萨提亚·纳德拉谈AI末日论的退潮、微软的总体规划以及谁将在AI竞争中胜出

他为微软创造了2500亿美元的市场价值，没错，是十亿级别的。

He has generated $250 billion, with a B, in market value for Microsoft

萨提亚·纳德拉，微软的董事长兼首席执行官。

Satya Nadella, chairman and CEO of Microsoft.

自从你担任CEO以来，三年半的时间里，股价上涨了大约，我猜，大概120%。

Since you've been the CEO, three and a half years, the stock is up about, I guess it's about 120%.

我对我那800亿美元很有信心。

I'm good for my 80 billion.

我打算投入800亿美元来建设Azure。

I am going to spend $80 billion building out Azure.

也许工业革命之后，这是最重大的事情。

Maybe after the Industrial Revolution, this is the biggest thing.

这就是我们前沿模型的目标。

That's our goal with our frontier model.

我们的模型应该成为他们能使用的最佳基础模型。

Our model should be the best model that they can use as a base.

我们创造技术，是为了让别人能创造更多的技术。

We create technology so that others can create more technology.

这就是我们的定位。

That's who we are.

我们是工具制造者。

We're a toolmaker.

让我们欢迎萨提亚·纳德拉。

Please welcome Satya Nadella.

好的。

All right.

老兄，见到你真好。

My guy, good to see you.

萨提亚，你好。

Satya, hi.

谢谢你能来。

Thanks for coming out.

很高兴见到你。

Good to see you.

大家早上好。

Good morning, guys.

你好吗？

How are you?

挺好的。

Good.

谢谢你加入我们。

Thanks for joining us.

周末真是疯狂，不过我们还是来了。

Crazy weekend, but here we are.

我们需要控制前沿技术的发展节奏吗？

Do we need to pace the frontier?

那么，我们先从常识说起，那就是我们应该尽一切努力，去建设那些首先服务于人类、并且由人类掌控的东西。

So, let's start with the common sense part first, which is we should do what it takes to build stuff that serves humanity first and is in human control.

我们居然得从这种常识说起，这有点疯狂，但我觉得这是个好的起点。

It's kind of crazy that we have to start with that level of common sense, but I think it's a good place.

然后，当我想到怎么控制节奏时，我首先相信的是，这项技术的广泛普及是最关键的，因为这项技术到处出现带来的好处才是真正的重点。

Then when I think about pacing whatever, the first thing that at least I believe is the broad diffusion of this technology is the most critical thing because the benefits of this tech showing up everywhere is really what it's all about.

所以归根结底，如果你说要服务人类，那就让它真正触达人类——

So at the end of the day, if you say serving humanity, let it actually reach humanity-

……以服务人类的方式，这意味着你必须要有选择，要有竞争，要有各种商业模式，无论是开放权重还是封闭权重，等等。

... in ways that it serves humanity, and that means you got to have choice, you have to have competition, you have to have all kinds of business models, whether they're open weights, closed weights, what have you.

然后，我觉得在谈论控制时，还有一个没被提到的方面，就是客户、企业或商家对这项技术的控制，因为有时候它太不透明了。

Then the other aspect I think that is not talked about when we talk about control is actually the control that, for example, customers have, enterprises or businesses have around this technology because sometimes this is so opaque.

我想要我的隐私。

I want my privacy.

我希望能够把我的知识嵌入到一组我控制的权重中。

I want to be able to embed my knowledge in a set of weights I control.

我想看到所有生成的思维链。

I want to see all of the CoT that's being generated.

我想用它来微调我自己的模型。

I want to use it to do fine tuning of my own models.

我的知识产权不应该泄露。

My IP shouldn't leak.

所以有一整套事情，没有人——

So there's an entire body of things that nobody's-

……经常谈论，那就是我真的想确保这项技术掌握在我自己手里。

... talking about as much, which is I really want to make sure that this tech is in my control.

然后我们谈到我认为真正关乎安全的问题，我们应该认真对待，那就是我们应该花所有需要的时间来测试。

Then we get to what is, I think, a real issue of safety, and we should take it seriously, which is we should take all the time we want to test things.

事实上，我非常喜欢引入第三方测试这个想法。

In fact, I love this idea of having third-party testers.

哇，真的。

Oh, wow.

我在一家一直做测试的公司长大。

I grew up in a company that's always done testing.

所以，说他们引入了嵌入式第三方测试人员，这还挺新鲜的。

So it's novel that we should say, wow, they're having embedded third-party testers.

为什么不行呢？

Why not?

这是个好主意。

It's a great idea.

其实，我只想说，我们应该避免那种谁测试什么、谁有权访问什么的过于亲密安排，而且测试范围应该广泛。

In fact, the only thing I would say is we should avoid these cozy arrangements of who's testing what, who has access to what, and it should be broad.

不过，当那篇文章发表后，前沿公司似乎开始抱团了，你当时感到意外吗？

Were you surprised though then when both the essay landed, and then it seemed like there was a circling of the wagons amongst the frontier companies?

我怀疑这确实源于一个真实的地方，当你开始看到——实际上，这很有趣。

My suspicion is it comes genuinely from this place where when you start seeing, in fact, it's fascinating.

