Dario Amodei — “We are near the end of the exponential”

We talked three years ago. In your view, what has been the biggest update over the last three years? What has been the biggest difference between what it felt like then versus now?

我们三年前聊过。在你看来,过去三年最大的变化是什么?当时和现在的感觉有什么最大的不同?

Broadly speaking, the exponential of the underlying technology has gone about as I expected it to go. There's plus or minus a year or two here and there.

从大方向来说,底层技术的指数增长基本符合我的预期。在时间上有正负一两年的误差。

I don't know that I would have predicted the specific direction of code. But when I look at the exponential, it is roughly what I expected in terms of the march of the models from smart high school student to smart college student to beginning to do PhD and professional stuff, and in the case of code reaching beyond that.

我不确定自己是否预测到了代码能力发展的具体方向。但当我看指数曲线时,模型从聪明的高中生水平进步到聪明的大学生水平,再到开始做PhD和专业级工作,在代码方面甚至超越了这个水平——这些大致符合我的预期。

The frontier is a little bit uneven, but it's roughly what I expected. What has been the most surprising thing is the lack of public recognition of how close we are to the end of the exponential.

前沿进展有些不均衡,但大体上符合我的预期。最令我惊讶的是,公众并没有意识到我们离指数增长的终点有多近。

To me, it is absolutely wild that you have people — within the bubble and outside the bubble — talking about the same tired, old hot-button political issues, when we are near the end of the exponential.

对我来说,非常疯狂的是,无论是圈内还是圈外的人,在我们接近指数增长终点的时刻,还在讨论那些老掉牙的政治热点话题。

I want to understand what that exponential looks like right now. The first question I asked you when we recorded three years ago was, "What's up with scaling and why does it work?"

我想了解现在这条指数曲线是什么样的。三年前我们录制时我问你的第一个问题是:「扩展定律是怎么回事,为什么会有效?」

I have a similar question now, but it feels more complicated. At least from the public's point of view, three years ago there were well-known public trends across many orders of magnitude of compute where you could see how the loss improves.

我现在有个类似的问题,但感觉更复杂了。至少从公众视角来看,三年前有众所周知的公开趋势,跨越多个数量级的算力你都能看到损失函数如何改善。

Now we have RL scaling and there's no publicly known scaling law for it. It's not even clear what the story is. Is this supposed to be teaching the model skills? Is it supposed to be teaching meta-learning? What is the scaling hypothesis at this point?

现在我们有了RL扩展,但没有公开已知的扩展定律。甚至连故事逻辑都不清楚。这是在教模型技能吗?还是在教元学习?现在的扩展假说到底是什么?

I actually have the same hypothesis I had even all the way back in 2017. I think I talked about it last time, but I wrote a doc called "The Big Blob of Compute Hypothesis."

其实我的假说和2017年时是一样的。我想上次聊过,我写过一份文档叫做「大算力团假说」。

It wasn't about the scaling of language models in particular. When I wrote it, GPT-1 had just come out. That was one among many things.

它并不是专门讲语言模型扩展的。我写这份文档时GPT-1刚刚发布,那只是众多事物之一。

Back in those days there was robotics. People tried to work on reasoning as a separate thing from language models, and there was scaling of the kind of RL that happened in AlphaGo and in Dota at OpenAI. People remember StarCraft at DeepMind, AlphaStar.

那时候有机器人技术。人们试图把推理作为独立于语言模型的东西来研究,还有AlphaGo和OpenAI的Dota中出现的那种RL扩展。人们还记得DeepMind的StarCraft项目AlphaStar。

It was written as a more general document. Rich Sutton put out "The Bitter Lesson" a couple years later. The hypothesis is basically the same.

它是作为一份更通用的文档写的。Rich Sutton几年后发表了「苦涩的教训」。假说基本上是一样的。

What it says is that all the cleverness, all the techniques, all the "we need a new method to do something" — that doesn't matter very much. There are only a few things that matter. I think I listed seven of them.

它说的是,所有的巧思、所有的技术、所有「我们需要新方法来做某事」——这些都不太重要。真正重要的东西只有几样。我记得列出了七个。

One is how much raw compute you have. The second is the quantity of data. The third is the quality and distribution of data. It needs to be a broad distribution.

第一是你有多少原始算力。第二是数据的数量。第三是数据的质量和分布。分布需要足够广泛。

The fourth is how long you train for. The fifth is that you need an objective function that can scale to the moon.

第四是训练时长。第五是你需要一个可以无限扩展的目标函数。

The pre-training objective function is one such objective function. Another is the RL objective function that says you have a goal, you're going to go out and reach the goal. Within that, there's objective rewards like...

预训练目标函数就是这样一个目标函数。另一个是RL目标函数,它告诉你有一个目标,你要去实现这个目标。在其中,有一些客观的奖励,比如……

You see in math and coding, and there's more subjective rewards like you see in RLHF or higher-order versions of that.

在数学和编程中能看到这种情况,还有更主观的奖励,比如 RLHF 或其更高阶的版本中看到的那样。

Then the sixth and seventh were things around normalization or conditioning, just getting the numerical stability so that the big blob of compute flows in this laminar way instead of running into problems.

然后第六和第七个要素是关于归一化或条件化的,就是确保数值稳定性,让这一大团计算以层流方式流动,而不是遇到问题。

That was the hypothesis, and it's a hypothesis I still hold.

这就是当时的假设,也是我至今仍然持有的假设。

I don't think I've seen very much that is not in line with it.

我认为还没有看到太多与之不符的情况。

The pre-training scaling laws were one example of what we see there. Those have continued going.

预训练的规模法则就是我们在那里看到的一个例子。这些规律一直在持续。

Now it's been widely reported, we feel good about pre-training.

现在已经被广泛报道,我们对预训练感到很有信心。

It's continuing to give us gains. What has changed is that now we're also seeing the same thing for RL.

它持续给我们带来收益。变化的是,现在我们也在 RL 中看到了同样的情况。

We're seeing a pre-training phase and then an RL phase on top of that.

我们看到先有一个预训练阶段,然后在此基础上有一个 RL 阶段。

With RL, it's actually just the same.

对于 RL 来说,实际上完全一样。

Even other companies have published things in some of their releases that say, "We train the model on math contests — AIME or other things — and how well the model does is log-linear in how long we've trained it."

甚至其他公司也在他们的一些发布中公开过这样的内容,说「我们在数学竞赛上训练模型——AIME 或其他比赛——模型表现的好坏与训练时间呈对数线性关系」。

We see that as well, and it's not just math contests.

我们也看到了这一点,而且不仅仅是数学竞赛。

It's a wide variety of RL tasks. We're seeing the same scaling in RL that we saw for pre-training.

这是各种各样的 RL 任务。我们在 RL 中看到了与预训练相同的规模效应。

You mentioned Rich Sutton and "The Bitter Lesson". I interviewed him last year, and he's actually very non-LLM-pilled.

你提到了 Rich Sutton 和「痛苦的教训」。我去年采访过他,他实际上非常不看好 LLM。

I don't know if this is his perspective, but one way to paraphrase his objection is: Something which possesses the true core of human learning would not require all these billions of dollars of data and compute and these bespoke environments, to learn how to use Excel, how to use PowerPoint, how to navigate a web browser.

我不知道这是否是他的观点,但可以这样转述他的反对意见:如果真正掌握了人类学习的核心,就不需要投入数十亿美元的数据和算力,以及这些定制化的环境,来学习如何使用 Excel、如何使用 PowerPoint、如何浏览网页。

The fact that we have to build in these skills using these RL environments hints that we are actually lacking a core human learning algorithm.

我们必须使用这些 RL 环境来构建这些技能的事实,暗示我们实际上缺乏核心的人类学习算法。

So we're scaling the wrong thing. That does raise the question. Why are we doing all this RL scaling if we think there's something that's going to be human-like in its ability to learn on the fly?

所以我们在扩展错误的东西。这确实引出了一个问题:如果我们认为会有某种东西能像人类一样即时学习,那我们为什么还要做这么多 RL 扩展呢?

I think this puts together several things that should be thought of differently.

我认为这把几个应该分开思考的东西混在了一起。

There is a genuine puzzle here, but it may not matter.

这里确实有一个真正的谜题,但它可能并不重要。

In fact, I would guess it probably doesn't matter. There is an interesting thing. Let me take the RL out of it for a second, because I actually think it's a red herring to say that RL is any different from pre-training in this matter.

事实上,我猜它可能并不重要。这里有个有趣的点。让我先把 RL 从中拿出来,因为我实际上认为说 RL 在这个问题上与预训练有什么不同是个误导。

If we look at pre-training scaling, it was very interesting back in 2017 when Alec Radford was doing GPT-1.

如果我们看预训练的规模扩展,2017 年 Alec Radford 做 GPT-1 的时候就很有意思。

The models before GPT-1 were trained on datasets that didn't represent a wide distribution of text.

GPT-1 之前的模型是在不能代表广泛文本分布的数据集上训练的。

You had very standard language modeling benchmarks. GPT-1 itself was trained on a bunch of fanfiction, I think actually.

当时有非常标准的语言建模基准。GPT-1 本身是在一堆同人小说上训练的,我记得确实是这样。

It was literary text, which is a very small fraction of the text you can get.

那是文学文本,只占你能获得的文本的很小一部分。

In those days it was like a billion words or something, so small datasets representing a pretty narrow distribution of what you can see in the world.

那时候大概是十亿个词之类的,这么小的数据集代表的是你在世界上能看到的相当狭窄的分布。

It didn't generalize well. If you did better on some fanfiction corpus, it wouldn't generalize that well to other tasks.

它的泛化能力不好。如果你在某个同人小说语料库上表现更好,它也不会很好地泛化到其他任务上。

We had all these measures. We had all these measures of how well it did at predicting all these other kinds of texts.

我们有各种各样的指标。我们有各种指标来衡量它在预测其他各种文本时的表现。

It was only when you trained over all the tasks on the internet — when you did a general internet

只有当你在互联网上的所有任务上进行训练时——当你做了一个通用的互联网

Scrape from something like Common Crawl or scraping links in Reddit, which is what we did for GPT-2 — that you started to get generalization.

从 Common Crawl 这类来源抓取数据,或者像我们为 GPT-2 做的那样抓取 Reddit 链接——这样你才开始获得泛化能力。

I think we're seeing the same thing on RL.

我认为我们在强化学习上也看到了同样的现象。

We're starting first with simple RL tasks like training on math competitions, then moving to broader training that involves things like code.

我们首先从简单的强化学习任务开始,比如在数学竞赛上训练,然后转向更广泛的训练,涉及代码之类的内容。

Now we're moving to many other tasks.

现在我们正在扩展到许多其他任务。

I think then we're going to increasingly get generalization.

我认为这样我们就会越来越多地获得泛化能力。

So that kind of takes out the RL versus pre-training side of it.

所以这在某种程度上解决了强化学习与预训练的对比问题。

But there is a puzzle either way, which is that in pre-training we use trillions of tokens.

但无论如何都存在一个困惑,那就是在预训练中我们使用了数万亿个 token。

Humans don't see trillions of words. So there is an actual sample efficiency difference here.

人类不会看到数万亿个词。所以这里确实存在样本效率的差异。

There is actually something different here.

这里确实有些不同之处。

The models start from scratch and they need much more training.

模型从零开始,它们需要更多的训练。

But we also see that once they're trained, if we give them a long context length of a million — the only thing blocking long context is inference — they're very good at learning and adapting within that context.

但我们也看到,一旦它们训练完成,如果我们给它们一百万的长上下文长度——阻碍长上下文的唯一因素是推理——它们非常擅长在该上下文中学习和适应。

So I don't know the full answer to this.

