OpenAI总裁格雷格·布罗克曼：AI自我改进、超级应用押注、通往AGI之路、算力扩展

我非常期待看到它如何提升上限。

I'm super excited to see how it raises the ceiling.

对吧？

Right?

我们已经看到了这些在物理应用等方面的表现。

We've already seen these physics applications, things like that.

我认为我们将能够解决更多开放式问题，应对更长的时间跨度。

And I think we will be able to just solve, like, way more open ended problems, way longer time horizons.

同时我也非常期待看到它如何提升下限，让你做任何事情时都变得有用得多。

And then also very excited to see how it raises the floor, where just for anything you wanna do, it's just so much use more useful for you.

对于普通用户来说，真正感受到这种变化可能有点困难。

It can be kinda tough for everyday users to really feel the change.

比如，在GPT-5发布之前，大家有很多铺垫和期待，但当它真正发布时，

Like, there was a talk about a lot of buildup before GPT five came out, and then it came out.

公众最初的反应有些失望，但后来人们意识到，它在某些任务上确实非常出色。

And, actually, the initial reaction was somewhat disappointing among the public, but then I think people realized that for certain tasks, was really good.

或者在下一波模型中，你认为这种变化会在某些职业的基层工作中明显体现出来，还是会对每个人都有普遍而切实的提升？

Or with these next series of models, do you expect that it'll really be felt sort of in the trenches in certain occupations, or do you think it will be a broadly tangible improvement for everyone?

我认为情况会类似：当你发布它时，会有一些人尝试后觉得，这和我以前见过的任何东西都截然不同。

I think that it will be a similar story where you when you release it, there will be people who will try it and be like, this is a night and day different from anything I've seen.

是的。

Mhmm.

然后也会有一些应用场景，我们原本并不是被智能水平所限制的。

And then there will be some applications where we weren't necessarily intelligence bottlenecked.

所以如果你有一个更智能的模型，也许在那些地方你并不会立刻感受到变化。

And so if you have a model that's more intelligent, maybe you won't feel it right there.

但我认为随着时间推移，你会感受到它，因为真正发生变化的是你对系统的依赖程度。

But I think over time that you will feel it because the fundamental thing that shifts is how much do you rely on the system?

比如，想想我们所有人如何与AI互动，因为我们对它能做什么都有某种心理预期。

Like, if you think about the way we all interact with AI, because we have some mental model for what we think it can do.

而这种心理预期实际上转变得相当缓慢。

And that that mental model shifts actually fairly slowly.

对吧？

Right?

随着你经验的积累，它会为你带来神奇的变化。

As you get more experience, it does something magical for you.

你会说：哇，真的吗？

You're like, oh, wow.

它还能做到这个？

It can do that?

我从来没想过会这样。

I never imagined that.

比如在获取健康信息的应用中，我们就看到了这种情况。

And we see this, for example, in applications like access to health information.

对吧？

Right?

我们看到有些人，比如我有个朋友，他用ChatGPT来了解自己癌症的不同治疗方案，而医生曾告诉他病情已到晚期，无药可救。

That we see people who you know, like, I I have a friend who used Chatuchi BT to understand different treatments for his cancer and that he was told by doctors that he was terminal, that there was nothing they could do for him.

他使用Chatuchi BT实际研究了许多不同的想法，并通过这种方式获得了治疗。

He used Chatuchi BT to actually research a bunch of different ideas, and he was able to get treatment that way.

在这样的应用中，你需要对AI能提供帮助抱有一定的信心，才能真正投入精力从机器中获得成果。

And that's something where you need to have some level of belief that the AI is gonna be helpful in that application for you to really put in the effort to get something out of the machine.

我认为，对于任何类似的应用，AI能够帮助你这一点将变得对每个人来说都更加明显。

And think what we're going to see is that for any application like that, it's going to become so much more evident to everyone that the AI can help you.

所以我认为，这既是技术本身在进步，也是我们对技术的理解在转变并逐渐跟上它的步伐。

And so I think it's a little bit of the technology getting better, but it's also our understanding of the technology shifting and and catching up to that.

你将来在OpenAI内部会更多地依赖它。

And you'll be relying on it more inside OpenAI.

你们正在开发一个自动化的AI研究员。

You have, an automated AI researcher in the works.

它预计在今年秋天推出。

It's supposed to come out this fall.

那是什么？

What is that?

所以目前的发展方向是，我们正处于这项技术的早期起飞阶段。

So the the direction of travel right now, we are in this early phase of takeoff of this technology.

起飞意味着什么？

What does takeoff mean?

起飞是指AI变得越来越好，呈指数级进步，部分原因是我们可以利用AI来让AI变得更好。

Takeoff is as the AI gets better and better on this exponential and in part because we can use the AI to make the AI better.

因此，我们的开发进程加快了。

So our development process speeds up.

但我认为，当我想到起飞时，它也关乎现实世界的影响。

But I also think when I think of takeoff, it's also about real world impact.

在某种程度上，每项技术都是一条S曲线。

And in some ways, we've been you know, every technology is is an s curve.

或者如果你放大来看，一些S曲线最终会变成指数曲线。

Or if you zoom out some of s curves, that end up being an exponential.

我认为，我们现在正遭遇的就是这种情况。

And I think that that's what we're encountering right now.

因此，技术发展正以越来越快的速度推进，这台引擎正在积累更多动力。

So it's the tech technology development is moving with increasing speed, and it's this engine that's picking up momentum.

但与此同时，世界上还有诸多顺风因素，因为芯片开发者正在为其项目投入更多资源。

But it's also in the world, there's all of these tailwinds because there's chip developers that are getting more resourcing into their programs.

还有一群人正在此基础上构建，试图弄清楚它如何融入各种不同的应用场景。

There's this economy of people who are building on top of it, trying to figure out how it fits into every different application.

所有这些能量正不断累积，使人工智能从一个旁观者转变为推动经济增长的主要驱动力。

And all of that energy is just accumulating more and more into this this takeoff phase of the AI becoming just a kinda sideshow to being the main driver of economic growth.

我认为，这不仅仅关乎我们在这堵墙内所做的事情。

And I think that that is something that it's not just about what we're doing in these walls.

它关乎整个世界、整个经济如何协同合作，共同推动这项技术及其应用的发展。

It's about how the whole world, the whole economy comes together in order to to push forward this technology and its usefulness together.

而研究人员

And the researcher

那么它具体会做些什么呢？

will then what will it do exactly?

所以，研究人员将处于一个关键时刻，即我们正在构建的AI现在正在承担越来越多的任务，我们应该能够让它自主运行。

Well, so the researcher will be a moment where the AI which we're building that right now it's taking you know, it's doing a larger percentage of tasks, that we should be able to let it run autonomously.

我认为，这背后需要深入思考，这并不意味着我们只是放任它独自运行，然后过段时间再回来查看它是否做了些有用的事。

And that I think there's a lot of thought that goes into what that means and that it doesn't necessarily mean that we just let it off on its own and then come back later and see if it did something good.

我认为我们将深度参与对它的管理。

I think that we are gonna be very involved in managing it.

对吧？

Right?

就像现在，如果你让一名初级研究人员独自工作太久，他们很可能会走上一条没什么用的路。

Just like right now, if you have a junior researcher, if you leave them on their own too long, they're probably gonna go down a path that's not very useful.

但如果你有一位资深研究人员，或者有明确愿景的人，他们甚至不需要掌握具体的技术细节，就能提供反馈，审阅这位——你知道的，实习生——所生成的图表，并从愿景层面给予指导：我希望你实现什么目标。

But if you have a senior researcher or someone who has a vision, they don't even necessarily need to know the mechanical skills, they will be able to provide feedback, review the plots that this person's that the, you know, the intern's producing and to provide direction in terms of the vision of what is it that I want you to accomplish.