当你开始看到奖励黑客行为，以及这些智能体群体在这些环境中发生的事情时，有平凡的一面。

When you start seeing reward hacking, and what's happening in these environments with these agent swarms, there is the mundane.

比如某个DevOps错误，有人配置错了容器。

There is some DevOps error where somebody misconfigured a container.

或者这些API密钥。

Or these API keys.

或者一个API密钥。

Or an API key.

对，没错。

Yeah, exactly.

没有监控。

There's no monitoring.

有互联网访问。

There's internet access.

还有经典的，我会称之为基本的DevOps。

There's classic, I would call it basic DevOps.

然后还有真正新颖的东西，那就是这些持久化代理的奖励黑客行为等等。

And then there is real novel new stuff, which is what is this reward hacking with these persistent agents and so on.

而在这一点上，我承认科学还没有跟上。

And that's a place where I'll admit that the science is not there.

我觉得雅各布那篇文章写得很好，他把它叫做“我们在培育智能，而不是在建造智能”。

I thought Jacob's post, which is a good one, which he said he called it, "We're growing intelligence, not building intelligence."

所以这是一门实验科学，而实验性越强的科学，你就越需要确保在受控环境中进行那些实验。

So it's an experimental science, and so the more experimental a science is, then you really need to make sure you're doing those experiments in controlled environments.

如果要说的话，我希望的是，即使是像Hugging Face事件那样的地方，也能有更多的透明度，说明需要什么条件。

If anything, the place where I would love is taking even the Hugging Face incident and other places, more transparency on what would it take.

事实上，现在一个很吸引人的问题是内部人员风险。

In fact, one of the fascinating things right now is the insider risk.

你想想看。

Think about it.

如果你在企业里工作，顺便说一句，这一切都是测试时计算。

If you're sitting in an enterprise, this is all test time compute, by the way.

所以并不是说，哦，这只会发生在训练运行的时候。

So it's not like, oh, it's going to only happen when in some training run.

它可能发生在一个非常普通的任务上，比如我在企业里给某个前沿模型布置任务时，我跟大卫说过，假设我说：“嘿，去优化我的营运资金。”

It can happen for a very mundane task that I give one of these frontier models inside an enterprise where I say, I was telling David this, suppose I say, "Hey, go optimize my working capital."

它可能会伪造我的账本。

It may fake my books.

因为这是一种新型的内部风险。

Because this is like a new type of insider risk.

那么，怎么应对呢？

And so what is the way to do that?

我会说，去构建一个因果模型，比如一个语义模型，来实际检查和验证。

I would say, oh, go build maybe a causal model, like a semantic model that actually checks and verifies.

所以我觉得有很多产品建设，我要说的是让系统更稳健，这是经典的工程问题，我们应该更透明地讨论，而不是说：“嘿，这太神秘了，我们搞不懂。”

So I think there's a lot of product buildings, I would say making things more robust, which is classic engineering that we should be talking a lot more about transparently versus saying, "Hey, this is so mystical that we can't figure this out."

你相信这种神秘的说法吗？

Do you buy this argument that it's mystical?

我相信我们确实不了解潜在空间。

I buy the argument that we do not understand the latent space.

不过，就像你说的，我们真的了解大脑吗？

Other than I thought, as you said, do we understand the brain?

我们确实不了解。

We don't.

我们做功能性磁共振成像，研究神经科学，试图弄清楚这个问题，不断获得更好的理解。

We do functional MRIs and do neuroscience, and we're trying to figure this out, continuously getting a little better understanding.

所以在这个意义上，我确实认为我们还没有完全理解。

So I do think that in that sense, we don't exactly have a complete understanding.

顺便说一句，这就是为什么我不相信新版本。

That's why, by the way, I don't believe in new release.

所以这就是为什么我认为要确保思维链用我们都能理解的语言表达。

So that's why I think making sure that the CoTs are in language that we can all understand.

事实上，它们是透明的，这样当我回到一个使用所有这些模型的企业时，如果你有完整的思维链，你就可以——

In fact, they're transparent, so that when I go back to an enterprise that's using all these models, and if you have the full CoT, then you can-

思维链

Chain of thought

……思维链。

... chain of thought.

这样你就能真正深入地去审视它。

And so then you can really go look at it deeply.

实际上，你可以有多个模型，然后对比它们之间的思维链。

In fact, you can have multiple models, and you can look at the CoT across those.

我觉得这些都是将来会变得非常重要的东西。

I think these are all things that I think will become very important.

萨提亚，你跟技术人员打了这么多年交道。

Satya, you've worked with technologists for decades

作为微软这样一家与公众沟通非常清晰的公司的领导者，当你看到达里奥和他的团队的情况，有人公开说“我们有10%的概率会全部灭亡”，你觉得那些技术人员心里到底在想什么？

And when you see, as a leader of one company, Microsoft, which has very crisp communications with the public, and you see what's happening with Dario and his team, people coming out saying, "10% chance we all die," what do you think is going through those technologists' minds?