所以我不知道这个问题的完整答案。

I think there's something going on where pre-training is not like the process of humans learning, but it's somewhere between the process of humans learning and the process of human evolution.

我认为有些现象正在发生,预训练不像人类学习的过程,而是介于人类学习过程和人类进化过程之间。

We get many of our priors from evolution.

我们的许多先验知识来自进化。

Our brain isn't just a blank slate. Whole books have been written about this.

我们的大脑不仅仅是一块白板。关于这个已经有整本书写过了。

The language models are much more like blank slates.

语言模型更像是白板。

They literally start as random weights, whereas the human brain starts with all these regions connected to all these inputs and outputs.

它们字面上是从随机权重开始的,而人类大脑一开始就有所有这些区域连接到所有这些输入和输出。

Maybe we should think of pre-training — and for that matter, RL as well — as something that exists in the middle space between human evolution and human on-the-spot learning.

也许我们应该把预训练——以及强化学习——看作是存在于人类进化和人类即时学习之间的中间空间的某种东西。

And we should think of the in-context learning that the models do as something between long-term human learning and short-term human learning.

而我们应该把模型所做的上下文学习看作是介于人类长期学习和短期学习之间的东西。

So there's this hierarchy. There's evolution, there's long-term learning, there's short-term learning, and there's just human reaction.

所以存在这样一个层次结构。有进化,有长期学习,有短期学习,还有人类的即时反应。

The LLM phases exist along this spectrum, but not necessarily at exactly the same points.

大语言模型的各个阶段存在于这个谱系之中,但不一定恰好在相同的点上。

There's no analog to some of the human modes of learning. The LLMs are falling in between the points. Does that make sense?

有些人类的学习模式没有对应物。大语言模型落在这些点之间。这样说有意义吗?

Yes, although some things are still a bit confusing.

是的,尽管有些事情仍然有点令人困惑。

For example, if the analogy is that this is like evolution so it's fine that it's not sample efficient, then if we're going to get super sample-efficient agents from in-context learning, why are we bothering to build all these RL environments?

例如,如果这个类比是说这就像进化,所以样本效率低也没关系,那么如果我们要从上下文学习中获得超高样本效率的智能体,我们为什么还要费力构建所有这些强化学习环境呢?

There are companies whose work seems to be teaching models how to use this API, how to use Slack, how to use whatever. It's confusing to me why there's so much emphasis on that if the kind of agent that can just learn on the fly is emerging or has already emerged.

有些公司的工作似乎是教模型如何使用这个 API,如何使用 Slack,如何使用各种东西。如果那种可以即时学习的智能体正在出现或已经出现,我就很困惑为什么还有这么多重点放在这上面。

I can't speak for the emphasis of anyone else. I can only talk about how we think about it.

我不能代表其他人的重点。我只能说说我们是怎么想的。

The goal is not to teach the model every possible skill within RL, just as we don't do that within pre-training.

目标不是在强化学习中教会模型所有可能的技能,就像我们在预训练中也不这样做一样。

Within pre-training, we're not trying to expose the model to every possible way that words could be put together.

在预训练中,我们不是试图让模型接触到词语组合的所有可能方式。

Rather, the model trains on a lot of things and then reaches generalization across pre-training.

相反,模型在很多东西上训练,然后在预训练中达到泛化。

That was the transition from GPT-1 to GPT-2 that I saw up close. The model reaches a point.

这就是我近距离看到的从 GPT-1 到 GPT-2 的转变。模型达到一个临界点。

These moments where I was like, "Oh yeah, you just give the model a list of numbers — this is the cost of the house, this is the square feet of the house — and the model completes the pattern and does linear regression." Not great, but it does it, and it's never seen that exact thing before.

有那么几次我就觉得,「哦对了,你只需要给模型一串数字——这个是房价,这个是房子的面积——然后模型就能识别出模式,做线性回归。」虽然效果不算特别好,但它确实做到了,而且它之前从未见过这个具体的任务。

So to the extent that we are building these RL environments, the goal is very similar to what was done five or ten years ago with pre-training. We're trying to get a whole bunch of data, not because we want to cover a specific document or a specific skill, but because we want to generalize.

所以我们现在构建这些强化学习环境,目标其实和五年或十年前做预训练时很相似。我们试图获取大量数据,不是为了覆盖某个特定文档或某项特定技能,而是为了让模型能够泛化。

I think the framework you're laying down obviously makes sense. We're making progress toward AGI. Nobody at this point disagrees we're going to achieve AGI this century. The crux is you say we're hitting the end of the exponential. Somebody else looks at this and says, "We've been making progress since 2012, and by 2035 we'll have a human-like agent."

我觉得你提出的这个框架显然是合理的。我们确实在朝着AGI的方向前进。现在没人会否认我们会在本世纪实现AGI。关键分歧在于,你说我们正在触及指数增长的尽头。而另一些人看到这些进展会说,「我们从2012年就开始取得进展了,到2035年我们就会有类人的智能体。」

Obviously we're seeing in these models the kinds of things that evolution did, or that learning within a human lifetime does. I want to understand what you're seeing that makes you think it's one year away and not ten years away.

显然我们在这些模型中看到了进化所做的事情,或者说人类在一生中学习所做的事情。我想理解的是,你看到了什么,让你认为这件事一年之后就会发生,而不是十年之后。

There are two claims you could make here, one stronger and one weaker. Starting with the weaker claim, when I first saw the scaling back in 2019, I wasn't sure. This was a 50/50 thing. I thought I saw something. My claim was that this was much more likely than anyone thinks. Maybe there's a 50% chance this happens.

这里可以提出两种主张,一个更强,一个较弱。先说较弱的主张,2019年我第一次看到scaling规律时,我并不确定。当时对我来说是五五开。我觉得我看到了一些迹象。我当时的判断是,这件事发生的可能性比任何人想的都要高得多。也许有50%的概率会发生。

On the basic hypothesis of, as you put it, within ten years we'll get to what I call a "country of geniuses in a data center", I'm at 90% on that. It's hard to go much higher than 90% because the world is so unpredictable. Maybe the irreducible uncertainty puts us at 95%, where you get to things like multiple companies having internal turmoil, Taiwan gets invaded, all the fabs get blown up by missiles.

对于你提到的基本假设,也就是十年内我们会实现我所说的「数据中心里的天才之国」,我有90%的把握。很难超过90%,因为这个世界太难预测了。也许不可消除的不确定性让我们最多到95%,剩下5%是那些像多家公司内部出现动荡、台湾遭到入侵、所有芯片厂被导弹炸毁这样的情况。

Now you've jinxed us, Dario.

你这是乌鸦嘴了,Dario。

You could construct a 5% world where things get delayed for ten years. There's another 5% which is that I'm very confident on tasks that can be verified. With coding, except for that irreducible uncertainty, I think we'll be there in one or two years. There's no way we will not be there in ten years in terms of being able to do end-to-end coding.

你可以构想一个5%概率的世界,在那里事情会被推迟十年。另外5%的不确定性在于,我对可验证任务非常有信心。对于编程来说,除了那些不可消除的不确定性,我认为一两年内我们就能做到。在端到端编程能力方面,十年内不可能做不到。

My one little bit of fundamental uncertainty, even on long timescales, is about tasks that aren't verifiable: planning a mission to Mars; doing some fundamental scientific discovery like CRISPR; writing a novel. It's hard to verify those tasks. I am almost certain we have a reliable path to get there, but if there's a little bit of uncertainty it's there.

我唯一的一点根本性不确定,即使放在长时间尺度上,是关于那些不可验证的任务:规划火星任务;做像CRISPR那样的基础科学发现;写小说。这些任务很难验证。我几乎确信我们有可靠的路径达到那里,但如果有一点点不确定性的话,就在这里。

On the ten-year timeline I'm at 90%, which is about as certain as you can be. I think it's crazy to say that this won't happen by 2035. In some sane world, it would be outside the mainstream.

对于十年时间线,我有90%的把握,这基本上是你能达到的最高确定性了。我认为说2035年之前不会发生是很疯狂的。在一个理性的世界里,这种观点应该是非主流的。

But the emphasis on verification hints to me a lack of belief that these models are generalized. If you think about humans, we're both good at things for which we get verifiable reward and things for which we don't.

但你对可验证性的强调,让我觉得你似乎不太相信这些模型已经实现了泛化。想想人类,我们既擅长那些能得到可验证反馈的事情,也擅长那些得不到可验证反馈的事情。

No, this is why I'm almost sure. We already see substantial generalization from things that verify to things that don't. We're already seeing that. But it seems like you were emphasizing—

不,这正是我几乎确信的原因。我们已经看到了从可验证任务到不可验证任务的大量泛化。我们已经在看到这种现象了。但你好像在强调——

This as a spectrum which will split apart which domains in which we see more progress.

这是一个光谱,它会分裂开来,显示我们在哪些领域看到更多进展。

That doesn't seem like how humans get better. The world in which we don't get there is the world in which we do all the verifiable things. Many of them generalize, but we don't fully get there.

这似乎不像人类变得更好的方式。我们无法达到那里的世界,是我们只做所有可验证事情的世界。其中许多会泛化,但我们并没有完全到达那里。

We don't fully color in the other side of the box. It's not a binary thing.

我们没有完全填满盒子的另一边。这不是一个非此即彼的事情。

Even if generalization is weak and you can only do verifiable domains, it's not clear to me you could automate software engineering in such a world.

即使泛化能力很弱,你只能做可验证的领域,我也不清楚在这样的世界里你能否自动化软件工程。

You are a software engineer in some sense, but part of being a software engineer for you involves writing long memos about your grand vision.

从某种意义上说你是一名软件工程师,但对你来说,作为软件工程师的一部分涉及撰写关于你宏大愿景的长篇备忘录。

I don't think that's part of the job of SWE. That's part of the job of the company, not SWE specifically.

我不认为那是 SWE 工作的一部分。那是公司工作的一部分,而不是 SWE 特有的。

But SWE does involve design documents and other things like that.

但 SWE 确实涉及设计文档和其他类似的事情。

The models are already pretty good at writing comments.

模型在编写注释方面已经相当不错了。

Again, I'm making much weaker claims here than I believe, to distinguish between two things.

再次强调,我在这里提出的主张比我实际相信的要弱得多,是为了区分两件事。

We're already almost there for software engineering.

我们在软件工程方面已经快到那里了。

By what metric? There's one metric which is how many lines of code are written by AI.

按什么标准?有一个标准是 AI 写了多少行代码。

If you consider other productivity improvements in the history of software engineering, compilers write all the lines of software.

如果你考虑软件工程历史上的其他生产力改进,编译器会写所有的软件代码行。

There's a difference between how many lines are written and how big the productivity improvement is. "We're almost there" meaning…

写了多少行代码和生产力提升有多大之间是有区别的。「我们快到那里了」的意思是……

How big is the productivity improvement, not just how many lines are written by AI?

生产力提升有多大,而不仅仅是 AI 写了多少行?

I actually agree with you on this. I've made a series of predictions on code and software engineering.

在这一点上我实际上同意你的看法。我对代码和软件工程做了一系列预测。

I think people have repeatedly misunderstood them.

我认为人们一再误解了它们。

Let me lay out the spectrum. About eight or nine months ago, I said the AI model will be writing 90% of the lines of code in three to six months.

让我阐述一下这个光谱。大约八九个月前,我说 AI 模型将在三到六个月内编写 90% 的代码行。

That happened, at least at some places. It happened at Anthropic, happened with many people downstream using our models.