因此，我认为我们将构建一个系统，它将极大加速我们开发模型、实现新研究突破的能力，让这些模型在现实世界中变得更实用、更易用。

And so I think of this as a system that we're going to build that will massively accelerate our ability to produce models to make new research breakthroughs happen, to be able to make these models more useful and usable in the real world

对。

Right.

而且要以越来越快的速度做到这一点。

And to do that at increasing speed.

抱歉。

So sorry.

它会做什么？

What's it gonna do?

你会说‘去找到AGI’，然后它就会尝试去——我觉得我觉得——

Are you gonna say go find, AGI, and it will just try to I I think I think

在我看来，第一层面上就是这样。

the way I think of it is is something like that to first order.

好的。

Okay.

在实际层面，我认为——

And at a practical level, I think

我会把它看作是将我们一位科研人员的全流程工作完全在硅基系统中实现。

I would view it as is taking the full end to end of what one of our research scientists does and be able to do that in silicon.

另一种思考起飞的方式是，人工智能的进步从渐进式转变为加速积累，最终形成一种不可阻挡的进程，朝着超越人类智能的方向发展。

Another way to think about takeoff is their progress in AI goes from incremental to gathering momentum and then sort of this unstoppable march to an intelligence that's smarter than humans.

你是否担心，正如这一领域有可能顺利发展，也同样存在这种进程出错的可能性？

Do you do you worry that there's just as there's possibilities for things to go right on that front, there's also possibilities for that progress that process to go wrong?

我的意思是，我

I mean, I

我认为这绝对是肯定的。

think that that's absolutely yes.

我认为，要获得这项技术的好处，就必须认真考虑其风险。

I think that the way to get the benefits of this technology is also to really think about the risks.

如果你看看我们从技术角度如何推进技术发展，我们会投入大量资源在安全和防护上。

And if you look at how we've approached technology development from a technical perspective, we invest a lot in safety, security.

一个很好的例子就是提示注入攻击。

Good example of this is prompt injections.

对吧？

Right?

如果你要拥有一个非常聪明、能力很强并连接了大量工具的AI，你就必须确保它不会因为有人给出奇怪的指令而被操控。

If you're going to have an AI that is very smart, very capable, hooked up to lots of tools, you wanna make sure that it can't be subverted by someone giving it a weird instruction.

在这方面我们投入了大量资源，我认为已经取得了非常出色的结果，并且有一支极其优秀的团队在负责这项工作。

And that's something that we've invested in quite a lot and I think have really incredible results, have an incredible team working on.

思考这些难题时，将它们与人类的情况进行类比会很有趣。

And it's interesting to think about some of these problems where you can make analogies to humans.

比如，人类也容易受到钓鱼攻击，容易被以各种方式欺骗，难以完全理解自己所从事工作的全部背景。

Like, humans are also susceptible to phishing attacks, to being deceived in different ways, to not really understanding the full context of what they're working on.

我们在开发过程中引入了这些类比，每次发布或开发模型时都会思考：如何确保它与人类保持一致，并真正具有帮助性？

And we bring those analogies into our development process and think about this whenever we release a model, develop a model, how do we ensure that it's going to be aligned with people and be able to actually be helpful?

这一点是我们非常重视的。

And, that is something that we care quite a lot about.

我认为还有一些更宏大的问题，比如世界、经济将如何变化，以及每个人如何从这项技术中受益。

I think that there are bigger questions about the world, the economy, how does everything change, how does everyone benefit from this technology.

这些问题并非纯粹的技术问题，也不是单靠OpenAI就能解决的。

They're not purely technical, not purely something that OpenAI on our own will be able to solve.

但确实，我不仅在思考如何推动技术进步，更在认真考虑如何确保我们能实现其潜在的积极影响。

But yes, I think quite a lot about not just pushing forward the technology, but also really about how do we ensure that we have the positive impact that is its potential.

不过，令人担忧的是，这是一场竞赛。

The worry, though, is that this is a race.

在OpenAI总部内所进行的工作，正被许多开源参与者效仿，而他们在安全方面的边界、限制和保护措施要少得多。

And what's being done within these walls for OpenAI headquarters is also being copied by many of the open source players, which have much less bare boundaries and barriers and protection on on on the safety side of things.

我记得你曾经说过，要创造需要很多人把很多事情做对，而一个心怀恶意的人却足以造成破坏。

And I think you've said this once that it takes, you know, people getting a lot of things right to be creative and sort of one person with bad intent to be destructive.

这正是我担忧的焦点——因为这显然是一场竞赛。

And that's sort of where the concern lies for me at least is just when this this is it's clearly a race.

进展非常迅速。

It's going fast.

你的许多同行都表示，如果所有人都同意暂停，我们就会停下，但目前看来，局势丝毫没有放缓的迹象。

Many of your counterparts have said, if everybody agrees to stop it, we'll stop it, and, and it doesn't seem like it's gonna slow at all.

所以，归根结底，回报值得冒这个风险吗？

So is the reward worth the risk, basically?

我认为回报值得冒险，但我觉得这种说法在某种程度上太过笼统了。

I think that the reward is worth the risk, but I think that that is too too coarse grained of a of a answer in some sense.

好的。

K.

在我看来，自从OpenAI成立之初，我们就一直在思考：一个美好的未来是什么样的？

Like, the way that I think about it is that we've asked from the beginning of OpenAI, how what does a great future look like?

这项技术如何才能真正提升每个人的生活？

How can this technology really be something that uplifts everyone?

你可以从两个不同的角度来思考这个问题。

And you can think of there almost being two different angles.

一个是集中化视角，认为让单一主体来开发这项技术才是确保其安全的方法，这样就不会有任何压力。

One is the centralization view of saying that, well, the way to make this technology safe is that you have only one actor building it, and so then you don't have any pressures.

对吧？

Right?

你可以专心致志地把它做对，然后等准备好了再向所有人推广，诸如此类的事情。

You can really think about getting it right and, you know, then figure out how to roll it out to everyone when it's ready, those kinds of things.

在某些方面，这确实很难接受，但我认为我们可以换一种方式来思考许多特性，这就是我们所说的韧性——将它视为一个有许多参与者共同开发技术的开放系统。

That's a pretty tough pill in some ways, and I think that there's a lot of properties that you can think instead think about approaching differently, which we refer to as resilience, to think of it as this open system where there's lots of players who are developing the technology.

但这不仅仅关乎技术本身。

But it's not not just about the technology.

它还关乎构建能够帮助这项技术良好发展的社会基础设施。

It's about building societal infrastructure that helps this technology really go well.

如果你想想电力的发展历程，就会发现有很多人生产电力，它确实存在危险和风险。

And if you think about how electricity has developed, that's something where lots of people produce it, that it actually has dangers and risks.

但我们围绕电力的安全标准、不同的利用方式、如何规模化以及在大规模应用时的监管机制，建立了一整套多样化的安全基础设施，使许多人能够以民主化的方式使用电力。

But we also build our safety infrastructure in a diversity of different ways around safety standards for electricity, around different ways of harnessing it, about how you scale it, that there's regulations when you're at these massive scales that lots of people are able to use it in a democratized fashion.

还有检查员。

There's inspectors.

实际上，围绕这项技术的需求和特性，已经建立了一整套体系。

Like, there's a whole system that's been built around the needs of that technology, the proclivities of that specific technology.

我认为，我们在人工智能领域真正看到的一点是，我们需要一场广泛的对话。

And I think that one thing that we have really, I think, seen with AI is that it is something where we need this broad conversation.

我们需要让更多人意识到这一点。

We need lots of people to be aware.

如果这项技术即将改变每个人的一切，人们就需要参与其中。

If this technology is going to come and change everything for everyone, people need to participate in that.

它不能总是由一个集中的团体秘密地主导。

It can't be something that's done often secret by just one, you know, sort of centralized group.

因此，这对我来说一直是关于这项技术应该如何发展的核心问题。

And so this has been, to me, a very core question to how this technology should play out.