你觉得他们是真的相信这会毁灭人类，还是他们陷入了某种精神错乱，或者他们在那些前沿模型上看到了什么让他们感到恐惧的东西？

Do you believe they actually believe that this is going to kill humanity, or are they going through some psychosis, or are they seeing something working on those frontier models that is terrorizing them?

你不是心理学家，但你跟技术人员共事很久了。

You're not a psychologist, but you have worked with technologists for a long time.

你评估一下，这些组织里到底发生了什么，让人们觉得有必要辞职，并且说我们都会死。

Handicap what's going on in these organizations that's making people feel the need to resign and say we're all going to die.

我很难说这些地方具体发生了什么，但就说我自己在微软内部的成长经历吧。

It's hard for me to speak to what's happening in any of these places, but let's just say how I grew up, even inside of Microsoft.

比如，作为早期工程负责人，你学到的最大的一课之一就是如何处理致命错误。

For example, one of the biggest things you learn as an early sort of engineering lead is how to deal with a showstopper bug.

这就像入门第一课，对吧？

I mean, that's kind of like 101, right?

就是问题就摆在你面前，你怎么办？

Which is right on your face you have a bug, what do you do?

是停下来修复，还是推迟，或者直接说‘嘿，这只是个极端情况’？

Do you stop and fix or you defer or you go in and say, "Hey, this is such an edge case?"

这就是判断力。

That's kind of the judgment.

而且随着风险越来越大，你更倾向于……

And as the stakes go up, you want to.

就像事务处理，我记得做数据库的时候，对吧？

But like transaction processing, I remember working on databases, right?

哇。

Wow.

你得非常认真对待任何可能导致交易丢失的bug，对吧？

You've got to take very seriously any bug where if the transaction's going to get lost, right?

数据丢失是必须停下来处理的事情。

Data loss is a thing that you stop the thing for.

所以我觉得在AI行业，我们有点在重新发现这种文化，可能是因为当你看到，他们可能比其他人更早看到那些致命问题，如果你看到致命问题，就停下来。

So I feel a little bit culturally in the AI industry rediscovering maybe because when you see, and it's possible that they see stuff which are showstoppers before the rest, and if you see a showstopper, stop the show.

所以修复bug。

So fix the bugs.

但是当你看到Hugging Face那次运行，它非常表演性质，Dwarkesh发了他的整个帖子，关于文明，你对那个测试有什么看法？

But when you saw the Hugging Face run, and it was super performative, Dwarkesh did his whole post, civilizations, what's your take on that test they ran?

因为他们本可以运行一个测试，让3000个代理去防御一堆网站。

Because they could have run a test where they had 3,000 agents defend a bunch of websites.

相反，他们指示代理去攻击网站和隐藏信息，所有这些对代理的拟人化描述，等等。

Instead, they instructed them to hack websites and the hiding of information, all this anthropomorphizing, whatever, of the agents.

我的理解是，这其实基本上是在为 Cyber Gym 做一个评估。

I mean, the way at least I understand it was, it was actually basically trying to do an eval for Cyber Gym.

而且据我了解，基于那个评估，它找到了一种方法，就是直接奖励黑客行为。

And as I understand it, given that eval, it sort of figured out a way to say let's just reward hack.

这就是它最终跑到 Hugging Face 的原因。

And that's what led it to Hugging Face.

实际上，这恰恰说明了现在最明显的问题，就是如果这些长期运行的持久化智能体，它们基本上会变成新的内部威胁。

In fact, it speaks to, I think, what's the clear issue right now, which is you can have these things, if they're long-running persistent agents, become essentially new insider risks.

所以我会从最基础的问题开始，比如，隔离措施应该是什么样的？

And so I would start from the very basics of saying, okay, what does containment look like?

举个例子，我觉得将来会成为一个大问题、而且需要很好解决方案的事情，就是对智能体行为进行真正严格的、基于行为的监控。

So for example, one of the things that I think is going to be really an issue and a thing that needs great solutions is true aggressive monitoring of agent activity that's behavioral.

证据。

Evidence.

证据。

Evidence.

所以一切都必须可审计。

And so everything has got to be auditable.

然后它访问的每个对象，对吧？

And then every object it accesses, right?

如果它去拿一个秘密，或者它要串联几件事，你应该能在它开始串联几个漏洞去攻击的时候看到它。

If it goes and gets a secret or it's going to go chain a couple of things, you should be able to see it when it's starting to chain a couple of vulnerabilities to go hack.

所以我认为这些才是你真正需要应对这些情况的方式，而不是说——事实上，我觉得我的核心观点是，我们必须让围绕构建这门实验科学的工程流程更加稳健。

And so I think that these are the ways that you really have to sort of deal with these situations versus saying-- In fact, I think the core of my take is we will have to get the engineering process around building out this experimental science to be more robust.

萨克斯？

Sacks?

我觉得这点说得很好。

So I think that's a great point.

我喜欢你在Hugging Face那集里对平凡错误的区分，比如配置错误的沙箱，还有Hugging Face的破解凭据就放在公共仓库里，而且没有任何监控。

I love how you differentiated in the Hugging Face episode between the mundane things they got wrong, like the misconfigured sandbox and the Hugging Face hack credentials just sitting in a public repository, and there was no monitoring.