这已经发生了,至少在一些地方。它在 Anthropic 发生了,在许多使用我们模型的下游用户那里也发生了。

But that's actually a very weak criterion.

但这实际上是一个非常弱的标准。

People thought I was saying that we won't need 90% of the software engineers. Those things are worlds apart.

人们以为我是在说我们不需要 90% 的软件工程师。这些事情相差十万八千里。

The spectrum is: 90% of code is written by the model, 100% of code is written by the model. That's a big difference in productivity.

光谱是:90% 的代码由模型编写,100% 的代码由模型编写。这在生产力上是一个巨大的差异。

90% of the end-to-end SWE tasks — including things like compiling, setting up clusters and environments, testing features, writing memos — are done by the models.

90% 的端到端 SWE 任务——包括编译、设置集群和环境、测试功能、编写备忘录等——都由模型完成。

100% of today's SWE tasks are done by the models.

今天 100% 的 SWE 任务都由模型完成。

Even when that happens, it doesn't mean software engineers are out of a job.

即使发生这种情况,也不意味着软件工程师会失业。

There are new higher-level things they can do, where they can manage.

他们可以做新的更高层次的事情,他们可以管理。

Then further down the spectrum, there's 90% less demand for SWEs, which I think will happen but this is a spectrum.

然后在光谱的更远端,对 SWE 的需求减少 90%,我认为这会发生,但这是一个光谱。

I wrote about it in "The Adolescence of Technology" where I went through this kind of spectrum with farming.

我在《技术的青春期》中写过这个问题,我在那里用农业经历了这种光谱。

I actually totally agree with you on that.

在这一点上我完全同意你的看法。

These are very different benchmarks from each other, but we're proceeding through them super fast.

这些是彼此非常不同的基准,但我们正在超快地完成它们。

Part of your vision is that going from 90 to 100 is going to happen fast, and that it leads to huge productivity improvements.

你愿景的一部分是,从 90 到 100 的过程会很快发生,并且会带来巨大的生产力提升。

But what I notice is that even in greenfield projects people start with Claude Code or something, people report starting a lot of projects… Do we see in the world out there

但我注意到的是,即使在全新项目中,人们也会从 Claude Code 或类似工具开始,人们报告说启动了很多项目……我们在外面的世界中是否看到

A renaissance of software, all these new features that wouldn't exist otherwise? At least so far, it doesn't seem like we see that. So that does make me wonder.

软件的复兴,所有这些原本不会存在的新功能?至少到目前为止,我们似乎并没有看到这种情况。所以这确实让我困惑。

Even if I never had to intervene with Claude Code, the world is complicated. Jobs are complicated.

即使我从不需要干预 Claude Code,现实世界也是复杂的。工作也是复杂的。

Closing the loop on self-contained systems, whether it's just writing software or something, how much broader gains would we see just from that?

在自包含系统上闭环,无论是写软件还是其他什么,仅仅从这一点我们能看到多大的收益?

Maybe that should dilute our estimation of the "country of geniuses".

也许这应该削弱我们对「天才之国」的估计。

I simultaneously agree with you that it's a reason why these things don't happen instantly, but at the same time, I think the effect is gonna be very fast.

我一方面同意你的观点,这确实是为什么这些事情不会立即发生的原因,但同时,我认为这个效应会非常快。

You could have these two poles. One is that AI is not going to make progress. It's slow. It's going to take forever to diffuse within the economy.

你可以看到两个极端。一个是 AI 不会取得进展。它很慢。要在经济中扩散需要永远的时间。

Economic diffusion has become one of these buzzwords that's a reason why we're not going to make AI progress, or why AI progress doesn't matter.

经济扩散已经成为这些流行词之一,用来解释为什么我们不会在 AI 上取得进展,或者为什么 AI 进展无关紧要。

The other axis is that we'll get recursive self-improvement, the whole thing. Can't you just draw an exponential line on the curve?

另一个极端是我们会实现递归自我改进,整个过程。你不就是在曲线上画一条指数线吗?

We're going to have Dyson spheres around the sun so many nanoseconds after we get recursive. I'm completely caricaturing the view here, but there are these two extremes.

我们会在实现递归后的几纳秒内就在太阳周围建造戴森球。我在这里完全是在讽刺这种观点,但确实存在这两个极端。

But what we've seen from the beginning, at least if you look within Anthropic, there's this bizarre 10x per year growth in revenue that we've seen. So in 2023, it was zero to $100 million.

但从一开始我们所看到的,至少如果你看 Anthropic 内部,我们看到了这种惊人的每年十倍收入增长。所以在2023年,是从零到一亿美元。

In 2024, it was $100 million to $1 billion. In 2025, it was $1 billion to $9-10 billion.

2024年是从一亿美元到十亿美元。2025年是从十亿美元到九十亿到一百亿美元。

You guys should have just bought a billion dollars of your own products so you could just…

你们应该直接买十亿美元自己的产品,这样你们就可以……

And the first month of this year, that exponential is... You would think it would slow down, but we added another few billion to revenue in January.

而今年的第一个月,这个指数级增长……你会以为它会放缓,但我们在一月份又增加了几十亿美元的收入。

Obviously that curve can't go on forever. The GDP is only so large.

显然这条曲线不可能永远持续下去。GDP 总共就那么大。

I would even guess that it bends somewhat this year, but that is a fast curve. That's a really fast curve. I would bet it stays pretty fast even as the scale goes to the entire economy.

我甚至猜测今年会有所放缓,但这仍然是一条快速的曲线。这是一条非常快的曲线。我敢打赌即使规模扩展到整个经济体,它仍会保持相当快的速度。

So I think we should be thinking about this middle world where things are extremely fast, but not instant, where they take time because of economic diffusion, because of the need to close the loop.

所以我认为我们应该考虑这个中间状态的世界,事物发展极快,但不是瞬间完成,它们需要时间,因为经济扩散,因为需要闭环。

Because it's fiddly: "I have to do change management within my enterprise... I set this up, but I have to change the security permissions on this in order to make it actually work... I had this old piece of software that checks the model before it's compiled and released and I have to rewrite it. Yes, the model can do that, but I have to tell the model to do that. It has to take time to do that."

因为这很繁琐:「我必须在企业内部做变更管理……我设置好了这个,但我必须更改安全权限才能让它真正工作……我有这个旧软件在模型编译和发布前检查它,我必须重写它。是的,模型可以做到,但我必须告诉模型去做。这需要时间来完成。」

So I think everything we've seen so far is compatible with the idea that there's one fast exponential that's the capability of the model. Then there's another fast exponential that's downstream of that, which is the diffusion of the model into the economy.

所以我认为到目前为止我们看到的一切都符合这样的想法:有一条快速的指数曲线,那就是模型的能力。然后还有另一条快速的指数曲线,是前者的下游,也就是模型向经济中的扩散。

Not instant, not slow, much faster than any previous technology, but it has its limits.

不是瞬间的,也不慢,比以往任何技术都快得多,但它有其局限性。

When I look inside Anthropic, when I look at our customers: fast adoption, but not infinitely fast.

当我审视 Anthropic 内部,当我看我们的客户时:采用速度很快,但不是无限快。

Can I try a hot take on you? Yeah.

我能说个激进的观点吗?可以。

I feel like diffusion is cope that people say. When the model isn't able to do something, they're like, "oh, but it's a diffusion issue." But then you should use the comparison to humans.

我觉得扩散是人们用来自我安慰的说法。当模型做不到某事时,他们就说「哦,但这是扩散问题」。但你应该用与人类的对比来看。

You would think that the inherent advantages that AIs have would make diffusion a much easier problem for new AIs getting onboarded than new humans getting onboarded.

你可能会觉得,AI 本身具有的固有优势应该会让新 AI 的扩散和上手比新员工要容易得多。

An AI can read your entire Slack and your drive in minutes.

AI 可以在几分钟内读完你所有的 Slack 消息和云端文件。

They can share all the knowledge that the other copies of the same instance have.

它们可以共享同一实例的其他副本所拥有的全部知识。

You don't have this adverse selection problem when you're hiring AI, so you can just hire copies of a vetted AI model. Hiring a human is so much more of a hassle.

雇佣 AI 时不存在逆向选择问题,你只需要雇佣经过审核的 AI 模型的副本就行了。而雇佣人类要麻烦得多。

People hire humans all the time. We pay humans upwards of $50 trillion in wages because they're useful, even though in principle it would be much easier to integrate AIs into the economy than it is to hire humans.

但人们仍然一直在雇佣人类。我们每年支付给人类超过 50 万亿美元的工资,因为他们有用,尽管理论上把 AI 整合到经济中应该比雇佣人类容易得多。

The diffusion doesn't really explain.

光用扩散速度并不能真正解释这个现象。

I think diffusion is very real and doesn't exclusively have to do with limitations on the AI models.

我认为扩散问题是真实存在的,而且并不完全是因为 AI 模型本身的局限性。

Again, there are people who use diffusion as kind of a buzzword to say this isn't a big deal. I'm not talking about that.

有些人把「扩散」当成流行词,用来说这不是什么大事。我说的不是这个意思。

I'm not talking about how AI will diffuse at the speed of previous technologies.

我也不是说 AI 会以过去技术的扩散速度来普及。

I think AI will diffuse much faster than previous technologies have, but not infinitely fast.

我认为 AI 的扩散会比以往的技术快得多,但不会快到瞬间完成。

I'll just give an example of this. There's Claude Code. Claude Code is extremely easy to set up.

我举个例子。我们有 Claude Code 这个产品。Claude Code 设置起来非常简单。

If you're a developer, you can just start using Claude Code.

如果你是开发者,你可以直接开始使用 Claude Code。

There is no reason why a developer at a large enterprise should not be adopting Claude Code as quickly as an individual developer or developer at a startup.

理论上,大企业的开发者采用 Claude Code 的速度应该和独立开发者或创业公司的开发者一样快,没有理由更慢。

We do everything we can to promote it. We sell Claude Code to enterprises.

我们尽一切努力推广它。我们向企业销售 Claude Code。

Big enterprises, big financial companies, big pharmaceutical companies, all of them are adopting Claude Code much faster than enterprises typically adopt new technology.

大型企业、大型金融公司、大型制药公司,它们采用 Claude Code 的速度都比企业通常采用新技术的速度快得多。

But again, it takes time.

但即便如此,这仍然需要时间。

Any given feature or any given product, like Claude Code or Cowork, will get adopted by the individual developers who are on Twitter all the time, by the Series A startups, many months faster than they will get adopted by a large enterprise that does food sales.

任何特定功能或产品,比如 Claude Code 或 Cowork,会被那些天天刷 Twitter 的独立开发者、A 轮创业公司采用,而这比一家做食品销售的大型企业采用要快上好几个月。

There are just a number of factors. You have to go through legal, you have to provision it for everyone.

这里面有很多因素。你得走法务流程,得为所有人配置权限。

It has to pass security and compliance.

还得通过安全和合规审查。

The leaders of the company who are further away from the AI revolution are forward-looking, but they have to say, 'Oh, it makes sense for us to spend 50 million.'

公司领导层离 AI 革命比较远,他们虽然有前瞻性,但也得想明白:「我们花 5000 万美元是有意义的。」

This is what this Claude Code thing is. This is why it helps our company.

「这个 Claude Code 是做什么的,它为什么能帮到我们公司。」

This is why it makes us more productive. Then they have to explain to the people two levels below.

「它为什么能提高我们的生产力。」然后他们还得向下两级的人解释清楚。

They have to say, 'Okay, we have 3,000 developers. Here's how we're going to roll it out to our developers.'

他们得说:「好,我们有 3000 名开发者,我们要这样向他们推广。」

We have conversations like this every day. We are doing everything we can to make Anthropic's revenue grow 20 or 30x a year instead of 10x a year.