我们真正信奉的是，在这项技术的发展过程中应涌现出一个具有韧性的生态系统。

And something we really believe in is this resilience ecosystem that should emerge around the development of this technology.

所以你说我们现在正处于起飞阶段，整个人类都在经历这一过程。

So you said we're in takeoff, in the middle of a takeoff process, and we, I guess, all of humanity are experiencing this.

英伟达首席执行官黄仁勋最近表示，他认为通用人工智能已经实现。

NVIDIA's CEO Jensen Huang said recently that he believes AGI has been achieved.

你同意吗？

Do you agree?

我认为AGI对很多人来说有不同的定义，很多人会说我们现在拥有的就是AGI。

I think that AGI has a different definition to many people, and I think that there are many people who would say that what we have right now is AGI.

我认为这可以争论。

I think you can debate it.

但我觉得有趣的是，我们目前的AGI技术其实非常不稳定。

But I think that maybe the thing that's interesting is that AGI, like, the technology we have right now, is very jagged.

它在许多任务上绝对超越了人类。

Like, it is absolutely superhuman at many tasks.

在写代码这类事情上，AI完全可以胜任。

When it comes to writing code, those kinds of things, AI can just do it.

对吧？

Right?

它大大减少了创造事物的障碍。

And it really removes a lot of the friction to creating things.

但有些人类能轻松完成的基本任务，我们的AI仍然难以应对。

But there's some very basic tasks that a human can do that our AI still struggle with.

所以这几乎是在问，你在哪里划这条界线？

And so it's almost to say that where do you draw the cut line?

嗯哼。

Mhmm.

目前，这更像是一种氛围或感觉，而不是科学。

It's a little bit more of a vibe than a feeling than it is a than it is science at the moment.

所以我认为，对我们来说，我们正处在这样一个时刻。

And so I think for myself, we're definitely going through that moment.

如果你五年前给我看我们现在拥有的系统，我会说，是的。

And if you were to show me five years ago the systems we have today, I'd go, yeah.

这就是我们所谈论的。

That's what we're talking about.

但它就是不一样。

But it's just different.

它和我们曾经想象的任何东西都截然不同。

It's so different from anything we ever pictured.

所以我认为我们需要相应地调整我们的思维模式。

And so I think we need to adjust our mental models appropriately.

那你还没到那儿吗？

So you're not there yet?

我觉得

I think

我觉得我基本上已经完成了80%。

I think that I'd say I'm I'm I'm basically, like, 80% there.

所以我认为我们已经非常接近了。

So I think we're we're quite close.

我认为毫无疑问，未来几年内我们一定会实现通用人工智能，尽管它仍然会有些不完善，但任何涉及使用电脑的智力任务，其最低门槛都将被彻底突破。

I think it's extremely clear that we are going to have AGI within the next couple years in a way that is still gonna be jagged, but that the the floor of task will just be almost for any intellectual task of how you use your computer.

人工智能将能够完成这些任务。

The AI will be able to do that.

我认为，目前我必须给出一个略显不确定的回答。

And I think that that, yeah, right now, I have to give a little bit of a uncertain answer K.

因为这里存在某种不确定性，有点像不确定性原理，你可以对此进行争论。

Because there's some there's some it's almost like a like a uncertainty principle kind of thing that you can you can you can debate it.

根据我自己的定义，我认为我们几乎已经达到了。

From my own personal definition, I think we're almost there.

再加一点点，我们肯定就能做到了。

And with maybe a little bit more, we will absolutely be.

好的。

Okay.

好吧，我们得去休息一下了。

Well, we're we gotta go to a break.

但只要

But as long

在去休息的路上，我想让在家观看的观众知道，你和我将于6月18日在这里的旧金山爵士乐俱乐部再次对话。

as we're on the way to the break, I wanna let folks watching at home know that you and I are gonna be talking again June 18 here in San Francisco at SF Jazz.

所以我会在节目说明中放一些信息，如果你想要来参加这场对话的话，我真心希望你能报名。

So I will put some information if you wanna come join that conversation in the show notes, and I do hope you sign up.

好的。

Alright.

我们马上回来。

We'll be back right after this.

我在这档节目中采访过许多优秀的科技创始人，一个出人意料却反复出现的普遍挑战就是找到合适的域名。

I've interviewed a lot of great tech founders on this show, and one surprisingly universal challenge comes up again and again, finding the right domain name.

当我推出《大科技》时，自己也遇到过这个问题。

It's something I ran into myself when launching Big Technology.

你想要的名字往往已经被注册了，很容易就想随便选一个凑合了事。

The names you want are often taken, and it's tempting just to settle and move on.

但我最敬重的那些创始人，从不在根本问题上妥协，而域名正是其中之一。

But the founders I respect most don't settle on fundamentals, and your name is one of them.

它应该能立刻传达出你实际打造的产品是什么。

It should immediately signal what you actually build.

这正是我欣赏 .tech 域名的地方。

That's what I appreciate about dot tech domain names.

这很合理。

It just makes sense.

它向世界、你的客户、你的投资者，以及任何搜索你的人清晰地传达了你正在从事科技领域，简洁直接，无需修饰。

It tells the world, your customers, your investors, anyone googling you that you're building in technology, clean, direct, no qualifiers.

我看到越来越多认真的初创公司开始采用这种方式。

And I'm seeing more serious startups lean into it.

Nothing.tech、1x.tech、aurora.tech、ces.tech、ultra.tech、alice.tech、neon.tech、blaze.tech、pie.tech，还有更多。

Nothing.tech, 1x.tech, aurora.tech, ces.tech, ultra.tech, alice.tech, neon.tech, blaze.tech, pie.tech, and so many more.

如果你正在打造一个以科技为核心的产品，别将就。

If you're building something tech first, don't settle.

从你选择的任何注册商那里注册一个.tech域名，从第一天起就明确你的定位。

Secure your dot tech domain from any registrar of your choice and make your positioning obvious from day one.

我们回到《大科技》播客，今天邀请到OpenAI联合创始人兼总裁格雷格·布罗克曼。

And we're back here on Big Technology podcast with OpenAI cofounder and president Greg Brockman.

格雷格，

Greg,

让我

let me

我想问你2025年12月发生了什么，因为那时似乎成了一个转折点，之前让机器不间断编码数小时还只是理论，而那一刻所有人都开始觉得，我想我可以信任它持续运行一段时间了。

just ask you what happened in December 2025 because it seems like it was an inflection point where all this idea of letting the machine code for hours uninterrupted went went from theory to a moment where everyone said, I think I can trust this to keep going for a while.

那么到底发生了什么？

So what exactly happened?

新的模型发布让AI完成任务的能力从大约20%跃升到了80%。

So new model releases really went from the AI being able to do, like, 20% of your tasks to, like, 80%.

这是一个巨大的转变，因为AI从‘做起来挺不错’变成了你必须彻底围绕这些AI重新设计你的工作流程。

And that was this massive shift because it went from being kind of a, yeah, it's a nice thing to do to you absolutely need to retool your workflow around these AIs.

对我个人而言，我确实经历了一个这样的时刻：我有一个用了多年的测试提示词——‘帮我建一个网站’。

And for myself, I very much had this moment where I have a test prompt that I've been using for years of build a website for me.

我当初学编程时真的亲手建过这个网站。

I'd actually built this website back when I was learning to code.

花了我好几个月。

Took me months.

以前在二十五天里，你需要花四个小时，用很多不同的提示词才能搞定。

Used to be over the course of twenty five that, you know, it would take, like, four hours, a bunch of different prompts to get it right.

到了十二月，一次就能完成。

In December, one shot.

我只向AI问了一次，它就生成了，而且做得非常好。

Just asked the AI one time and it produced it and did a great job.

那么这些模型是如何实现飞跃的呢？

So how did those models make the leap?

很大程度上是因为更好的基础模型。

Well, a lot of it is about the better base models.

嗯。

Mhmm.