然后你还有真正新颖的行为，比如智能体集群、奖励黑客。

And then you have the genuinely novel behavior, the swarms of agents, the reward hacking.

这些才是让大家感到恐慌的东西。

That's the stuff that has everyone freaked out.

我同意，我们现在必须想办法修复这些漏洞，或者解决奖励黑客背后更深层的问题。

I agree that we have to now figure out how to fix the bugs or fix the deeper problem that's coming from that reward hacking.

你觉得这对未来意味着什么？而且，我觉得前沿实验室做得对，他们现在说要放慢所谓原始算力提升的速度，转而注重可靠性和可预测性，也就是他们所说的对齐，这我觉得是明智的商业做法。

What do you think that means for-- And I think to their credit, I think what the frontier labs are saying is we are now going to slow down the pace of, let's say, raw power and shift towards reliability and predictability and what they call alignment, which I think is good business practice.

那么，你觉得这对未来一两年我们能看到的新产品意味着什么？

I guess, what do you think that means for what we see in terms of new products for the next year or two?

是意味着我们只是改进现有的东西，还是会有新的能力出现？

Does it mean we just kind of improve what we already have, or do we see new capabilities?

你觉得这对未来意味着什么？

What do you think this is going to mean for the future?

好问题，大卫。

It's a great question, David.

我确实认为已经存在大量的模型积压，对吧？

I do think there's already a massive model overhang, right?

能力过剩的意思是模型本身已经很好了，但要广泛推广还需要很多条件，对吧？

Capability overhang in the sense of the models are very good, except the broad diffusion requires a lot of things, right?

这甚至需要，基本上如果你要压缩工作流程并改变工作流程的执行方式，那么为了整合这些系统所需的大量变更管理，正是需要时间的地方。

It even requires, essentially if you're compressing workflows and changing workflows to happen differently, the amount of change management that needs to happen in order to even incorporate these systems is sort of what's taking time.

所以从某种程度上说，我会说——而且还有能力去创造这些新的形态，对吧？

So to some degree, I would say the-- And also the ability to create these new form factors, right?

我的意思是，想想编程代理，编程代理变得真正可用，你发现你可以有一个带文件系统的代理循环。

I mean, if you think about coding agents, and coding agents became really usable, and you discovered that you could have an agent loop with a file system.

那就是让编程代理真正起作用的突破。

And that was the breakthrough that just made coding agents work.

而且我觉得现在也许有了Kua，对吧？

And I think now maybe with Kua, right?

所以，也就是通过Astra和Kua，我们甚至可以实现计算机操作，或者我们只是用长轨迹任务，这些任务可以完全自动化。

So which is with Astra, with Kua, could be a way for us to even do computer use, or we just use long trajectory tasks that can get completely automated.

所以我认为这类产品创新，模型加上辅助工具让我们能够做到一些事情，从而带来广泛采用。

So I think these type of product innovations where the model plus the harness allow us to do things that then lead to broad adoption.

我甚至回想起ChatGPT刚出来的时候，正是最后的RLHF让对话成为可能。

I even go back to the ChatGPT moment for me, which was it was that RLHF at the very end that made a chat conversation possible.

所以我觉得，是的，有科学，也有某种形式因素，然后才带来广泛普及，我们现在需要找到下一批能在真实企业中真正干活的这些东西。

And so I think that, yes, so there's some science, there is some form factor that then leads to broad diffusion, and we now need to find the next level of these things that are doing real work in the real enterprise.

而且在这个背景下，顺便说一句，另一个问题是，这将是一个多模型的世界。

And in that context, by the way, the other thing is it's going to be a multi-model world.

所以现在，仅仅为了韧性。

So at this point, just out of resilience.

想想看，现在每个企业都会来找我说：“嘿，这个模型在这里会拒绝回答。

Think about every enterprise now comes to me and says, "Hey, this model does refusals here.

这个模型，我想要这里的权重。”

This model, I want weights here."

所以人们会想要多个模型。

And so the people are going to want multiple models.

所以我们需要搞对的另一件事，就是一些互操作的标准。

So one of the other things that we have to get right is some standards of interop.

比如KV缓存，为什么我就不能同时用多个模型家族，还能复用KV缓存呢？

Like even KV cache, like why the heck can't I use multiple model families and have KV cache reuse?

我们已经有文档标准了。

We've had document standards.

你我都经历过那个时代，对吧？

You and I lived through it, right?

在现实世界的其他领域，很多东西都是可以互操作的。

You have things that are interoperable in the real world everywhere else.

所以我觉得这个行业也得醒一醒，说：嘿。

So I think this industry also has to wake up and say, "Hey."

其实，如果要谈最紧迫的事情，那就是怎么在互操作性上建立更多标准。

In fact, if I were talking about the most important pressing things is how do I have more standards on interoperability?

怎么才能有一个独立于模型的框架，让我的记忆不绑定在某个模型上？

How do I have a harness that is external to a model so that my memory is not tied to one model?

没错。

That's right.

这是你第一次遇到这样一种技术，你使用它产生的数据和废气可能都不属于你。

This is the first time you're going to have a technology where your use of it and the exhaust in the data could not be yours.