我们每天都在进行这样的对话。我们尽一切努力让 Anthropic 的收入每年增长 20 倍或 30 倍,而不只是 10 倍。

Again, many enterprises are just saying, 'This is so productive. We're going to take shortcuts in our usual procurement process.'

确实,很多企业都在说:「这太高效了,我们要在常规采购流程上走捷径。」

They're moving much faster than when we tried to sell them just the ordinary API, which many of them use.

它们的推进速度比我们向他们销售普通 API 时快得多,虽然很多企业也在用我们的 API。

Claude Code is a more compelling product, but it's not an infinitely compelling product.

Claude Code 是个更有吸引力的产品,但它也不是无限吸引人的产品。

I don't think even AGI or powerful AI or 'country of geniuses in a data center' will be an infinitely compelling product.

我认为即使是 AGI、强大的 AI,或者说「数据中心里的天才国度」,也不会是一个能瞬间征服所有人的产品。

It will be a compelling product enough maybe to get 3-5x, or 10x, a year of growth, even when you're in the hundreds of billions of dollars, which is extremely hard to do and has never been done in history before, but not infinitely fast.

这将是一个足够有吸引力的产品,也许能实现每年3到5倍、甚至10倍的增长,即使在规模已经达到数千亿美元的情况下——这极其困难,历史上从未有过先例——但不会是无限快速的增长。

I buy that it would be a slight slowdown.

我认可会有轻微的减速。

Maybe this is not your claim, but sometimes people talk about this like, "Oh, the capabilities are there, but because of diffusion... otherwise we're basically at AGI".

也许这不是你的观点,但有时人们会这样说:「哦,能力已经具备了,只是因为扩散速度的问题……否则我们基本上已经达到AGI了」。

I don't believe we're basically at AGI. I think if you had the "country of geniuses in a data center"... If we had the "country of geniuses in a data center", we would know it.

我不认为我们基本上已经达到AGI。我认为如果你真的拥有「数据中心里的天才之国」……如果我们真有「数据中心里的天才之国」,我们会知道的。

We would know it if you had the "country of geniuses in a data center". Everyone in this room would know it.

如果你真有「数据中心里的天才之国」,我们会知道的。这个房间里的每个人都会知道。

Everyone in Washington would know it. People in rural parts might not know it, but we would know it.

Washington的每个人都会知道。偏远地区的人可能不知道,但我们会知道。

We don't have that now. That is very clear.

我们现在没有那样的东西。这一点非常清楚。

Coming back to concrete prediction... Because there are so many different things to disambiguate, it can be easy to talk past each other when we're talking about capabilities.

回到具体的预测……因为有太多不同的东西需要区分,当我们谈论能力时,很容易出现各说各话的情况。

For example, when I interviewed you three years ago, I asked you a prediction about what we should expect three years from now. You were right. You said, "We should expect systems which, if you talk to them for the course of an hour, it's hard to tell them apart from a generally well-educated human."

例如,三年前我采访你时,我问你对三年后的预测。你说对了。你说:「我们应该期待这样的系统,如果你和它们交谈一个小时,很难将它们与受过良好教育的人类区分开来」。

I think you were right about that.

我认为你在这一点上是对的。

I think spiritually I feel unsatisfied because my internal expectation was that such a system could automate large parts of white-collar work.

但我在精神上感到不满足,因为我内心的期待是这样的系统能够自动化大部分白领工作。

So it might be more productive to talk about the actual end capabilities you want from such a system.

所以,讨论你希望这样的系统具备的实际最终能力可能更有成效。

I will basically tell you where I think we are. Let me ask a very specific question so that we can figure out exactly what kinds of capabilities we should think about soon.

我基本上会告诉你我认为我们处于什么位置。让我问一个非常具体的问题,这样我们就能弄清楚应该很快考虑哪些能力。

Maybe I'll ask about it in the context of a job I understand well, not because it's the most relevant job, but just because I can evaluate the claims about it. Take video editors. I have video editors.

也许我会在一个我很了解的工作背景下来问这个问题,不是因为这是最相关的工作,而是因为我能够评估关于它的说法。就拿视频编辑来说吧,我有视频编辑。

Part of their job involves learning about our audience's preferences, learning about my preferences and tastes, and the different trade-offs we have.

他们工作的一部分涉及了解我们受众的偏好,了解我的偏好和品味,以及我们面临的各种权衡。

They're, over the course of many months, building up this understanding of context.

在几个月的时间里,他们逐渐建立起对这些背景的理解。

The skill and ability they have six months into the job, a model that can pick up that skill on the job on the fly, when should we expect such an AI system?

他们在工作六个月后具备的技能和能力,一个能够在工作中即时学会这种技能的模型,我们什么时候应该期待这样的AI系统?

I guess what you're talking about is that we're doing this interview for three hours.

我想你说的是我们正在进行这个三小时的访谈。

Someone's going to come in, someone's going to edit it.

会有人进来,会有人编辑它。

They're going to be like, "Oh, I don't know, Dario scratched his head and we could edit that out."

他们会说:「哦,我不知道,Dario挠了挠头,我们可以把那个剪掉」。

"Magnify that." "There was this long discussion that is less interesting to people. There's another thing that's more interesting to people, so let's make this edit."

「放大那个」。「这段冗长的讨论对观众来说不太有意思。还有另一个东西对观众更有意思,所以我们来做这个剪辑」。

I think the "country of geniuses in a data center" will be able to do that.

我认为「数据中心里的天才之国」将能够做到这一点。

The way it will be able to do that is it will have general control of a computer screen.

它能够做到这一点的方式是,它将拥有对计算机屏幕的通用控制能力。

You'll be able to feed this in.

你将能够把这些内容输入进去。

It'll be able to also use the computer screen to go on the web, look at all your previous interviews, look at what people are saying on Twitter in response to your interviews, talk to you, ask you questions, talk to your staff, look at the history of edits.

它还能够使用计算机屏幕上网,查看你所有以前的访谈,查看人们在Twitter上对你访谈的评论,与你交谈,向你提问,与你的员工交谈,查看编辑历史。

That you did, and from that, do the job. I think that's dependent on several things.

你之前做的事情,然后基于此来完成工作。我认为这取决于几个因素。

I think this is one of the things that's actually blocking deployment: getting to the point on computer use where the models are really masters at using the computer.

我认为这是实际阻碍部署的问题之一:要让模型在计算机使用方面达到真正精通的程度。

We've seen this climb in benchmarks, and benchmarks are always imperfect measures.

我们看到基准测试分数在不断攀升,当然基准测试总是不完美的衡量标准。

But I think when we first released computer use a year and a quarter ago, OSWorld was at maybe 15%. I don't remember exactly, but we've climbed from that to 65-70%.

但我记得一年零三个月前我们刚发布 computer use 功能时,OSWorld 的分数大概是15%。我记不太清了,但我们已经从那时攀升到了65-70%。

There may be harder measures as well, but I think computer use has to pass a point of reliability.

可能还有更难的衡量标准,但我认为 computer use 必须达到一定的可靠性水平。

Can I just follow up on that before you move on to the next point?

在你继续下一个点之前,我能先追问一下这个问题吗?

For years, I've been trying to build different internal LLM tools for myself. Often I have these text-in, text-out tasks, which should be dead center in the repertoire of these models. Yet I still hire humans to do them.

多年来,我一直在尝试为自己构建各种内部 LLM 工具。我经常有一些文本输入、文本输出的任务,这些本应是这些模型的核心能力范围。但我还是雇人来做这些事。

If it's something like "identify what the best clips would be in this transcript", maybe the LLMs do a seven-out-of-ten job on them. But there's not this ongoing way I can engage with them to help them get better at the job the way I could with a human employee.

比如「找出这份文字稿中最好的片段」这样的任务,LLM 可能能做到七分的水平。但我没法像对待人类员工那样持续与它们互动,帮助它们在工作中不断进步。

That missing ability, even if you solve computer use, would still block my ability to offload an actual job to them.

这种缺失的能力,即使你解决了 computer use 的问题,仍然会阻碍我把真正的工作交给它们。

This gets back to what we were talking about before with learning on the job. It's very interesting. I think with the coding agents, I don't think people would say that learning on the job is what is preventing the coding agents from doing everything end to end.

这又回到我们之前讨论的在工作中学习的问题。这很有意思。我认为对于编程 agent 来说,人们不会说在工作中学习的能力是阻碍编程 agent 端到端完成所有任务的原因。

They keep getting better. We have engineers at Anthropic who don't write any code.

它们一直在变得更好。我们 Anthropic 有些工程师已经不写任何代码了。

When I look at the productivity, to your previous question, we have folks who say, "This GPU kernel, this chip, I used to write it myself. I just have Claude do it." There's this enormous improvement in productivity.

回到你之前的生产力问题,我们有人说「这个 GPU kernel,这个芯片,我以前都是自己写的。现在我就让 Claude 来做。」生产力有了巨大的提升。

When I see Claude Code, familiarity with the codebase or a feeling that the model hasn't worked at the company for a year, that's not high up on the list of complaints I see.

当我观察 Claude Code 时,对代码库的熟悉程度,或者那种模型好像没在公司工作过一年的感觉,这些并不是我看到的主要抱怨。

I think what I'm saying is that we're kind of taking a different path.

我想说的是,我们正在走一条不同的路径。

Don't you think with coding that's because there is an external scaffold of memory which exists instantiated in the codebase? I don't know how many other jobs have that.

你不觉得编程之所以如此,是因为存在一个外部的记忆支架,它实例化在代码库中?我不知道还有多少其他工作具备这种条件。

Coding made fast progress precisely because it has this unique advantage that other economic activity doesn't.

编程之所以进展迅速,正是因为它具有其他经济活动所不具备的这种独特优势。

But when you say that, what you're implying is that by reading the codebase into the context, I have everything that the human needed to learn on the job.

但你这么说的时候,你暗示的是,通过把代码库读入上下文,我就拥有了人类在工作中需要学习的所有东西。

So that would be an example of—whether it's written or not, whether it's available or not—a case where everything you needed to know you got from the context window.

所以这就是一个例子——无论它是否被写下来,是否可获取——你需要知道的一切都能从上下文窗口中获得。

What we think of as learning—"I started this job, it's going to take me six months to understand the code base"—the model just did it in the context.

我们所认为的学习——「我刚开始这份工作,理解代码库需要六个月」——模型只是在上下文中就完成了。

I honestly don't know how to think about this because there are people who qualitatively report what you're saying.

说实话我不知道该如何看待这个问题,因为确实有人在定性地报告你说的这种情况。

I'm sure you saw last year, there was a major study where they had experienced developers try to close pull requests in repositories that they were familiar with. Those developers reported an uplift. They reported that they felt more productive with the use of these models.

我相信你去年看到过,有一项重要研究,让有经验的开发者尝试在他们熟悉的代码仓库中关闭 pull request。那些开发者报告说有提升。他们说使用这些模型后感觉更高效了。

But in fact, if you look at their output and how much was actually merged back in,

但事实上,如果你看他们的产出以及实际有多少被合并回去,

There was a 20% downlift. They were less productive as a result of using these models.

生产力下降了20%。使用这些模型反而降低了他们的生产效率。

So I'm trying to square the qualitative feeling that people feel with these models versus, one, on a macro level, where is this renaissance of software?

所以我在试图理解人们对这些模型的主观感受，与以下问题之间的矛盾：第一，从宏观层面看，软件行业的复兴在哪里？

And then two, when people do these independent evaluations, why are we not seeing the productivity benefits we would expect?