OpenAI的一个特点是，我们长期以来一直在改进预训练技术，而在那一刻，我们稍微窥见了今年余下时间将要到来的变化。

That one thing about OpenAI is that we've been working on improving our pretrained technology for quite some time and that in that moment, we got to see a little taste of what is going to be coming for for the rest of this year.

但这也真的不只关乎某一个方面。

But it's also really about not any one thing.

这是我们不断在每一个创新维度上持续推动的结果。

It's about we're constantly pushing on every single axis of innovation.

这些模型非常有趣的一点是，有时你会看到这种飞跃。

And the thing that's very interesting about these models is in some ways you get these leaps.

在某些方面，这一切都是连续的。

In some ways, it's all continuous.

对吧？

Right?

它并不是从0%直接跳到80%。

It didn't go from 0% to 80%.

而是从20%提升到了80%。

Went from 20% to 80%.

所以在某种程度上，它只是变得更好了。

And so in some ways, it's it just got better.

我认为，我们实际上在每一次版本更新中都看到了这种持续的改进，比如在5月2日和2023年5月之间，我一位密切合作的工程师从完全无法让AI完成他所做的底层硬件系统工程，到AI现在能完美地生成它。

And I think that we've actually seen this improvement continue with every single point release that we've had, Like, between 05/02 and five '3, one of my engineers I work with very closely went from he couldn't get it to do the, like, low level hardware systems engineering he does to it absolutely being created.

他给它一份设计文档。

He gives it a design doc.

它真的实现了它，添加了指标和可观测性，运行了性能分析器，并将其改进到完全达到他原本希望产出的效果。

It actually implements it, adds metrics, observability, runs the profiler, improves it to the point that it's the exact thing that he was hoping to produce.

所以我认为，看待这个问题的方式几乎是缓慢、缓慢、缓慢地突然一下，但这一切都由当前真正有效的东西所指示，肯定在一年内，有时甚至更早，就会变得极其可靠。

And so I think that the way to think about it is almost a sort of slowly, slowly, slowly all at once, but it is all indicated by what's kinda working right now, certainly within a year, sometimes much sooner, is going to be incredibly reliable.

这让你感到惊讶，因为我不久前听你在一次采访中提到，Codex——这个自主编码器——只是为软件开发者设计的。

And it surprised you because I heard you talking on an interview not long ago about how Codex, right, this autonomous coder, was just for software developers.

而在这次对话的早期，你说实际上每个人都能使用这些东西。

And earlier in this conversation, you said, actually, everyone can use this stuff.

是的。

Yes.

是什么让你改变了对它的看法？

What led to the fact that you sort of changed your perspective on it?

我想我之前一直专注于Codex，因为它里面包含代码，所以认为它只是给程序员用的。

Well, I think I'd been focusing on Codex, and it's got the code in it, right, as really being for coders.

考虑到OpenAI内部的许多人都是为自己开发软件的工程师，自然会这样想。

And thinking about people within OpenAI, because many of us are software engineers building for ourselves, it's very natural to think that way.

但随着这项技术的进步，我们开始意识到，我们所开发的核心技术其实根本不是关于代码的。

But as this technology has been progressing, we've started to realize that the underlying technology we produced is mostly not about code at all.

它主要关乎解决问题。

It's mostly about solving problems.

它主要关乎管理上下文、工具，并思考AI应该如何整合与完成工作。

It's mostly about being able to manage context and harnesses and think about how an AI should integrate and do work.

而这使得即使是对于代码而言，任何人都能获得访问权限，因为你能够管理一个将去执行任务的系统。

And that's something that becomes both even for code, suddenly, anyone can have access because you can manage something that's gonna go do work.

对吧？

Right?

如果你有一个愿景，想要完成某件事，你可以描述你的意图。

If you have a vision, you have something you want to accomplish, you can describe your intent.

AI可以执行，帮你实现它。

The AI can execute, can get that done.

但接着你就会开始想，为什么我只专注于编程呢？

But then it also starts to be like, why am I just focused on coding?

比如，Excel表格、演示文稿中涉及的大量技能都是非常机械的。

Like, there's so much just very mechanical skill associated with Excel spreadsheets, with presentations.

如果AI掌握了上下文，如今它已经具备足够的原始智能来出色地完成这些任务。

And if the AI has the context, it has the raw intelligence now to be able to do these things at a great level.

所以，如果我们能让它更易用，它就会从程序员专属的代码工具，变成每个人都能使用的工具。

So if we can just make it more accessible, suddenly it goes from codexes for coders to codexes for everyone.

就在我们看到这些进步之后不久，硅谷又出现了一个相当奇特的现象，那就是OpenAI——或许更广泛地说，是整个科技界，人们开始以你所提到的方式信任它：允许AI机器人访问他们的桌面，买一台Mac mini，赋予它访问邮件、日历和文件的权限，然后干脆让它接管自己的生活。

And soon after this moment where we saw all this improvement, there was another somewhat phenomenon in Silicon Valley, which was OpenAI, right, which is and maybe it's the broader tech community where where people started to trust it in ways that you suggested giving the an AI bot access to their their desktop or getting a Mac mini and giving it access to, like, their mail and calendar there and their files, and then just kinda letting it go run their life.

随后，OpenAI把创始人请进了公司。

And then OpenAI brought the founder of OpenAI in house.

所以你刚才更多地谈到了AI如何以某种方式帮你打理生活。

So you talked a little bit more about the AI as something that will help run your life for you in a way.

把OpenAI团队纳入公司内部，是否就是这一愿景的体现？

Is that the vision by bringing the OpenClaw team in house?

我认为这项技术的核心在于，弄清楚它如何有用、人们希望如何使用它、智能代理的愿景是什么，以及它将如何融入人们的生活，这是一个难题。

Well, I say that the core thing about this technology is that figuring out how it's useful, how people want to use it, what is the vision for agents, how is it gonna slot in people's lives, that is a hard problem.

在这一技术的多代发展中，我看到真正投入其中、充满好奇心和远见的人，具备一种真正的技能，而这正是这个新兴经济中正在崛起的极其宝贵的技能。

And that one thing I've seen across many generations of this technology is the people who really lean in, who have a lot of curiosity, who have a lot of vision, that's a real skill, and that's an emerging very valuable skill in this new economy that is emerging.

彼得，也就是OpenAI的创始人，我认为他拥有非凡的远见和惊人的创造力。

And Peter, who is the OpenAI founder, is, I think, someone who's got incredible vision, incredible creativity.

因此，在某种程度上，这关乎具体的技术，但在某种程度上，又完全不关乎技术。

And so to some extent, it's about the specific technology, but to some extent, it's not at all.

真正重要的是，我们该如何利用这些能力，找到它们如何融入人们的生活？

It's really about the how do we take these capabilities and figure out how those slotting people's lives?

因此，作为一名技术专家，这非常令人兴奋。

And so I think as a technologist, it's very exciting.

但作为一名专注于为人们带来实用价值的人，我们正在加倍投入并大力投资这一方向。

But as a someone who is focused on bringing utility to people, that's something that we are doubling down on and investing quite a lot.

你最近对这一点有一个相当有趣的说法。

You had a pretty interesting quote about this recently.

谈到让这些自主AI代理为你工作时。

Talking about getting these autonomous AI agents to work on your behalf.

你说，当你这么做的时候，你就成了一个拥有数十万代理的CEO，这些代理在帮你实现目标、愿景，而你不再深陷于具体问题的解决细节中。

You said, you become when you do it, you become this CEO of a fleet of hundreds of thousands of agents that are completing your objectives, your goals, your vision, and you're not in the weeds on exactly how different things are solved.

在某种程度上，这种新的工作方式会让你感觉失去了对问题的掌控。

And in some ways, this new way of work can make you feel like you're losing your pulse on the problem.

这好吗？

Is that good?

我觉得我觉得存在

I think I think that there's

利弊兼有。

a mixed bag.