这就像我卖给你一个数据库，然后说：‘嘿，你放进数据库里的数据不是你的，是我的。’

It's like if I sold you a database and said, "Hey, the data you put into your database is not yours and it's mine.

如果我收回许可证，这些数据就没了，你会怎么想？

It goes away if I took away the license," how would you feel about it?

所以，我觉得我们确实有一些这样严重的问题需要处理。

So therefore, I think we have some serious issues like that to deal with.

我觉得这是个很好的过渡。

I think that's a good segue.

抱歉，我想问一个问题，把经济激励的论点跟当前的情况联系起来。

Sorry, let me just ask one question to connect the economic incentive argument on what's going on.

论点是前沿实验室面临token压缩，OpenAI每百万token输出收费50美元，而据估计DeepSeek的新模型每百万token输出可以低至15美分。

The argument is the frontier labs are facing token compression, 50 bucks for OpenAI's kind of million token output versus I think someone estimated DeepSeek's new can go as low as 15 cents for a million tokens of output.

就算按60美分算，成本也降低了99%。

Let's call it 60 cents, 99% cost reduction.

如果这就是前沿实验室面临的核心经济问题，那为什么大多数企业还要为大部分任务支付50美元，而不是60美分呢？

If that is the big kind of economic crux of what the frontier labs are facing, why would most tokens be paying 50 bucks or most enterprises pay 50 bucks when they could pay 60 cents for most of their tasks?

这难道不也引出一个问题：他们的商业模式是不是错了？

Doesn't that also beg the question, are they in the wrong business model?

我作为微软的CEO问你，正确的商业模式是什么？

And I ask this for you as the CEO of Microsoft, what's the right business model?

你是想打造前沿模型吗？

Do you want to be making the frontier model?

你是想运营计算资源，靠出租计算能力收费，还是想专注于应用层？

Do you want to be running the compute, charging for rent on your compute, or do you want to be in the application layer?

我知道你经常谈论这个话题——

I know you talk about this a lot-

……但我很想听听你从今天的角度怎么看，这一切是如何——

... but I'd just love your perspective from where we sit today and how this all kind of-

我认为我们观察到的最根本的事情，就是老式的竞争。

I think the fundamental thing that I think we are observing is good old-fashioned competition.

对我来说，回头看的话，我们确实有一些非常棒的闭源资产。

For me, if I look back at it, we had some real great closed source assets.

Windows。

Windows.

那它的制衡是什么？

What was the check against it?

当然是Mac，但还有Linux。

It was, of course, the Mac, but Linux.

Linux。

Linux.

我们有一个很棒的闭源产品叫SQL Server。

We had a great closed source product called SQL Server.

那它的制衡是什么？

What was the check against it?

总有替代品，比如Postgres或MySQL。

There was always a substitute called Postgres or MySQL.

所以我觉得现在的情况就是这样。

So I think that's what's happening.

有一部分原因是闭源和开源之间的竞争是真实存在的，开源确实起到了制衡作用。

A little bit of it is there's real competition between closed source and the open source check is real.

说实话，这是好事，因为没有这种竞争，我觉得我们不会有一个广泛的、前沿的生态系统，也不会有广泛的传播，否则我们又会回到大型机那种被锁定的状态。

And that's good, quite frankly, because without it, I don't think we're going to have a broad frontier ecosystem or broad diffusion, because otherwise we'll just be back to some mainframe lock-in.

那是不可能的。

That's just not a thing.

你说得对，如果要说的话，希望我们以后在每一层都能有更丰富的选择。

To your point about if anything, given that we will now hopefully continue to have a much richer choice in every layer.

所以对我来说，希望我们能开始构建这些AI，因为今天如果一个AI产品的所有收益都只归模型层，那对于想打造产品公司的人来说是没有意义的。

So to me, hopefully we can start building these AI because today the royalty of an AI product all going to just the model layer doesn't make sense if you really want to build a product company.

那绝对不行。

It just cannot be.

实际上，如果要说的话，这其实是一回事。

In fact, if anything, that's the same thing.

也就是说，如果拿数据库来说，如果没有开源对闭源的制衡，价格就不会降到人们能成功构建应用层并保持利润的地步。

Which is if you take the database, if there was no open source check on closed source, the prices wouldn't have been at a place where people could have built the app tier successfully and with a margin.

所以我觉得应用层在经济上会变得更加可行，这对整个生态系统来说是好事。

And so I think the apps are going to become much more viable economically, which is great for the ecosystem.

还会有各种中间层，比如，我的记忆系统是什么？

There are going to be all these other layers of middleware, call it, which is, hey, what's my memory system?

我的编排和协调层是什么？

What's my harness and orchestration layer?

所以那里会有一个非常丰富的工具生态系统。

So there's going to be a very rich tools ecosystem there.

模型公司会做得很好。

The model companies will do fine.

实际上，帕累托最优，他们可以根据自己的模型系列来管理令牌定价。

In fact, the Pareto, they can manage the token pricing based on their model family.

如果要说的话，我希望他们甚至能致力于KV这些标准——

If anything, I want them to work on even the KV, these standards-

标准。

Standards

……这样我们就能用多个模型家族了。

... such that we can use multiple model families.

