs TPUs.

对吧？

Right?

但你知道，计算机让这个问题反而变得更加突出。

And so but, you know, computers made the question ironically much more front of mind.

顺便说一句，在我所有做的事情中，我非常期待AI作为伙伴至少能为这项任务提供输入。

By the way, of all the things I do, I'm really looking forward to how AI, as a companion, at least gives inputs to this task.

你知道的吧？

You know?

我认为，一旦我们能把所有数据连接并顺畅流动起来，模型本身已经具备能力了。

And I think once we can actually get all the data connected and flowing through, I think the models are already capable.

更重要的是，我觉得把所有数据都释放出来会很有帮助。

It's more, you know, getting all the data unlocked, I think, will be helpful.

所以我在这方面有这种感觉。

So I feel it there.

历史上，我认为谷歌的一个优势是我们有时会在周期初期就做出这些决策。

Historically, I think at Google, one of the advantages we have had is sometimes we make these decisions very early in the cycle.

所以这几乎像是回归到那种以深度技术为导向的初心。

So it's almost like going back to that truth as a a deep technology orientation.

而且，你知道，我们其实经常思考你刚才问的那个问题，比如那些长期性的事情。

And, you know, we actually think about the question you were asking a lot a bit ago about, like, what are those longer term things?

是的。

Mhmm.

所以我认为，在这个阶段思考更容易，因为初期的资金投入可以更少。

And so I think thinking at that stage, it's easier because your initial funding amounts can be smaller.

但接着，你知道，你要长期坚持，同时确保以深入的方式取得进展。

But then, like, you know, you stay committed for the long term, but you're making sure you're, like, making progress in a deep way.

所以只要你看到底层技术的发展，比如量子计算就是一个例子。

So as long as you're seeing that underlying tech like, take quantum, for example.

我们该如何评判它呢？

How do we judge it?

比如，我们是在评判底层技术的进步。

Like, we're judging the underlying Yes.

你知道，你会设定一些目标，比如什么时候能实现逻辑量子比特的纠错，达到稳定的大规模逻辑量子比特阈值，以及团队是否能够实现这些目标。

Like, you know, so you have goals around, you know, what logical qubit error corrected, large stable logical qubit threshold by when you're gonna get to, and is the team able to do that.