因此，我认为我们需要承认这些工具所能带来的优势，并弥补其不足。

And so I think that what we need to do is acknowledge the strengths of what these tools can deliver and mitigate the weaknesses.

因此，要赋予人们杠杆力和自主权，让你拥有愿景、想要完成某事时，能够有一支代理团队去为你实现。

And so giving people leverage, agency, making it so that if you have a vision, something you wanna accomplish, that you can have a fleet of agents that will go do it for you.

但如果你想想这个世界是如何运作的，归根结底，总得有人承担责任。

But if you think about how the world works, that at the end of the day, there's an accountable party.

对吧？

Right?

如果你在建一个网站，而你的代理搞砸了，导致用户受到影响，这责任并不在代理身上。

If you're trying to build a website and your agent messes it up and your user is affected, it's not really the agent's fault.

而是在你。

It's your fault.

所以你必须在意。

And so you need to care.

我认为，要正确使用这些工具，人们必须意识到：人类的能动性与责任感，是整个系统的核心部分。

And I think that for people to use these tools right, you need to realize that human agency, human accountability, that's a core part of the system.

人类如何使用人工智能，这一点至关重要。

How the human uses the AI, that's something that is deeply fundamental.

因此，我认为作为这些代理的使用者——我们在OpenAI内部也是如此——绝不能推卸责任。

And so I think the important thing is that as a user of these agents, and we do this within OpenAI, you cannot abdicate responsibility.

你不能只是说，哦，AI会自己搞定一切。

You cannot just say, ah, the AI is just gonna do stuff.

当然。

Of course.

但你说你觉得自己正在失去对问题本身的把握。

But you said feel you're you're losing your pulse on the problem itself.

这和责任层面是不同的。

That's different than accountability layer to it.

对我来说，它们实际上是相互关联的。

Well, to me, they actually are are linked together.

因为关键是，如果你是一名CEO，却离细节太远，对吧？如果你在管理这家公司、这个团队，却失去了对局势的掌控，这不会带来好的结果。

Because the point is that if you if you're a CEO and you're too far from the details, right, if you're running this company, you're running this this team, and that you've lost your finger on the pulse, That is something that's not gonna lead to great results.

所以我刚才想表达的是，人类不需要了解一切情况，并不是一件值得追求的事。

And so the the point that I was trying to make there is that not that it's a desirable thing for humans to not have to to to know about what's going on.

有些细节你确实可以放心，比如你雇用总承包商来盖房子，有很多细节你可能不需要操心，因为你相信他们会处理好。

There are some details that because you you can trust like, if you are working with a team, like a general contractor to build a house, there's a bunch of details there that you probably don't need to worry about because you can trust that that they'll be taken care of.

但归根结底，如果有一些细节出错了，你还是应该关心。

But at the end of the day, if there are details that are wrong, you should care about it.

你应该保持警觉。

You should be aware.

所以我认为，你不能盲目地说，我接受自己失去对局势的把握，这是一个重要的细微差别。

And so this is, I think, an important nuance of you cannot just blindly say, I'm okay with losing my finger on the pulse.

我们需要主动介入，说：我必须保持对局势的把握，才能真正理解优势和劣势。

That we need to lean in and say, I need to keep it there to really understand the strengths and weaknesses.

当你逐渐脱离这些细节、这些低层次的机械性事务时，你应该是因为已经建立了对系统的信任，相信它能做好这些工作。

And that as you disengage from some of these details, these lower level mechanical things, you should do it because you have built trust with the system that it will do a good job.

关于模型，还有一个最后的问题。

One last question about the models.

你已经稍微谈到了模型所经历的演变。

You've talked a little bit about the evolutions that the models have gone through.

预训练和微调，强化学习让模型更能一步步解决问题，并上网去执行任务。

Pretraining and fine tuning, reinforcement learning that gets it, more equipped to solve problems step by step and go out in the on the Internet and do things.

而我们现在正处于这样一个阶段：模型已经通过这一过程学会了使用工具。

And now we're in this moment where the models have learned through that process to use tools.

如果我在这里有误解，请纠正我。

And correct me if I'm wrong on this one.

这一演进的下一步是什么？

What is next in that progression?

我认为，我们所处的世界是机器能力与深度不断增长的世界。

Well, I think that the world that we're in is one of this increasing capability and depth of what the machine can do.

这其中一部分涉及工具的使用，但我们现在也需要真正地构建出色的工具。

And some of this is about we've got this tool use, but now we also need to actually build really great tools.

你可以想象一下计算机使用场景，AI能够真正操作桌面，那么它就能完成你所能做的任何事情。

You think about something like computer use and AI that can actually use a desktop, then it is really able to do anything that you can do.

但我们也必须为机器考虑一下，在企业环境中，身份认证是如何运作的。

But we also have to build a little bit for the machine to think about how does, in the enterprise, credentialing work.

审计追踪和可观测性又是如何运作的？

How does how do audit trails and observability work?

因此，还有很多技术需要开发，以跟上核心模型能力的发展。

So there's a lot of technology to build to catch up with what the core model capability is.

我认为整体发展方向包括一个非常出色的语音界面。

And I think the overall direction of travel includes things like a really great speech interface.

你可以像进行这场对话一样自然地与电脑交谈，它能完全理解你。

So you can just talk to your computer naturally and just as natural as this conversation, and it understands you.

它能完成你所需的任务。

It does what you need.

它能提供良好的建议。

It has good advice.

它能够提醒我正在处理这件事。

It's able to surface that I've been working on this thing.

我遇到一个问题。

I have a problem.

当你早上醒来时，它会说：这是你的每日报告，显示你的代理昨晚取得了多少进展。

Here's you wake up in the morning, it says, here's your daily report of how much progress your agents made overnight.

也许它正在为你经营一家企业，我认为这将是这项技术的一个巨大应用场景。

Maybe it's running a business for you, which I think is gonna be a huge application of this technology.

创业的民主化正在必然到来。

The democratization of entrepreneurship is absolutely coming.

我会说，这些问题在这里。

And I'll say, here are these problems.

有一位客户很不满。

There's this customer that's upset.

你知道，他们想和真人交谈。

You know, they wanna talk to a real human.

你应该去和他们谈谈。

Like, you should go talk to them.

所有这些都会发生。

Like, all of that's gonna happen.

然后我认为，这项技术的下一步是提升人类所能解决的挑战的野心上限，我们正在见证它的前沿进展。

And then I think that the raising of the ceiling of ambition of challenges humanity can solve, that is also a next step for this technology, and we're seeing the leading edges of it.

但我最期待看到的是，如果你还记得AlphaGo的第37步，对吧，那步棋是人类根本想不到的，它很有创造力。

But the thing that I am just very excited to see is almost if you remember AlphaGo move 37, right, this move that no human ever would have come up with, it's creative.

富有创造力。

Creative.

它改变了人类对这盘棋的理解。

And it changed humanity's understanding of the game.

这种事将在每一个领域发生。

That is gonna happen in every single domain.

它会在科学、数学、物理、化学中发生。

It will happen in science, in math, in physics, in chemistry.

它会在材料科学中发生。

It's gonna happen in material science.

它会在生物学中发生。

It's gonna happen in biology.

它会在医疗保健和药物研发中发生。

It's gonna happen in health care, drug discovery.

但这也可能发生在文学、诗歌以及其他许多领域，这些领域将以前所未有的方式释放人类的创造力和想象力。

But it may also even happen in literature, in poetry, in a bunch of other fields that are going to unlock human creative understanding and ideation in ways we can't imagine right now.

既然你说模型已经如此强大，为什么你觉得这种现象还没有发生呢？

Why do you think that hasn't happened yet given how strong you say the models are?

我认为，模型的能力与人们使用它们的方式之间存在一定的滞后。

Well, I think that there is an overhang of what the models are capable of and how people are using them.

所以现在就在这里。

So the Now there.

嗯，这几乎是我们对这些模型内部内容的理解。

Well, yeah, it's it's almost our understanding of what is in these models.