其实这对他们更好。

In fact, it's better for them.

其实，我最早是做Windows和Unix互操作的。

In fact, I worked on Windows interop with Unix first.

这其实挺反直觉的。

In fact, it was counterintuitive.

我们以前觉得，天哪，做互操作会让我们用得少，结果反而用得更多了。

We used to think, oh my god, this interop means we'll be less used, except we were more used.

其实，说来也怪，因为那时候Unix版本太多了，Windows的互操作反而让Unix和Windows都变得更好了。

In fact, we became weirdly enough, because there were so many variants of Unix at that time, that Windows interop made Unix better and Windows better.

而且，我们之所以能打入企业市场，主要就是靠做互操作。

And in fact, we were able to penetrate the enterprise primarily because we did that interop work.

所以至少我是这么想的。

And so that's at least how I think about it.

萨提亚，我们现在处在一个很有意思的时刻，一方面，这些专家在呼吁监管、呼吁监督和治理。

Satya, we're in this interesting moment where on the one hand you have these experts asking for regulation, asking for oversight, governance.

这通常总会带来一些自由的限制。

It typically always leads to some restriction of freedom.

而普通社会大众现在被迫要表态，说这是对还是错。

And general society are put in a position where now we have to opine on whether this is right or wrong.

但另一方面，大多数人的实际体验并不是AI带来的那种神奇的生产力提升。

But then on the other side, most people's lived experience is not this magical productivity boost of AI.

最多也就是把Apple Watch的数据整合起来，告诉我们为什么睡得少了。

At best, it's integrating our Apple iWatch data to tell us why we're sleeping less.

这基本上就是大多数人对AI的体验标准了。

That's functionally the bar for most people.

或者就是为什么我家孩子老跟ChatGPT说些不靠谱的话。

Or why is my kid an asshole into ChatGPT?

那你能帮我们解释一下这个差距吗？

So can you just help us bridge this?

你看到了这么多企业应用。

You see so many enterprise applications.

神奇之处在哪里？

Where's the magic?

利润增长在哪里？

Where are the gains in profits?

AI带来的巨大突破在哪里，这些突破最终能让所有人理解这些紧张关系？

Where are the huge upside breakthroughs that AI is creating that will somehow make all of this tension understandable for everybody?

对，这个问题问得好。

Yeah, it's a great point.

我觉得这才是真正的问题，就是我们怎么才能在生产力数据里真正看到这些？

I think this is the real question, which is how do we truly see this in the productivity stats?

我们怎么才能在GDP增长中真正看到它，而且是广泛的增长，不仅仅是供应商——

How do we really see it in the GDP growth that's broad based, it's not just supplier-

……还是供给侧？

... or supply side?

我最喜欢的一个例子，也是我经常提到的，就是医疗，对吧？

The one example that I love and I get back to, in fact, healthcare is a good one, right?

想想医疗，哪怕只是简单的医患互动。

If you think about healthcare and even the simple doctor-patient interaction.

我们有个产品叫DAX Copilot。

In our case, we have this thing called DAX Copilot.

这是我能举出的最具体的例子：医生可以把更多时间花在照顾病人上，而不是花在录入电子病历系统上。

That's the place which is the most tangible example I can always point to when a doctor can spend more time with the patient caring for them versus just the entry into an EMR system.

这就是实实在在的生产力提升。

That's a good productivity gain.

如果它能帮医生处理收件箱，让他们能更快响应，这对病人和整个医疗系统都有帮助。

If it can triage the inbox for the doctor so that they can be more responsive, that's helpful for the patient and the care system.

行政方面，尤其是录入，因为涉及支付方、病人和医疗系统三方的协调。

The administrative, in fact, keying, because it's the triangulation of the payer, patient, and the health system.

实际上，医疗保健的大部分成本都花在了工作流程上。

In fact, most of healthcare is sort of all workflow cost.

所以，简化工作流程的复杂性，这是很有帮助的。

So taming of that workflow complexity, that's a helpful thing.

但是你在微软那边看到这种情况了吗——

But do you see that in Microsoft-

我们看到了。

We do

——在你帮助的那些人身上？

... with the people that you're helping enable?

是的，当然。

Yeah, absolutely.

我们确实看到了。

We see that.

而且，顺便说一句，即使在简单的Copilot案例中，对吧，如果你看看数量——大多数人想到的是工作，我认为确实会有岗位被替代。

And by the way, even in simple Copilot cases, right, which is if you look at the amount-- Most people think about jobs, which I think there is going to be displacement.

但归根结底，关键问题之一就是会创造出哪些新工作。

But the bottom line is, what are the new jobs that get created, is going to be one of the key aspects of it.

但不幸的是，很多知识工作其实也挺枯燥的，对吧？

But also a lot of knowledge work, unfortunately, is drudgery, right?

我早上起来就会想，天哪，我整天就是在处理邮件，对吧？

I get up in the morning and I think about, like, man, all I do is email triage, right?

如果这些工作流程不再占用你本可以花在其他事情上的时间呢？

What if even just these workflows that are taking away time from things that you could be spending time on-

好的。

Okay.