第二，当人们进行这些独立评估时，为什么我们没有看到预期的生产力提升？

Within Anthropic, this is just really unambiguous. We're under an incredible amount of commercial pressure and make it even harder for ourselves because we have all this safety stuff we do that I think we do more than other companies.

在 Anthropic 内部，这一点非常明确。我们面临着巨大的商业压力，而且因为我们做了大量安全方面的工作——我认为比其他公司做得更多——这让我们的处境更加艰难。

The pressure to survive economically while also keeping our values is just incredible. We're trying to keep this 10x revenue curve going.

在保持我们价值观的同时实现经济生存的压力是巨大的。我们正在努力维持这条10倍收入增长曲线。

There is zero time for bullshit. There is zero time for feeling like we're productive when we're not.

我们没有时间搞虚的。我们没有时间自欺欺人地觉得自己很高效。

These tools make us a lot more productive. Why do you think we're concerned about competitors using the tools?

这些工具确实让我们的生产力大幅提升。你觉得我们为什么会担心竞争对手使用这些工具？

Because we think we're ahead of the competitors. We wouldn't be going through all this trouble if this were secretly reducing our productivity.

因为我们认为自己领先于竞争对手。如果这些工具实际上在降低我们的生产力，我们不会费这么大劲。

We see the end productivity every few months in the form of model launches. There's no kidding yourself about this.

我们每隔几个月就能通过模型发布看到最终的生产力成果。这方面没法自欺欺人。

The models make you more productive. One, people feeling like they're productive is qualitatively predicted by studies like this.

这些模型确实提升了生产力。第一，人们感觉自己更高效这一点，在类似研究中已经定性地预测到了。

But two, if I just look at the end output, obviously you guys are making fast progress.

但第二，如果我只看最终产出，显然你们确实在快速进步。

But the idea was supposed to be that with recursive self-improvement, you make a better AI, the AI helps you build a better next AI, et cetera, et cetera.

但原本的设想是通过递归自我改进，你造出更好的 AI，这个 AI 帮你构建下一代更好的 AI，如此循环往复。

What I see instead—if I look at you, OpenAI, DeepMind—is that people are just shifting around the podium every few months.

但我看到的情况是——如果看你们、OpenAI、DeepMind——大家只是每隔几个月在领奖台上轮换位置。

Maybe you think that stops because you've won or whatever. But why are we not seeing the person with the best coding model have this lasting advantage if in fact there are these enormous productivity gains from the last coding model?

也许你认为这种情况会因为你们获胜而停止。但如果最新的编程模型真的带来了巨大的生产力提升，为什么我们没有看到拥有最佳编程模型的公司获得持久优势？

I think my model of the situation is that there's an advantage that's gradually growing. I would say right now the coding models give maybe, I don't know, a 15-20% total factor speed up. That's my view. Six months ago, it was maybe 5%.

我对这种情况的理解是，优势在逐渐增长。我认为目前编程模型带来的全要素生产率提升大概是，我也不确定，15-20%。这是我的看法。六个月前大概只有5%。

So it didn't matter. 5% doesn't register. It's now just getting to the point where it's one of several factors that kind of matters.

所以当时没什么影响。5%根本不明显。现在才刚刚达到一个程度，成为几个重要因素之一。

That's going to keep speeding up. I think six months ago, there were several companies that were at roughly the same point because this wasn't a notable factor, but I think it's starting to speed up more and more.

而且这个提升速度会持续加快。我认为六个月前，有几家公司处于大致相同的水平，因为这还不是一个显著因素，但我认为现在开始加速得越来越明显。

I would also say there are multiple companies that write models that are used for code and we're not perfectly good at preventing some of these other companies from using our models internally.

我还要说的是，有多家公司在开发用于编程的模型，而且我们并不能完全阻止其中一些公司在内部使用我们的模型。

So I think everything we're seeing is consistent with this kind of snowball model. Again, my theme in all of this is all of this is soft takeoff, soft, smooth exponentials, although the exponentials are relatively steep.

所以我认为我们看到的一切都符合这种滚雪球模型。再次强调，我对这一切的主题是：这都是软起飞，平缓流畅的指数增长，尽管这些指数曲线相对陡峭。

So we're seeing this snowball gather momentum where it's like 10%, 20%, 25%, 40%. As you go, Amdahl's law, you have to get all the things that are preventing you from closing the loop out of the way.

所以我们看到这个雪球在积累动能，从10%、20%、25%到40%。随着进展，根据 Amdahl 定律，你必须把所有阻碍你形成闭环的因素都清除掉。

But this is one of the biggest priorities within Anthropic. Stepping back, before in the stack we were talking about when do we get this on-the-job learning?

但这是 Anthropic 内部最重要的优先事项之一。退一步说，之前在讨论技术栈时，我们谈到什么时候能实现这种在岗学习？

It seems like the point you were making on the coding thing is that we actually don't need on-the-job learning. You can have tremendous productivity improvements, you can have potentially trillions of dollars of revenue for AI companies, without this basic human ability to learn on the job.

你在编程这件事上的观点似乎是,我们其实不需要在岗学习能力。即使没有人类这种基本的在工作中学习的能力,AI 公司也能获得巨大的生产力提升,也能创造数万亿美元的潜在收入。

Maybe that's not your claim, you should clarify. But in most domains of economic activity, people say, "I hired somebody, they weren't that useful for the first few months, and then over time they built up the context, understanding."

也许这不是你的观点,你应该澄清一下。但在大多数经济活动领域,人们会说:「我雇了一个人,头几个月他们并不是很有用,然后随着时间推移,他们积累了上下文和理解力。」

It's actually hard to define what we're talking about here. But they got something and then now they're a powerhouse and they're so valuable to us.

其实很难定义我们这里在讨论的是什么。但他们掌握了某种东西,然后现在变成了一个强大的生产力,对我们来说非常有价值。

If AI doesn't develop this ability to learn on the fly, I'm a bit skeptical that we're going to see huge changes to the world without that ability.

如果 AI 不发展出这种即时学习的能力,我对没有这种能力就能给世界带来巨大变革这一点持怀疑态度。

I think two things here. There's the state of the technology right now. Again, we have these two stages.

我认为这里有两点。首先是技术的现状。我们还是有这两个阶段。

We have the pre-training and RL stage where you throw a bunch of data and tasks into the models and then they generalize. So it's like learning, but it's like learning from more data and not learning over one human or one model's lifetime.

我们有预训练和强化学习阶段,在这个阶段你把大量数据和任务喂给模型,然后它们进行泛化。所以这像是学习,但这是从更多数据中学习,而不是在一个人或一个模型的生命周期内学习。

So again, this is situated between evolution and human learning. But once you learn all those skills, you have them.

所以这又是介于进化和人类学习之间的。但一旦你学会了所有这些技能,你就拥有了它们。

Just like with pre-training, just how the models know more, if I look at a pre-trained model, it knows more about the history of samurai in Japan than I do. It knows more about baseball than I do. It knows more about low-pass filters and electronics, all of these things. Its knowledge is way broader than mine.

就像预训练一样,模型知道的更多,如果我看一个预训练模型,它对日本武士历史的了解比我多。它对棒球的了解比我多。它对低通滤波器和电子学的了解,所有这些东西,它的知识面比我广得多。

So I think even just that may get us to the point where the models are better at everything.

所以我认为仅凭这一点就可能让我们达到模型在所有方面都更优秀的程度。

We also have, again, just with scaling the kind of existing setup, the in-context learning. I would describe it as kind of like human on-the-job learning, but a little weaker and a little short term.

我们还有,通过扩展现有的设置,上下文学习。我会把它描述为有点像人类的在岗学习,但稍微弱一些,也更短期一些。

You look at in-context learning and if you give the model a bunch of examples it does get it. There's real learning that happens in context.

你看上下文学习,如果你给模型一堆例子,它确实能掌握。在上下文中确实发生了真正的学习。

A million tokens is a lot. That can be days of human learning.

一百万个 token 是很多的。这可能相当于人类好几天的学习量。

If you think about the model reading a million words, how long would it take me to read a million? Days or weeks at least. So you have these two things.

如果你想想模型阅读一百万个词,我读一百万个词需要多长时间?至少好几天或几周。所以你有这两样东西。

I think these two things within the existing paradigm may just be enough to get you the "country of geniuses in a data center". I don't know for sure, but I think they're going to get you a large fraction of it.

我认为在现有范式内,这两样东西可能就足以让你得到「数据中心里的天才之国」。我不能确定,但我认为它们能让你获得很大一部分。

There may be gaps, but I certainly think that just as things are, this is enough to generate trillions of dollars of revenue. That's one.

可能还有差距,但我确实认为就目前的情况而言,这足以产生数万亿美元的收入。这是第一点。

Two, is this idea of continual learning, this idea of a single model learning on the job. I think we're working on that too.

第二点,是持续学习的想法,也就是单个模型在工作中学习的想法。我认为我们也在研究这个。

There's a good chance that in the next year or two, we also solve that. Again, I think you get most of the way there without it.

很有可能在未来一两年内,我们也能解决这个问题。但我再说一遍,我认为没有它你也能达到大部分目标。

The trillions of dollars a year market, maybe all of the national security implications and the safety implications that I wrote about in "Adolescence of Technology" can happen without it.

每年数万亿美元的市场,也许我在《技术的青春期》中写到的所有国家安全影响和安全影响,都可以在没有它的情况下发生。

But we, and I imagine others, are working on it. There's a good chance that we will get there within the next year or two.

但我们,我想其他人也是,正在研究这个问题。很有可能我们会在未来一两年内实现它。

There are a bunch of ideas. I won't go into all of them in detail, but

有很多想法。我不会详细讨论所有这些,但

One is just to make the context longer. There's nothing preventing longer contexts from working. You just have to train at longer contexts and then learn to serve them at inference.

一种方向就是让上下文变得更长。更长的上下文本身没有什么技术障碍,你只需要在更长的上下文上训练模型,然后学会如何在推理时提供服务就行了。

Both of those are engineering problems that we are working on, and I would assume others are working on them as well.

这两个都是工程问题,我们正在解决,我想其他公司应该也在做同样的事情。

This context length increase—it seemed like there was a period from 2020 to 2023 where from GPT-3 to GPT-4 Turbo, there was an increase from 2,000 context lengths to 128K.

关于上下文长度的增长,从2020年到2023年这段时间,从 GPT-3 到 GPT-4 Turbo,上下文长度从2000增加到了128K。

I feel like for the two-ish years since then, we've been in the same-ish ballpark.

但我感觉从那之后的大约两年时间里,我们基本还是在这个范围内徘徊。

When context lengths get much longer than that, people report qualitative degradation in the ability of the model to consider that full context.

当上下文长度远超这个范围时,人们反映模型处理完整上下文的能力出现了质量下降。

So I'm curious what you're internally seeing that makes you think "10 million contexts, 100 million contexts to get six months of human learning and building context."

所以我很好奇你们内部看到了什么,让你觉得「1000万上下文、1亿上下文,才能达到人类六个月的学习和积累上下文」。

This isn't a research problem. This is an engineering and inference problem.

这不是一个研究问题,这是一个工程和推理问题。

If you want to serve long context, you have to store your entire KV cache.

如果你想提供长上下文服务,就必须存储整个 KV cache。

It's difficult to store all the memory in the GPUs, to juggle the memory around.

要把所有这些内存都存在 GPU 里、在各处调度内存,是很困难的。

I don't even know the details. At this point, this is at a level of detail that I'm no longer able to follow, although I knew it in the GPT-3 era—"These are the weights, these are the activations you have to store..."