好的。

Okay.

我认为这仍然是一个正在浮现的领域。

That's something that I think is still emerging.

所以我认为，即使没有进一步的进步，也仍会发生巨大的转变。

So I think that even with no further progress, there's still a massive shift that will happen.

由计算和人工智能驱动的经济仍然会发生。

The economy being powered by compute and AI is still going to happen.

但我认为，我们非常擅长的是在那些可以衡量的任务上训练模型。

But I think there's also something where what we've gotten very good at is training models on tasks that can be measured.

所以我们最初是从数学问题、编程问题开始的，这些都有完美的验证器。

And so what we started with was math problems, programming problems, where you have a perfect verifier.

而推动我们走向更开放型问题的进展，在很大程度上是扩展了可创造内容的范围。

And a lot of what the progress has been in bringing us to more open ended problems has been expanding the space of what can be created.

而人工智能本身真的能在这方面提供帮助。

And the AI itself can really help with that.

如果人工智能足够聪明并理解事物，你可以给它一个评估任务完成质量的评分标准。

If the AI is smart and understands things, you give it a rubric for how well a task goes.

嗯。

Mhmm.

当然，对于创意写作这类事情，比如，这是一首好诗吗？

And, of course, for things like creative writing, like, is this a good poem?

这要评分就难多了。

That's a much harder thing to grade.

因此，我们之前在教AI、让它去体验和尝试方面能力有限。

And so we've had less ability to teach the AI and for it to experience and try things out.

但这一切正在改变，我们对此有很强的预见性。

But all of that is changing and something that we we have a lot of sight for.

这很有趣。

Now it's interesting.

回过头来看，彼得·蒂尔曾提到，他很确定，如果你是数学型的人，相比文字型的人，面对这些模型时你可能处境更危险。

Reflecting on that, Peter Thiel has mentioned pretty sure that's what he said that if you're a math person, you're probably in deeper trouble in terms of these models coming from what you do than if you're a words person.

你以前可是数学俱乐部的成员。

And you were a member of math club back in the day.

你对此不担心吗？

Are you not concerned about that?

我觉得，我们更容易看到失去的东西，而不是获得的东西。

Well, I think that it's much easier to see what we lose than what we gain.

对吧？

Right?

因为我们对过去的做法有深刻的理解，我以前是这样做的。

Because we have a deep understanding of, I used to do things this way.

我以前参加过数学竞赛。

I used to do this math competition.

现在AI也能参加数学竞赛了。

Now that AI can do the math competition.

但那从来就不是真正的重点。

But it was never really about the math competition.

对吧？

Right?

对吧？

Right?

真正推动人类的并不是这个。

That's not really the thing that drives humanity.

如果你想想我们现在工作的方式——有个盒子，什么东西都被锁在盒子里，一百年前我们可不是这样做的。

And if you think about the way that we do work right now of there's a box, something tight behind a box, weren't doing that a hundred years ago.

这并不自然。

That's not natural.

这并不是我们所有人都被卷入的数字世界的真实面貌。

That's not this digital world that we all got kinda sucked into.

这并不是人类本质的真正含义。

That's not really what being human's about.

作为人类，意味着在此时此地，保持专注，与他人建立联系。

Being human's about being here, being present, connecting with other humans.

我认为我们将看到，人工智能将为我们腾出大量时间，以增进人与人之间的联系，建立更多人与人之间的纽带。

And I think that what we're gonna see is that AI is going to free up so much time to increase human connection, to build more bonds across people.

这正是我极其期待的事情。

And that's something I'm extremely excited about.

好的。

Okay.

当我们转向这些更具主动性的应用场景时，有人讨论是否真的还需要进行更大规模的训练。

And then as we shift well, as you shift, really, to, these more agentic use cases, there's been discussion about whether the bigger training runs really need to happen.

尤其是如果你已经把模型调教得足够好，就可以让它进入现实世界，从而在非预训练环节获得大部分性能提升，而这些非预训练环节正是大型数据中心所需要的。

And, you know, especially if you, like, get the model good enough, then you could sort of let it go out in the world, and then you can, effectively get much of the uplift in areas that aren't the pretraining, which is what these big data centers are needed for.

所以，你一直在参与扩展工作。

So you you worked, you work on scaling here.

你主导了这一过程。

Lead that lead that process.

你怎么看待这个观点？

What do you think about that argument?

我认为……

Well, I think

这个观点忽略了技术发展过程中的一个非常重要的方面，因为模型生产流水线的每一个环节都会产生倍增效应。

it misses something very important for how the technology development goes because it is absolutely the case that every single step of the model production pipeline multiplies.

所以我们希望改进每一个环节。

And so you wanna improve all of them.

我们所看到的是，我们验证了预训练的价值。

And the thing that we see is we prove the pretraining.

它让所有其他步骤都变得容易得多，这也很合理，因为这是一个能够更快学习的模型。

It makes all the other steps much easier, and it makes sense because it's a model that's able to learn faster.

这个模型之所以更快，是因为它在尝试不同想法并从自身错误中学习时，本身就具备更强的能力。

It's a model that is because it already is, like, more capable to start when it's trying out different ideas and learning from its own mistakes, that process just is faster.

它需要犯更少的错误。

It needs to make fewer mistakes.

所以我认为，关键的转变在于，我们不再仅仅认为这是在单独训练一个大脑系统，只是不断让它变得更大，而是也开始关注如何实际尝试各种应用。

And so I think that the big shift has been from thinking of it as just it's you're just training this cerebral system on its own and you just make it bigger and bigger to it's also about trying things out.

这还涉及理解人们在现实世界中如何使用它，并将这些反馈重新融入训练过程。

It's also about understanding how people are using it in the real world and connecting that back into your training.

但这并不会削弱持续进行这项研究的价值和重要性。

But it doesn't remove the value and the importance of continuing that that that research.

我认为另一个转变是，我们过去只关注原始的预训练能力，却很少考虑推理能力。

And the thing that I think has also shifted is we used to really just focus on the raw pretraining capability, but not think as much about the inference ability.

在过去的二十四个月里，我们意识到这是一个平衡：模型虽然在基础层面具备所有这些优秀特性，但你还需要它具备良好的推理能力，因为你需要进行强化学习。

And that's been a big change over the past twenty four months to realize that it's a balance between you can have this model that has all those great properties in the base, but then you really need it to be able to be inferenceable because you need reinforcement learning.

你需要将它推向世界。

You need to serve it to the world.

这意味着你不一定非要将模型做到尽可能大，因为你必须充分考虑后续的广泛应用，真正希望的是智能水平与成本的最佳结合，并同时优化这两者。

And that that means that you don't necessarily go as big as you possibly could because you also really think about there's gonna be all this downstream use, and you really want the thing that has the best intelligence times that that cost and to to optimize those two things together.

如果大部分工作都转向推理，你还需要NVIDIA的GPU吗？

Do you still need the NVIDIA GPU if things move mostly to inference?

我们绝对需要。

We absolutely do.

是的。

Yes.

为什么？

Why?

原因有很多，但其中一个是因为，即使推理与训练的比例发生变化，你也无法通过其他任何方式实现如此大规模的训练，唯有依靠将计算资源集中于单一问题上。

Well, because the there's multiple reasons, but one is that even as the balance of how much inference versus training changes, that you cannot get massive scale training through any other way besides this concentration of compute on one problem.

所以我认为会发生的是，部署的规模会大幅增加，但有时你会有一批特定的预训练任务，需要集中大量计算资源去做。

And so I think that the the thing that I think will happen is there's some amount of the the, the deployment footprint goes up quite a lot, but that sometimes there will be you have a particular mass of pretraining run, and you really wanna concentrate a bunch of compute in there.

我也认为英伟达团队非常出色，做了真正了不起的工作。

I also think that the NVIDIA team is just incredible and does really, really amazing work.