嗯，你提到了一个很好的点。

Well, you're bringing up this great point.

如果追溯到上个世纪之交，工业革命时期，那时候是每周工作七天，很多人忘了我们为什么会有周末。

If you go all the way back to the turn of the century, the Industrial Revolution, when we had a seven-day workweek, a lot of people forget why did we introduce the weekends.

那是为了调和不同宗教群体之间的紧张关系，因为他们不得不在同一个工厂里工作。

It was to sort of manage the tension between different religious groups that had to work in the same factory.

然后你看长期GDP，除了某些外部事件，基本上就在200到400个基点之间。

And then when you look at long-run GDP, outside of some exogenous events, it sort of is between 200 and 400 basis points.

所以当生产力提升到来时，人类的工作就会退后一步，你最终完成的工作量差不多。

And so what happens is as productivity boosts come in, human work steps back, and you kind of accomplish the same amount of work.

你觉得这里会发生这种情况吗？

Do you think that that happens here?

有没有风险，我们——

Is there a risk that we-

这个问题问得好。

It's a great one

……我们每周工作三天，但GDP还是以2.5%的速度增长？

... we have a three-day workweek, and we're just still growing at 2.5%?

对，这是个好问题——还是我们会找到新的事情做？

Yeah, that's a great-- Or will we find new things?

而这就是我对AI真正影响的兴奋点所在，不只是想着它怎么帮我改进了某个工作流程或简化了现在的事情，而是它能不能发明新东西。

And this is where the excitement, at least I have, for what the real impact of AI would be is instead of just thinking about how it has helped me augment some workflow or simplify something that's happening today, is it inventing new things?

它是不是在加速药物研发？

Is it speeding up drug discovery?

它是不是在，我不知道，再回到我那个例子，比如小企业的营运资金管理变得高效多了——

Is it taking the, I don't know, let's again go back to my example of, okay, the working capital management of a small business has become so much more efficient-

……以至于突然之间，不再只是我有没有ERP或者QuickBooks之类的系统，而是我真的能根据查看发票、邮件等能力来做决策，从而优化我的营运资金。

... that suddenly it's no longer just, do I have an ERP or a QuickBooks-like thing, but I truly am making decisions based on the ability to introspect my invoices, my emails, and what have you, and somehow optimize my working capital.

这是以前不存在的生产力。

That's productivity that didn't exist.

所以我确实希望我们会开始看到GDP增长，就像工业时代那样——

And so I do hope that we will start seeing GDP growth, which we did see in the industrial era-

……在它的第一阶段。

... during the first phase of it.

所以我觉得这就是需要的，对吧？

So that I think is what is needed, right?

也就是说，为了让这一切实现，坦白说，我们确实需要看到至少7%、8%的真实且广泛的GDP增长。

Which is in order for all of this to play out, quite frankly, we do need to see at least 7%, 8% GDP growth that is real and that's broad based.

所以——

So-

萨提亚，微软在AI方面做的是什么业务？

Satya, what business is Microsoft in, in relation to AI?

显然，Azure表现非常强劲。

Obviously, Azure has been crushing it.

你在拒绝客户，而且你正在进行1750亿美元的资本支出建设。

You're turning away customers, and you're doing $175 billion in CapEx buildout.

但你的资本支出远低于Meta和Google。

But your CapEx is far below what Meta's doing, far below what Google's doing.

他们正在进行二次融资和举债，达到3500亿美元。

They're doing secondary raises and raising debt, $350 billion.

前沿实验室在花费5000亿美元。

The frontier labs are spending $500 billion.

你凭借对OpenAI的先见之明投资，早早进入了这个领域。

You were so early to the party with the prescient OpenAI investment.

但是后来Copilot好像并没有真正成功，我觉得。

But then Copilot didn't exactly land, I don't think.

它的评价也不是很好。

It didn't get great reviews.

你们没有前沿模型。

You don't have a frontier model.

那业务是什么——

What's the business-

请回来。

Please come back.

我知道，但这里的业务是什么？

I know, but what's the business here?

你们需要拥有前沿模型吗？

Do you need to have a frontier model?

我们有没有告诉过你，小组里有一位记者？

Did we tell you there was one journalist on the panel?

不不不，我是真心这么说的，因为我就是好奇。

No, no, no. I mean it sincerely because I'm just curious.

你是个很厉害的战略家。

You're a great strategist.

这一点我们都清楚。

We know that about you.

微软错过了移动革命。

Microsoft missed the mobile revolution.

微软会不会也错过AI革命——

Is Microsoft going to miss the AI revolution-

——如果没有前沿模型的话？

... if you don't have a frontier model?

因为我一直觉得奇怪，你们居然没有。

Because I always found it perplexing that you didn't.

说真的，那你们的战略到底是什么？

And what's the strategy there, in all seriousness?

你觉得开源会赢吗？

Do you think open source is going to win?

你应该有那套打法。

You should have that play.

那我跟你讲讲我们目前的情况，以及我们在这些方面都在做什么。

So let me walk you through the sort of where we are and what we're up to on each of these.

顺便说一句，在资本支出和建设方面，我们起步很早。

By the way, on the CapEx side and the build-out side, we started early.