我甚至不知道具体细节了。现在这些细节已经到了我跟不上的程度,虽然在 GPT-3 时代我还知道「这些是权重,这些是你要存储的激活值……」

But these days the whole thing is flipped because we have MoE models and all of that.

但现在整个局面都变了,因为我们有了 MoE 模型之类的东西。

Regarding this degradation you're talking about, without getting too specific, there's two things.

关于你提到的性能下降问题,不说得太具体的话,有两件事。

There's the context length you train at, and there's a context length that you serve at.

一个是你训练时用的上下文长度,另一个是你服务时提供的上下文长度。

If you train at a small context length and then try to serve at a long context length, maybe you get these degradations.

如果你在短上下文上训练,然后试图在长上下文下提供服务,可能就会出现这些性能下降。

It's better than nothing—you might still offer it—but you get these degradations.

这总比没有好,你可能仍然会提供这种服务,但确实会有性能下降。

Maybe it's harder to train at a long context length.

也许在长上下文上训练本身就更难。

I want to, at the same time, ask about maybe some rabbit holes.

我想同时问一下,是否可能陷入一些兔子洞。

Wouldn't you expect that if you had to train on longer context length, that would mean that you're able to get less samples in for the same amount of compute?

你不觉得如果必须在更长的上下文上训练,那意味着在同样的算力下你能处理的样本数会更少吗?

Maybe it's not worth diving deep on that. I want to get an answer to the bigger picture question.

也许不值得深入探讨这个。我想得到更宏观问题的答案。

I don't feel a preference for a human editor that's been working for me for six months versus an AI that's been working with me for six months—what year do you predict that that will be the case?

当我对一个为我工作了六个月的人类编辑和一个与我协作了六个月的 AI 没有偏好时,你预测这会发生在哪一年?

My guess for that is there's a lot of problems where basically we can do this when we have the "country of geniuses in a data center."

我的猜测是,有很多这样的问题,基本上当我们拥有「数据中心里的天才之国」时就能做到这一点。

My picture for that, if you made me guess, is one to two years, maybe one to three years. It's really hard to tell. I have a strong view—99%, 95%—that all this will happen in 10 years.

如果让我猜的话,我的预期是一到两年,也许是一到三年。真的很难说。我有一个强烈的看法——99%、95% 的把握——这一切会在10年内发生。

I think that's just a super safe bet.

我认为这是一个非常保险的预测。

I have a hunch—this is more like a 50/50 thing—that it's going to be more like one to two, maybe more like one to three.

我有一种直觉——这更像是五五开——会更接近一到两年,也许更像一到三年。

So one to three years—country of geniuses, and the slightly less economically valuable task of editing videos.

所以是一到三年——天才之国,以及经济价值稍低一些的视频编辑任务。

It seems pretty economically valuable, let me tell you.

让我告诉你,这看起来经济价值还挺高的。

It's just there are a lot of use cases like that. There are a lot of similar ones.

只是有很多这样的使用场景,有很多类似的情况。

So you're predicting that within one to three years.

所以你预测是在一到三年内。

And then, generally, Anthropic has predicted that by late '26 or early '27 we will have AI systems that "have the ability to navigate interfaces available to humans doing digital work today, intellectual capabilities matching or exceeding that of Nobel Prize winners, and the ability to interface with the physical world."

总的来说,Anthropic 预测到 2026 年底或 2027 年初,我们将拥有这样的 AI 系统:「能够操作人类现在用于数字工作的界面,拥有匹敌甚至超越诺贝尔奖得主的智力能力,并且能够与物理世界交互。」

You gave an interview two months ago with DealBook where you were emphasizing your company's more responsible compute scaling as compared to your competitors.

两个月前你接受 DealBook 采访时,强调你们公司在算力扩张方面比竞争对手更负责任。

I'm trying to square these two views. If you really believe that we're going to have a country of geniuses, you want as big a data center as you can get. There's no reason to slow down.

我试图理解这两种观点怎么协调。如果你真的相信我们将拥有一个「天才之国」,那你会想要尽可能大的数据中心,没理由放慢脚步。

The TAM of a Nobel Prize winner, that can actually do everything a Nobel Prize winner can do, is trillions of dollars. So I'm trying to square this conservatism, which seems rational if you have more moderate timelines, with your stated views about progress.

一个真正能做到诺贝尔奖得主所能做的一切事情的 AI,其总体可获市场规模是数万亿美元。所以我想理解这种保守态度——如果时间线更温和的话似乎合理——如何与你对进展的公开观点相符。

It actually all fits together. We go back to this fast, but not infinitely fast, diffusion. Let's say that we're making progress at this rate. The technology is making progress this fast.

其实这一切都能自洽。我们回到那个「快速但并非无限快」的扩散过程。假设我们正以这样的速度取得进展,技术正以这样的速度发展。

I have very high conviction that we're going to get there within a few years. I have a hunch that we're going to get there within a year or two.

我非常确信我们会在几年内达到那个目标。我有种直觉,我们会在一两年内就达到。

So there's a little uncertainty on the technical side, but pretty strong confidence that it won't be off by much. What I'm less certain about is, again, the economic diffusion side.

所以技术层面有些不确定性,但我很有信心偏差不会太大。我不太确定的是,经济扩散这一面。

I really do believe that we could have models that are a country of geniuses in the data center in one to two years.

我真的相信,我们可能在一到两年内就能在数据中心里拥有相当于一个天才之国的模型。

One question is: How many years after that do the trillions in revenue start rolling in? I don't think it's guaranteed that it's going to be immediate.

问题在于:那之后多少年,数万亿美元的收入才会开始滚滚而来?我不认为保证会立即发生。

It could be one year, it could be two years, I could even stretch it to five years although I'm skeptical of that. So we have this uncertainty. Even if the technology goes as fast as I suspect that it will, we don't know exactly how fast it's going to drive revenue.

可能是一年,可能是两年,甚至我可以拉长到五年,虽然我对此持怀疑态度。所以我们面临这种不确定性。即使技术发展像我预期的那样快,我们也不确切知道它会以多快的速度带动收入增长。

We know it's coming, but with the way you buy these data centers, if you're off by a couple years, that can be ruinous. It is just like how I wrote in "Machines of Loving Grace."

我们知道它会到来,但以购买这些数据中心的方式来看,如果时间偏差几年,可能是灾难性的。就像我在「Machines of Loving Grace」里写的那样。

I said I think we might get this powerful AI, this "country of genius in the data center." That description you gave comes from "Machines of Loving Grace." I said we'll get that in 2026, maybe 2027. Again, that is my hunch. I wouldn't be surprised if I'm off by a year or two, but that is my hunch.

我说过我认为我们可能会得到这种强大的 AI,这个「数据中心里的天才之国」。你提到的那个描述就来自「Machines of Loving Grace」。我说过我们会在 2026 年得到它,也许是 2027 年。再次强调,这是我的直觉。如果偏差一两年我也不会惊讶,但这就是我的直觉。

Let's say that happens. That's the starting gun. How long does it take to cure all the diseases? That's one of the ways that drives a huge amount of economic value. You cure every disease. There's a question of how much of that goes to the pharmaceutical company or the AI company, but there's an enormous consumer surplus because—assuming we can get access for everyone, which I care about greatly—we cure all of these diseases.

假设那真的发生了,那就是发令枪响。治愈所有疾病需要多长时间?这是带动巨大经济价值的方式之一。你治愈了所有疾病。有个问题是这些价值有多少归制药公司还是 AI 公司,但会产生巨大的消费者剩余,因为——假设我们能让所有人都获得,这是我非常关心的——我们治愈了所有这些疾病。

How long does it take? You have to do the biological discovery, you have to manufacture the new drug, you have to go through the regulatory process. We saw this with vaccines and COVID. We got the vaccine out to everyone, but it took a year and a half.

这需要多长时间?你必须进行生物学发现,必须生产新药,必须通过监管流程。我们在疫苗和 COVID 上见识过这个过程。我们把疫苗推广到所有人,但花了一年半时间。

My question is: How long does it take to get the cure for everything—which AI is the genius that can in theory invent—out to everyone? How long from when that AI first exists?

我的问题是:把治愈一切疾病的方法——AI 这个天才理论上能发明出来——推广到所有人需要多长时间?从那个 AI 首次出现开始算起要多久?

In the lab to when diseases have actually been cured for everyone?

从实验室研发出疗法,到疾病真正在所有人群中被治愈,需要多久?

We've had a polio vaccine for 50 years. We're still trying to eradicate it in the most remote corners of Africa.

我们有小儿麻痹症疫苗已经50年了,但我们仍在努力在非洲最偏远的角落根除这种疾病。

The Gates Foundation is trying as hard as they can. Others are trying as hard as they can. But that's difficult.

Gates 基金会在竭尽全力,其他机构也在竭尽全力,但这很困难。

Again, I don't expect most of the economic diffusion to be as difficult as that. That's the most difficult case. But there's a real dilemma here.

当然,我不认为大多数经济扩散会像那样困难。那是最困难的情况。但这里确实存在一个真实的困境。

Where I've settled on it is that it will be faster than anything we've seen in the world, but it still has its limits.

我得出的结论是,它的扩散速度会比我们在世界上见过的任何事物都快,但仍然有其局限性。

So when we go to buying data centers, again, the curve I'm looking at is we've had a 10x a year increase every year.

所以当我们考虑购买数据中心时,我看到的增长曲线是:我们每年都有10倍的年增长。

At the beginning of this year, we're looking at $10 billion in annualized revenue.

今年年初,我们的年化收入是100亿美元。

We have to decide how much compute to buy. It takes a year or two to actually build out the data centers, to reserve the data center.

我们必须决定购买多少算力。实际建设数据中心、预留数据中心需要一到两年的时间。

Basically I'm saying, in 2027, how much compute do I get?

基本上我在问:2027年,我能获得多少算力?

I could assume that the revenue will continue growing 10x a year, so it'll be $100 billion at the end of 2026 and $1 trillion at the end of 2027.

我可以假设收入继续以每年10倍的速度增长,那么2026年底会是1000亿美元,2027年底会是1万亿美元。

Actually it would be $5 trillion dollars of compute because it would be $1 trillion a year for five years.

实际上需要5万亿美元的算力,因为那是每年1万亿美元,持续五年。

I could buy $1 trillion of compute that starts at the end of 2027.

我可以购买1万亿美元的算力,在2027年底开始启用。

If my revenue is not $1 trillion dollars, if it's even $800 billion, there's no force on earth, there's no hedge on earth that could stop me from going bankrupt if I buy that much compute.

如果我的收入不是1万亿美元,哪怕是8000亿美元,如果我购买那么多算力,世界上没有任何力量,没有任何对冲手段能阻止我破产。

Even though a part of my brain wonders if it's going to keep growing 10x, I can't buy $1 trillion a year of compute in 2027.

尽管我脑子里有一部分在想它是否会继续保持10倍增长,但我不能在2027年购买每年1万亿美元的算力。

If I'm just off by a year in that rate of growth, or if the growth rate is 5x a year instead of 10x a year, then you go bankrupt.

如果我对增长速度的判断哪怕只偏差一年,或者增长率是每年5倍而不是10倍,那你就会破产。

So you end up in a world where you're supporting hundreds of billions, not trillions.

所以你最终会处于一个支持数千亿美元而非数万亿美元的世界。

You accept some risk that there's so much demand that you can't support the revenue, and you accept some risk that you got it wrong and it's still slow.

你要接受一些风险,即需求太大以至于你无法支撑收入增长,同时也要接受一些风险,即你判断错了而增长仍然缓慢。

When I talked about behaving responsibly, what I meant actually was not the absolute amount.