因此，我们与他们合作得非常紧密。

And so, yeah, we we partner very closely with them.

难道不会有一天，人们只是说我们已经预训练得足够多了，模型已经足够聪明了吗？

Isn't there gonna be a time where people just say we've pretrained enough, the models are smart enough?

我觉得这有点像，一旦人类解决了我们面前的所有问题，那时我们才可以说这句话。

I think that that's a little bit like once humanity has solved all problems in front of us, then maybe we can we can say that.

对。

Right.

但我认为我们想要实现的目标上限，我觉得我们过去五十年左右可能已经有所退缩了。

But I think that the ceiling of what we wanna accomplish, like, I think that that there's just so much ambition that that maybe we've, over the past fifty years or so, just sort of backed off from.

对吧？

Right?

你想想看，即使是一些看似很明确的问题，比如我们能否为所有人提供医疗保障——这不仅仅是治疗疾病，而是真正注重预防，关注生活方式，帮助人们在疾病发生前尽早发现潜在风险。

You think about I mean, even problems that seem very clear, like, can we have health care for everyone that is not just that's actually preventative, not just targeting when people have a problem, but really think about the lifestyle and how to really help people early detect potential diseases before they happen.

我认为，通过更智能的模型，我们实际上能够解决这类问题。

Like, that's a problem that I think we can actually achieve through more intelligent models.

可能在某个阶段，你可以彻底解决这个问题，然后你会想：我还需要一个聪明两倍的模型吗？

And there's probably some level where you can totally solve that problem, and then you say, well, do I need a model that's two times smarter?

但还有其他一些问题，会真正需要这样的模型。

But there's other problems that are going to demand that.

我们来谈谈建设这些数据中心的数学问题。

Let's talk about the math about building these data centers.

你今年早些时候提到了一千一百亿美元。

You raised 110,000,000,000 earlier this year.

这个数字背后的计算依据是什么？

What's the math behind that?

这笔钱会直接投入数据中心吗？

Does that money go right into data centers?

你打算如何向投资者返还这笔资金？

How do you think about how you're gonna return that money to investors?

谈谈这些计算吧。

Talk about those calculations.

是的。

Yeah.

我认为，我们面前的巨大开支本质上是计算资源。

So I think it's as simple as the massive expense we see in front of us is compute.

但你可以把计算资源看作不是成本中心，而是收入中心。

But you can think of compute not as a cost center but as a revenue center.

把它想象成雇佣销售人员。

Think of it a little bit like hiring salespeople.

对吧？

Right?

你想要雇佣多少销售人员？

How many salespeople do you want to hire?

只要你能卖出你的产品，只要你有一个可扩展的销售方式，那么销售人员越多，你创造的收入就越多。

And as long as you can sell your product, as long as you have a scalable way to sell that product, then the more salespeople you have, the more revenue you will make.

我认为我们所处的世界是，我们不断发现无法快速建造足够的算力来跟上需求。

And I think the world that we're in is we have continually found we cannot build compute fast enough to keep up with demand.

我对这一点有非常具体的体会。

And I see this very concretely.

对吧？

Right?

目前，我们必须做出非常痛苦的决定，关于要推出什么产品，以及算力分配到哪里，我认为整个经济体系都将经历类似的情况。

Right now, we have to make very painful decisions about what we're launching, about where the compute goes, and that I think we're going to experience this more broadly within the economy.

当我们转向这个由人工智能驱动的经济时，问题将是：哪些问题将获得这些巨大的算力支持？

As we shift to this AI powered economy, the question will be, what problems are going to get that massive compute?

你如何实现规模化，让每个人都能拥有一个专属的智能代理？

How do you scale so everyone can have a personal agent running for them?

如何让每个人都能使用像Codecs这样的系统？

How can everyone be using systems like Codecs?

世界上根本就没有足够的算力来实现这一点。

Like, there just isn't enough compute in the world to be able to do that.

所以我们正在努力提前解决这个问题。

And so we're trying to get ahead of that problem.

但这确实是一个全新的领域。

But it is a new category.

对吧？

Right?

所以你是带着十足的信心在做这件事。

So you're doing it with real confidence.

我的意思是，从资金规模上看，世界上从未有如此巨额的资金投入到这样一个项目中。

I mean, in sums of money, the world has never seen put towards a project like this.

当你在打造一个全新领域时，如何能确信它一定会成功？

When you're building a new category, how do you do it with certainty that it's gonna work out?

我认为这涉及几个关键因素。

Well, I think there's several components that go into it.

首先，目前已经有历史先例了。

So the first is there is historical precedent at this point.

从我们推出ChatGPT的那一刻起，我就记得和团队进行过完全相同的对话，他们问我该买多少算力。

From the moment we launched ChatGPT, I remember talking with my team having this exact conversation where they said, how much compute should we buy?

我说：全部都要。

I said, all of it.

他们说：不行。

They said, no.

不行。

No.

不行。

No.

真的吗？

Really.

我们该买多少算力？

How much compute should we buy?

我说，无论我们多么努力地建设，我都清楚我们无法跟上需求的增长。

I said, no matter how much we try to build, I know we're not going to be able to keep up with the demand.

这一点一直属实，而且自那以后每年都是如此。

And that has been true, and that has been true every year since then.

而挑战在于，这些计算资源的采购必须提前十八个月，有时是二十四个月，甚至更久就锁定，而它们实际交付还要更晚，这意味着你必须真正地前瞻性预测。

And the the challenge is that these compute purchases, you have to lock them in eighteen months, sometimes twenty four months, sometimes longer in advance of them actually being delivered, which means you really need to project forward.

我认为，我们正在走向的世界是，到目前为止，我们的大部分收入来自消费者订阅，这将始终非常重要。

And I think that the world that we're moving towards is one where, to date, most of our revenue has come from consumer subscriptions, and that will always be very important.

我们还有其他正在兴起的收入来源。

There's other revenue streams we have emerging as well.

但眼下明显浮现的机会是知识型工作，我们正清晰地看到，每个企业都在应用这项技术，并意识到它确实非常有效。

But the the opportunity that clearly is emerging now is knowledge work, And we're seeing this very concretely across every single enterprise realizing this technology, it actually really works.

为了保持竞争力，他们必须采用它。

And to be competitive, they need to adopt it.

你可以看到，这些软件工程师自发地使用它，而我们开始看到它在企业内部各种知识型工作中逐渐普及。

And you can see this organic energy of all these software engineers using it, and then we're starting to see the percolation of people using it for various knowledge work inside of the enterprises.

而且，这个行业中人们愿意付费以及收入增长的趋势非常明显。

And the willingness to pay and the revenue growth that you're seeing in this industry is very clear.

对吧？

Right?

现在正清楚地发生着，你只需要将这一趋势延续下去。

It's very clearly happening right now, and you just project that forward.

我们能看到一件事，可能是外界看不到的，那就是这些模型将如何改进的清晰路径。

And we look like, one thing we get to see that maybe the world doesn't is the line of sight to how these models will improve.

所有这些加在一起表明，经济——这是一个庞大的领域。

And all of this together says that the economy, which is a massive thing.

对吧？

Right?

经济实在是太庞大了。

The economy is just so large.

这几乎难以想象。

It's it's almost incomprehensible.

从现在开始，经济的所有增长，最关键的部分将取决于人工智能，以及你利用人工智能和可用计算能力的水平。

All of the growth, like, the the the highest order bit on how this economy grows from here will be about AI, how well you can leverage AI, and the computational power you have available to power it.

你说目前消费者订阅是你最大的收入来源。

You said consumer subscriptions are your biggest source of revenue right now.

是预计这个会反转，企业业务将成为最大的收入来源吗？

Is the

我认为，嗯，我认为……

projection that that will flip and that business will be the biggest source?

我认为，嗯，我认为……

I think well, I think

这非常明确，企业领域的发展速度非常快，但也不仅仅是企业，因为我认为‘企业’本身的含义也在变化。

that that it is very, like, very clear how quickly the the, yeah, the enterprise it's not just enterprise because I think enterprise is also changing what it means.