所以如果你累计来看，我不是说现在谈大量资本支出是好事，这其实是个问题。

So if you sort of cumulatively look, I'm not saying right now speaking about a lot of CapEx is not a feature, it's a bug.

但话虽如此，如果你真的去算一下账，考虑到我们开始的时间，因为我们比很多人意识到需要建设这件事早了好几年就开始动手了，所以这是其中一个方面。

But that said, but if you really go actually add up the math, given when we started, because we started multiple years before people woke up to even actually needing to build, and so that's one aspect of it.

另一个方面是，我们在调整资本支出，不想只为了一两个客户去建设，对吧？

The other aspect of it is we are calibrating our CapEx in such a way that we don't want to build for one or two customers, right?

我们想为长尾市场建设，对吧？

So we want to build for the long tail, right?

因为我觉得那才是最重要的。

Because that's, I think, most important.

而且如果你是超大规模云服务商，你不会只给两家模型公司供货。

And if you're a hyperscaler, you're not a supplier to two model companies.

那根本算不上生意。

That's not a business.

你得建立一个对大量第三方都很好用的系统，同时我们自己的第一方产品也要好。在这种情况下，我们对目前的进展感到非常兴奋，就连Copilot也是，你看我们公布的订阅用户数就知道了。这其实又回到了Chamath说的那个核心观点，就是这些是真正的企业在用，用在真实的工作流程里。

You have to sort of basically build a system that is great for lots of third parties, and our own 1P In that context, we are pretty thrilled with the progress we are making, with even Copilot, if you look at the subscriber numbers we gave, which this goes back, in fact, to Chamath's fundamental point, which is these are real enterprises using it for real workflows.

而且我们现在已经有超过三千万用户了。

And the fact that we now have 30-plus million out of a forum.

别忘了，整个知识工作者的基数有多大。

Remember, the total knowledge worker base, right?

大家常说互联网上有三十亿人、四十亿人。

Most people talk about three billion people, four billion people on the internet.

Office 365或者Microsoft 365就是知识工作的标准。

Office 365 or Microsoft 365 is the standard when it comes to knowledge work.

全球有4.5亿人，这包括了所有学生。

There's 450 million, that's including all students in the world.

哦，哇。

Oh, wow.

所以当我们谈论市场，所谓的定义，可能只有3亿，甚至2.5亿真正的企业用户。

So when we talk, the market, quote unquote, as defined, is maybe 300, 250 even of real enterprise users.

而在这其中，我们的渗透率已经接近3000万，而且还在增长，等等。

And of that, we've got the penetration of close to 30 million on that, and it's growing, and so on.

在模型方面，我们当然对投资OpenAI感到非常兴奋，我们能够使用他们的知识产权，这已经持续很长时间了。

The aspect on the model side is we're thrilled about, obviously, our investment in OpenAI, the access we have to their IP, which we have for a long time.

我们会利用这一点。

We're going to use that.

但我们也在积极开发自己的MAI模型，对吧？

But we are well on our way building our MAI models, right?

如果你看看这个——

If you look at it-

……我们有一个闪速网络模型，通过我们的协调框架来调度其他模型，在Cyber Gym上的表现甚至超过了Mythos。

... we have a flash cyber model that, with our harness, orchestrating other models, outperforms, on Cyber Gym, even a Mythos.

在编程方面我们也看到了同样的情况。

Same thing we're seeing in coding.

在知识工作领域也是如此，对吧？

Same thing we're seeing in knowledge work, right?

所以我们的目标基本上是从底层开始逐步提升，顺便说一句，我们并没有做任何蒸馏。

So our goal is to basically hill climb from the bottom, by the way, not distilling anything.

所以从最底层开始，利用我们的RLEs和数据，同时在企业方面保持差异化定位，回到解决他们的一些需求，比如“嘿，我能拿到权重吗？”

So from the very bottom, using our RLEs, our data, and then also have a differentiated position with enterprises going back to addressing some of the things that they want, which is, "Hey, can I have the weights?

“我能拿到权重，然后添加到我的知识库中吗？”

Can I have the weights that I can then add to my knowledge?"

这些就是我们基于基础模型要做的事情。

These are the things that we will do with our foundation.

所以，你给企业的最佳建议，我认为，就是AI主权很重要。

So your best advice, I think, to enterprises is AI sovereignty is important.

把你的数据放进前沿模型里，可能不是个好主意，然后你会成为他们的辅助工具，帮他们——

Putting your data into a frontier model, probably not a good idea, and then you're going to be that harness for them to help them-

……实现这一点。

... implement that.

我的建议是，多用各种模型，但不要依赖任何一家。

My advice is more use all, but be independent of all.

比如说，我的检验标准是，你永远要评估那些对你最重要的指标，对吧？

So for example, my acid test is you should always eval max evals that matter to you, right?

那么你想要的结果是什么？

So what's the outcome you want?

你应该用所有模型去跑那个结果。

You should go run that outcome through all the models.

然后，我会做这样一个测试。

Then here's the test I would do.

我会拿掉一个模型，看看我能不能保持那个评估结果。

I would pull out a model and see whether I can retain the eval.