当我谈到负责任的行为时,我实际指的不是绝对金额。

I think it is true we're spending somewhat less than some of the other players.

确实,我们的支出比其他一些参与者要少一些。

It's actually the other things, like have we been thoughtful about it or are we YOLOing and saying, we're going to do $100 billion here or $100 billion there?

实际上是其他方面,比如我们是否经过深思熟虑,还是在随意冒险说「我们要在这里投1000亿美元或在那里投1000亿美元」?

I get the impression that some of the other companies have not written down the spreadsheet, that they don't really understand the risks they're taking.

我感觉其他一些公司并没有做详细的电子表格分析,他们并不真正理解自己承担的风险。

They're just doing stuff because it sounds cool. We've thought carefully about it.

他们只是因为听起来很酷就去做。我们则经过了仔细思考。

We're an enterprise business. Therefore, we can rely more on revenue. It's less fickle than consumer.

我们是企业业务,因此可以更依赖收入。它比消费者业务更稳定,不那么善变。

We have better margins, which is the buffer between buying too much and buying too little.

我们有更好的利润率,这是购买过多和购买过少之间的缓冲。

I think we bought an amount that allows us to capture pretty strong upside worlds.

我认为我们购买的数量能让我们抓住相当强劲的上行空间。

It won't capture the full 10x a year. Things would have to go pretty badly for us to be in financial trouble.

它无法完全捕捉每年10倍的增长。但情况得变得相当糟糕,我们才会陷入财务困境。

So we've thought carefully and we've made that balance. That's what I mean when I say that we're being responsible.

所以我们经过仔细思考并做了这种平衡。这就是我说我们负责任的意思。

So it seems like it's possible that we actually just have different definitions of the country of a genius in a data center.

所以看起来我们可能对「数据中心里的天才之国」有不同的定义。

Because when I think of actual human geniuses, an actual country of human geniuses in a data center,

因为当我想到真正的人类天才,想到数据中心里有一整个国家的人类天才时,

I would happily buy $5 trillion worth of compute to run an actual country of human geniuses in a data center.

我会非常乐意花5万亿美元购买算力来运行这样一个数据中心里的天才国度。

Let's say JPMorgan or Moderna or whatever doesn't want to use them.

假设 JPMorgan 或 Moderna 或其他公司不想使用他们。

I've got a country of geniuses.

我有一整个国家的天才啊。

They'll start their own company. If they can't start their own company and they're bottlenecked by clinical trials…

他们可以自己创办公司。如果他们无法创办自己的公司,又受限于临床试验……

It is worth stating that with clinical trials, most clinical trials fail because the drug doesn't work. There's no efficacy.

值得说明的是,对于临床试验,大多数临床试验失败是因为药物本身无效,没有疗效。

I make exactly that point in "Machines of Loving Grace". I say the clinical trials are going to go much faster than we're used to, but not infinitely fast.

我在「Machines of Loving Grace」中正是提出了这一点,我说临床试验会比我们习惯的速度快得多,但不会无限快。

Okay, and then suppose it takes a year for the clinical trials to work out so that you're getting revenue from that and can make more drugs.

好的,那么假设临床试验需要一年时间才能完成,这样你才能从中获得收入并制造更多药物。

Okay, well, you've got a country of geniuses and you're an AI lab.

好吧,你有一整个国家的天才,而且你是一家 AI 实验室。

You could use many more AI researchers.

你可以使用更多 AI 研究人员。

You also think there are these self-reinforcing gains from smart people working on AI tech.

你也认为聪明人研究 AI 技术会产生这种自我强化的收益。

You can have the data center working on AI progress.

你可以让数据中心致力于推进 AI 的发展。

Are there substantially more gains from buying $1 trillion a year of compute versus $300 billion a year of compute?

每年购买1万亿美元的算力相比每年3000亿美元的算力,是否会带来显著更多的收益?

If your competitor is buying a trillion, yes there is.

如果你的竞争对手购买了1万亿,那确实会有。

Well, no, there's some gain, but then again, there's this chance that they go bankrupt before.

嗯,不,会有一些收益,但话说回来,也有他们在此之前就破产的风险。

Again, if you're off by only a year, you destroy yourselves. That's the balance.

再说一遍,如果你仅仅偏差一年,就会毁掉自己。这就是需要平衡的地方。

We're buying a lot. We're buying a hell of a lot.

我们买了很多。我们买了非常多。

We're buying an amount that's comparable to what the biggest players in the game are buying.

我们购买的规模与这个领域最大玩家的购买量相当。

But if you're asking me, "Why haven't we signed $10 trillion of compute starting in mid-2027?"...

但如果你问我「为什么我们没有签约从2027年中开始的10万亿美元算力?」……

First of all, it can't be produced. There isn't that much in the world.

首先,这根本生产不出来。世界上没有那么多。

But second, what if the country of geniuses comes, but it comes in mid-2028 instead of mid-2027? You go bankrupt.

其次,如果天才国度确实到来,但它是在2028年中而不是2027年中到来呢?你就会破产。

So if your projection is one to three years, it seems like you should want $10 trillion of compute by 2029 at the latest?

所以如果你的预测是一到三年,似乎你应该最迟在2029年前想要10万亿美元的算力?

Even in the longest version of the timelines you state, the compute you are ramping up to build doesn't seem in accordance.

即使按照你所说的最长时间线版本,你正在扩建的算力规模似乎也不相符。

What makes you think that?

你为什么这么认为?

Human wages, let's say, are on the order of $50 trillion a year—

比如说,人类工资总额大约是每年50万亿美元量级——

So I won't talk about Anthropic in particular, but if you talk about the industry, the amount of compute the industry is building this year is probably, call it, 10-15 gigawatts.

我不具体谈 Anthropic,但如果说整个行业,今年行业正在建设的算力规模大概是,姑且说,10到15吉瓦。

It goes up by roughly 3x a year. So next year's 30-40 gigawatts. 2028 might be 100 gigawatts. 2029 might be like 300 gigawatts.

每年大约增长3倍。所以明年是30到40吉瓦。2028年可能是100吉瓦。2029年可能会达到300吉瓦左右。

I'm doing the math in my head, but each gigawatt costs maybe $10 billion, on the order of $10-15 billion a year.

我在心算,但每吉瓦的成本大约是100亿美元,大概每年在100到150亿美元的量级。

You put that all together and you're getting about what you described. You're getting exactly that. You're getting multiple trillions a year by 2028 or 2029.

你把这些加在一起,就得到了你所描述的数字。完全就是那样。到2028或2029年,每年会达到数万亿美元。

You're getting exactly what you predict. That's for the industry.

这正是你预测的。这是整个行业的数据。

That's for the industry, that's right. Suppose Anthropic's compute keeps 3x-ing a year, and then by 2027-28, you have 10 gigawatts.

那是整个行业的,没错。假设 Anthropic 的算力持续每年3倍增长,那么到2027-28年,你们会有10吉瓦。

Multiply that by, as you say, $10 billion. So then it's like $100 billion a year.

按你说的乘以100亿美元。那么每年就是大约1000亿美元。

But then you're saying the TAM by 2028 is $200 billion.

但你又说到2028年的 TAM 是2000亿美元。

Again, I don't want to give exact numbers for Anthropic, but these numbers are too small.

再说一次,我不想给出 Anthropic 的确切数字,但这些数字都太小了。

Okay, interesting. You've told investors that you plan to be profitable starting in 2028.

好的,很有意思。你告诉投资者,你们计划从2028年开始实现盈利。

This is the year when we're potentially getting the country of geniuses as a data center.

这一年我们可能会拥有相当于一个天才之国规模的数据中心。

This is now going to unlock all this progress in medicine and health and new technologies.

这将释放医学、健康和新技术领域的大量进展。

Wouldn't this be exactly the time where you'd want to reinvest in the business and build bigger "countries" so they can make more discoveries?

这难道不正是你们想要再投资于业务、建造更大的「国家」以实现更多发现的时候吗?

Profitability is this kind of weird thing in this field.

在这个领域,盈利能力是一个有点奇怪的概念。

I don't think in this field profitability is actually a measure of spending down versus investing in the business.

我认为在这个领域,盈利能力实际上并不是衡量减少开支和投资业务之间取舍的指标。

Let's just take a model of this.

我们来建立一个模型。

I actually think profitability happens when you underestimated the amount of demand you were going to get, and loss happens when you overestimated the amount of demand you were going to get, because you're buying the data centers ahead of time.

我实际上认为,盈利发生在你低估了需求量的时候,而亏损发生在你高估了需求量的时候,因为你是提前购买数据中心的。

Think about it this way. Again, these are stylized facts. These numbers are not exact. I'm just trying to make a toy model here.

这样想吧。再说一遍,这些是简化的假设。这些数字并不精确。我只是在这里构建一个简单模型。

Let's say half of your compute is for training and half of your compute is for inference.

假设你一半的算力用于训练,一半用于推理。

The inference has some gross margin that's more than 50%.

推理部分有超过50%的毛利率。

So what that means is that if you were in steady state, you build a data center and if you knew exactly the demand you were getting, you would get a certain amount of revenue.

这意味着,如果你处于稳定状态,建造了一个数据中心,并且准确知道你会获得的需求量,你就会获得一定数量的收入。

Let's say you pay $100 billion a year for compute. On $50 billion a year you support $150 billion of revenue.

假设你每年为算力支付1000亿美元。其中500亿美元支撑1500亿美元的收入。

The other $50 billion is used for training.

另外500亿美元用于训练。

Basically you're profitable and you make $50 billion of profit.

基本上你是盈利的,可以赚500亿美元利润。

Those are the economics of the industry today, or not today but where we're projecting forward in a year or two.

这就是这个行业的经济状况,或者说不是现在,而是我们对未来一两年的预测。

The only thing that makes that not the case is if you get less demand than $50 billion.

唯一使情况不同的是,如果你获得的需求少于500亿美元。

Then you have more than 50% of your data center for research and you're not profitable.

那么你就会有超过50%的数据中心用于研究,而你就不盈利了。

So you train stronger models, but you're not profitable.

所以你训练更强大的模型,但你不盈利。

If you get more demand than you thought, then research gets squeezed, but you're kind of able to support more inference and you're more profitable.

如果你获得的需求比预期多,那么研究就会被压缩,但你可以支撑更多推理,从而更加盈利。

Maybe I'm not explaining it well, but the thing I'm trying to say is that you decide the amount of compute first.

也许我没解释清楚,但我想说的是,你首先决定算力的数量。

Then you have some target desire of inference versus training, but that gets determined by demand.

然后你对推理和训练的比例有一些目标期望,但这是由需求决定的。

It doesn't get determined by you.

不是由你决定的。

What I'm hearing is the reason you're predicting profit is that you are systematically underinvesting in compute?

我听到的是,你预测盈利的原因是你在系统性地对算力投资不足?

No, no, no. I'm saying it's hard to predict.

不不不。我是说这很难预测。

These things about 2028 and when it will happen, that's our attempt to do the best we can with investors.

关于2028年以及何时实现的这些事情,是我们尽力向投资者给出的最佳预测。

All of this stuff is really uncertain because of the cone of uncertainty.

所有这些事情都非常不确定,因为存在不确定性锥。

We could be profitable in 2026 if the revenue grows fast enough.

如果收入增长足够快,我们可能在2026年就能盈利。

If we overestimate or underestimate the next year, that could swing wildly.

如果我们高估或低估了明年的情况,结果可能会大幅波动。

What I'm trying to get at is that you have a model in your head of a business that invests, invests, invests, gets scale and then becomes profitable.

我想说的是,你脑海中有一个商业模型,就是不断投资、投资、投资,获得规模,然后实现盈利。

There's a single point at which things turn around.