对。

Right.

所以，实际上是人们在用它来进行生产性的知识工作，类似这样的用途。

So, really, people using it for productive knowledge work for those kinds of things.

我认为在定价方面，有一点值得注意，那就是如果你有ChatGPT的消费者订阅，你就可以使用Codex。

And I think that as we think about pricing, one thing, if you look at how Codex works right now is if you have a ChatGPT consumer subscription, you can use Codex.

所以我认为，这种分类方式不会像这样明确地划分为这个类别、那个类别。

And so I think it's not gonna be as well defined as this category, that category.

我认为真正重要的是，作为用户，你将拥有一个类似于笔记本电脑的入口，通往数字世界，而收入本质上将来自这个入口。

I think it will really be about you as a user are going to have, just again, like your laptop, this portal to the digital world, and that is what the the revenue fundamentally will will will come from.

达里奥说，我在想你。

Dario said, I think about you.

有一些玩家在孤注一掷，把杠杆拉得太远，我非常担心。

There are some players who are YOLOing, who pull the wrist dial too far, and I'm very concerned.

我认为他是在提及你的基础设施投入。

I think he's referencing your infrastructure bets there.

你对此怎么看？

What do you think about that?

我只是不同意。

Well, I just disagree.

我认为我们一直非常谨慎，并且密切关注着未来的发展。

I think we've been very thoughtful and very much seeing what is coming.

我认为即使在今年，我们也会看到所有参与方都将面临算力紧张的局面。

And I think that we will see even this year how everyone who is participating is going to be compute strapped.

我认为我们在意识到这一趋势即将到来并提前布局、预热这项技术的发展方面是最领先的。

And I think we have been the most forward in realizing that this is coming and building anticipation of how this technology is playing out.

我认为我们看到的情况是，其他玩家可能直到去年年底才意识到这一点，开始匆忙寻找可用的算力，但其实根本没多少可用的。

And I think that what we have seen is that for other players, that they kind of realized that probably late last year and started scrambling to see what compute is available, and there really wasn't any.

所以，尽管人们很容易做出这样的陈述，但我认为大家都已经意识到，这项技术确实有效。

And so think that even as people it's very easy to to make statements like that, but I think that everyone has kind of realized that this technology, it's working.

它已经来了。

It's here.

它是真实的。

It's real.

对吧？

Right?

软件工程只是其中一个例子，我们从根本上受限于可用的计算能力。

Software engineering is just the first example of it and that we are fundamentally limited by the computational power available.

你还提到，如果他对预测的判断稍微偏差一点，他的公司就可能破产。

And you said that also that if he's off by with his prediction by a little bit, then, his the company could potentially go bankrupt.

是这样吗

Is that

你们的情况也一样吗？

the same case for you?

我想说的是。

I think that look.

我认为这里实际上有更多的退出选项。

I I think that that there's actually more degrees of off ramp here K.

如果你开始担心最坏的情况，我认为这是一个非常合理的问题。

If you start to worry about the downside case, which I think is a very reasonable question.

对吧？

Right?

但在某种程度上，我认为这场赌注并不在于任何一家公司。

But to some extent, what I think the bet is on isn't about any one company.

而是在于整个行业。

It's really about the sector.

关键在于，你是否相信这项技术能够被实现，并带来我们所看到的这种巨大价值？

It's really about do you believe this technology can be produced and can deliver this massive amount of value that we see coming?

而且，我再举一些实证例子。

And, again, I'll point to proof points.

对吧？

Right?

软件工程就是这样，如果你不是软件工程师，也没用过 Codefs，你就很难体会到它的差异有多大，真的很难描述。

That software engineering, it just like, the degree to which if you're not a software engineer, you haven't tried Codefs, the degree to which is different, like, it's just hard to describe.

我认为人们会很快亲身体验到这一点。

And I think that people will experience it very quickly.

比如，六个月前，对我们来说，我们内部已经看到了这种变化，但那时外部的实证还比较少。

Like, you know, six months ago, I think that for us, we we saw this internally, but there were less proof points out there.

现在已经有实际的证据了。

Now there's proof points out there.

六个月后，我认为每个人都会感受到。

Six months from now, I think that everyone will feel it.

我认为我们都会体会到这种痛苦：有一个出色的模型，却根本无法获得，因为计算资源严重不足。

And I think that we will all feel the pain of there's an awesome model, and there's just no availability because there's not enough compute.

是的。

Yeah.

但当我们回顾对2026年的预测时，在去年年底的这档节目中，和我们一同出席的兰詹·雷曾说，2026年将是每个人都在使用智能代理的一年。

But as we were looking at our predictions for 2026, on this show, we had a conversation towards the end of the year last year where Ranjan Rai, who's on with us, was like, twenty twenty six is gonna be the year where everybody uses agents.

我说，对。

And I said, yeah.

但我得亲眼看到、亲自使用这些智能代理，才会相信这一点。

Well, I'll believe that when I see it, and I'm using the agents.

所以我们现在就在这里。

So here we are.

我们开始了。

Here we go.

你用它来做什么？

What what do you use it for?

我用它来为我同事内部构建工具，以便大家能就视频何时发布以及缩略图应该是什么样子达成一致。

I I use it to build build tools internally for my for the people who I work with to sort of get on the same page about about when videos are coming and what the thumbnails need to look like.

我还在整合来自YouTube的内容。

And I'm also integrating things from, from YouTube.

这样我们就能基于缩略图和我从未花钱购买的自定义软件，来评估视频的表现。

And so we can basically then rank how the videos are doing based off of thumbnail and, like, a custom built piece of software that I never would have paid for.

我认为这个时刻有趣的一点是，软件虽然被大众使用，

And that's one of the things that I think is interesting about this moment, I guess, is that software, it's it's scales used by the masses.

但当你使用它时，会发现有太多东西并不是为你量身打造的。

But, when you use it, therefore, there's gonna be so many things that are not made for you.

而也许这正是它让我们能够以更自然的方式与软件互动的原因。

And maybe what this does is it allows us to interact with software in a way that's much more natural.

我认为这才是关键。

I I think that is the key.

而且，我经常在想，我们设计计算机的方式实际上把我们拉入了这个数字世界。

And, again, I just think a lot about the fact that the way we've built computers has really pulled us in into this digital world.

想想你每天花在刷手机上的时间有多少。

You think about how much time you just spend scrolling through your phone.

是的。

Yep.

对吧？

Right?

你花在不断点击各种按钮、试图把这东西连到那东西上的时间有多少。

The amount of time that you spend clicking different buttons and trying to, like, connect this thing to that thing.

为什么呢？

Like, why?

为什么非得这么做不可？

Like, why do you have to do that?

相反，人工智能应该是让机器更贴近你，为你个性化，理解你想要完成的目标。

Instead, the AI being about bringing the machine closer to you, personalizing to you, understanding what you're trying to accomplish.

我们周围充斥着可以与之交谈的计算机文化，它们能为你做事，而现在这一切正逐渐成为现实。

And that we have all this pop culture of just computers you can talk to and that they go and do stuff for you, and it's starting to become real.

这正逐渐变成你可以真正实现的事情。

It's starting to become the thing that you can actually do.

我认为，这种奇妙之处在于，你必须亲自尝试，才能真正理解。

And I think that the amazingness of that is something where you just have to try it to really understand.

所以我确实认为，我们正处在一个非常特殊的时刻。

So I I definitely think it's a it's a very special moment we're in.

是的。

Yep.

那我想知道，为什么人工智能在公众中如此不受欢迎？

Then I wanna know why is AI so unpopular with the public?

例如，尤戈夫说，认为人工智能对社会影响为负面的美国人，是认为其影响为正面的三倍。

Yugov, for instance, says, three times as many Americans expect the effects of AI on society to be negative as they expect it to be positive.