有一个单一的转折点。

I don't think the economics of this industry work that way.

我认为这个行业的经济逻辑不是那样运作的。

I see. So if I'm understanding correctly, you're saying that because of the discrepancy...

我明白了。所以如果我理解正确的话,你是说因为这种差异...

Between the amount of compute we should have gotten and the amount of compute we got, we were sort of forced to make profit. But that doesn't mean we're going to continue making profit. We're going to reinvest the money

我们应该获得的算力和实际获得的算力之间存在差距,这在某种程度上迫使我们盈利。但这并不意味着我们会继续保持盈利。我们会把这些钱再投资出去

because now AI has made so much progress and we want a bigger country of geniuses. So revenue is high, but losses are also high.

因为现在AI已经取得了很大进展,我们想要一个更大的天才国度。所以收入很高,但亏损也很高。

If every year we predict exactly what the demand is going to be, we'll be profitable every year.

如果我们每年都能准确预测需求会是多少,那我们每年都会盈利。

Because spending 50% of your compute on research, roughly, plus a gross margin that's higher than 50% and correct demand prediction leads to profit. That's the profitable business model that I think is kind of there, but obscured by these building ahead and prediction errors.

因为大约把50%的算力用于研究,加上高于50%的毛利率,再加上准确的需求预测,就能实现盈利。我认为这个盈利的商业模式其实是存在的,只是被提前建设和预测误差所掩盖了。

I guess you're treating the 50% as a sort of given constant, whereas in fact, if AI progress is fast and you can increase the progress by scaling up more, you should just have more than 50% and not make profit.

我想你是把这50%当作一个既定的常数,但实际上,如果AI进展很快,而且你可以通过加大投入来加速进展,那你就应该投入超过50%,而不是追求盈利。

But here's what I'll say. You might want to scale it up more. Remember the log returns to scale.

但我要说的是这个。你可能想要进一步扩大规模。记住对数收益规律。

If 70% would get you a very little bit of a smaller model through a factor of 1.4x... That extra $20 billion, each dollar there is worth much less to you because of the log-linear setup.

如果投入70%只能让你获得一个稍小一点的模型,通过1.4倍的因子... 那额外的200亿美元,由于对数线性的设定,每一美元对你的价值都要小得多。

So you might find that it's better to invest that $20 billion in serving inference or in hiring engineers who are better at what they're doing.

所以你可能会发现,把这200亿美元投资在推理服务上,或者招聘更优秀的工程师,会是更好的选择。

So the reason I said 50%... That's not exactly our target. It's not exactly going to be 50%. It'll probably vary over time. What I'm saying is the log-linear return, what it leads to is you spend of order one fraction of the business. Like not 5%, not 95%. Then you get diminishing returns.

所以我说50%的原因是... 这并不完全是我们的目标。不会正好是50%。它可能会随时间变化。我想说的是,对数线性收益导致的结果是,你会花费业务中相当大的一部分。比如不是5%,也不是95%。然后你就会遇到收益递减。

I feel strange that I'm convincing Dario to believe in AI progress or something.

我觉得很奇怪,我竟然在说服Dario相信AI进展之类的。

Okay, you don't invest in research because it has diminishing returns, but you invest in the other things you mentioned.

好吧,你不投资研究是因为它有收益递减,但你会投资你提到的其他事情。

I think profit at a sort of macro level— Again, I'm talking about diminishing returns, but after you're spending $50 billion a year.

我认为在宏观层面上的盈利—— 再说一次,我在谈收益递减,但那是在你每年花费500亿美元之后。

This is a point I'm sure you would make, but diminishing returns on a genius could be quite high. More generally,

这一点我确信你也会提出,但天才的收益递减点可能相当高。更广泛地说,

what is profit in a market economy? Profit is basically saying other companies in the market can do more things with this money than I can.

在市场经济中利润是什么?利润基本上是在说,市场上的其他公司能用这笔钱做比我更多的事情。

Put aside Anthropic. I don't want to give information about Anthropic. That's why I'm giving these stylized numbers. But let's just derive the equilibrium of the industry.

抛开Anthropic不谈。我不想透露Anthropic的信息。这就是为什么我在用这些简化的数字。但让我们来推导一下这个行业的均衡状态。

Why doesn't everyone spend 100% of their compute on training and not serve any customers? It's because if they didn't get any revenue, they couldn't raise money, they couldn't do compute deals, they couldn't buy more compute the next year.

为什么不是每个人都把100%的算力用于训练而不服务任何客户?因为如果他们没有任何收入,就无法融资,无法达成算力交易,无法在第二年购买更多算力。

So there's going to be an equilibrium where every company spends less than 100% on training and certainly less than 100% on inference.

所以会存在一个均衡点,每家公司在训练上的投入都会少于100%,在推理上的投入当然也少于100%。

It should be clear why you don't just serve the current models and never train another model, because then you don't have any demand because you'll fall behind. So there's some equilibrium. It's not gonna be 10%, it's not gonna be 90%. Let's just say as a stylized fact, it's 50%.

应该很清楚为什么你不能只提供当前模型而永远不训练新模型,因为那样你就会落后,就不会有任何需求。所以存在某个均衡点。它不会是10%,也不会是90%。作为一个简化的事实,我们就说它是50%。

That's what I'm getting at. I think we're gonna be in a position where that equilibrium of how much you spend on training is less than the gross margins that you're able to get on compute.

这就是我想说的。我认为我们会处于这样一个位置:你在训练上花费的均衡比例会低于你在算力上能够获得的毛利率。

So the underlying economics are profitable. The problem is you have this hellish demand prediction problem when you're buying the next year of compute and you might guess under and be very profitable but have no compute for research.

所以底层的经济模型是盈利的。问题在于,当你购买下一年的算力时,会面临一个极其棘手的需求预测问题,你可能会低估需求,结果是非常盈利但没有算力用于研究。

Or you might guess over and you are not profitable and you have all the compute for research in the world. Does that make sense?

或者你可能会高估需求,结果是不盈利,但拥有世界上所有可用于研究的算力。这样说清楚了吗?

Just as a dynamic model of the industry?

作为这个行业的动态模型来理解?

Maybe stepping back, I'm not saying I think the "country of geniuses" is going to come in two years and therefore you should buy this compute.

也许退一步说,我并不是说我认为「天才之国」会在两年内到来,因此你应该购买这些算力。

To me, the end conclusion you're arriving at makes a lot of sense.

对我来说,你得出的最终结论很有道理。

But that's because it seems like "country of geniuses" is hard and there's a long way to go.

但那是因为「天才之国」似乎很难实现,还有很长的路要走。

So stepping back, the thing I'm trying to get at is more that it seems like your worldview is compatible with somebody who says, "We're like 10 years away from a world in which we're generating trillions of dollars of value."

所以退一步说,我想要表达的更多是,你的世界观似乎与那些认为「我们距离创造数万亿美元价值的世界还有大约10年」的人是一致的。

That's just not my view. So I'll make another prediction. It is hard for me to see that there won't be trillions of dollars in revenue before 2030.

那不是我的观点。所以我再做一个预测。我很难想象在2030年之前不会有数万亿美元的收入。

I can construct a plausible world.

我可以构想一个合理的情景。

It takes maybe three years. That would be the end of what I think is plausible.

可能需要三年时间。那将是我认为合理的最长时限。

Like in 2028, we get the real "country of geniuses in the data center".

比如在2028年,我们实现了真正的「数据中心里的天才之国」。

The revenue's going into the low hundreds of billions by 2028, and then the country of geniuses accelerates it to trillions.

到2028年收入达到数千亿美元的低端,然后天才之国将其加速到数万亿。

We're basically on the slow end of diffusion.

我们基本上处于扩散的慢速端。

It takes two years to get to the trillions. That would be the world where it takes until 2030.

需要两年时间达到数万亿。那就是要到2030年的情景。

I suspect even composing the technical exponential and diffusion exponential, we'll get there before 2030.

我怀疑即使叠加技术指数增长和扩散指数增长,我们也会在2030年之前达到那个目标。

So you laid out a model where Anthropic makes profit because it seems like fundamentally we're in a compute-constrained world.

所以你阐述了一个模型,在这个模型中Anthropic能够盈利,因为从根本上看我们处在一个算力受限的世界。

So eventually we keep growing compute—

所以最终我们持续增长算力——

I think the way the profit comes is... Again, let's just abstract the whole industry here. Let's just imagine we're in an economics textbook.

我认为利润产生的方式是……再次强调,让我们把整个行业抽象化。让我们想象一下我们在经济学教科书里。

We have a small number of firms. Each can invest a limited amount.

我们有少数几家公司。每家都只能投资有限的金额。

Each can invest some fraction in R&D. They have some marginal cost to serve.

每家都可以投资一定比例在研发上。它们都有一定的边际服务成本。

The gross profit margins on that marginal cost are very high because inference is efficient.

基于边际成本的毛利率非常高,因为推理是高效的。

There's some competition, but the models are also differentiated.

存在一些竞争,但模型之间也有差异化。

Companies will compete to push their research budgets up.

公司会竞相提高它们的研究预算。

But because there's a small number of players, we have the... What is it called?

但由于参与者数量少,我们有那个……叫什么来着?

The Cournot equilibrium, I think, is what the small number of firm equilibrium is.

Cournot均衡,我想,就是少数企业均衡的那个概念。

The point is it doesn't equilibrate to perfect competition with zero margins.

重点是它不会达到零利润率的完全竞争均衡。

If there's three firms in the economy and all are kind of independently behaving rationally, it doesn't equilibrate to zero.

如果经济中有三家公司,并且都各自独立地理性行事,它不会达到零利润的均衡。

Help me understand that, because right now we do have three leading firms and they're not making profit. So what is changing?

帮我理解一下,因为现在我们确实有三家领先公司,但它们并没有盈利。那么是什么在改变?

Again, the gross margins right now are very positive.

再次强调,目前的毛利率是非常正向的。

What's happening is a combination of two things. One is that we're still in the exponential scale-up phase of compute. A model gets trained. Let's say a model got trained that cost $1 billion last year.

正在发生的是两件事的组合。一是我们仍然处于算力的指数扩张阶段。一个模型被训练出来。假设去年训练了一个花费10亿美元的模型。

Then this year it produced $4 billion of revenue and cost $1 billion to inference from. Again, I'm using stylized numbers here, but that would be 75% gross margins and this 25% tax.

然后今年它产生了40亿美元的收入,推理成本是10亿美元。再次强调,我这里用的是简化的数字,但那将是75%的毛利率和25%的成本占比。

So that model as a whole makes $2 billion.

所以这个模型整体赚了20亿美元。

But at the same time, we're spending $10 billion to train the next model because there's an exponential scale-up. So the company loses money. Each model makes money, but the company loses money.

但与此同时,我们要花100亿美元来训练下一个模型,因为存在指数级的扩张。所以公司是亏损的。每个模型本身都赚钱,但公司整体是亏损的。

The equilibrium I'm talking about is an equilibrium where we have the "country of geniuses in a data center", but that model training scale-up has equilibrated more. Maybe it's still going up. We're still trying to predict the demand, but it's more leveled out.

我所说的均衡状态是指,我们已经拥有了「数据中心里的天才国度」,但模型训练的规模扩张已经趋于平衡。也许仍在增长,我们还在努力预测需求,但增速已经趋于平稳了。

I'm confused about a couple of things there. Let's start with the current world.

我对其中几点有些困惑。我们先从当前的情况说起吧。

In the current world, you're right that, as you said before, if you treat each individual model as a company, it's profitable.

在当前的情况下,你说得对,正如你之前所说,如果把每个单独的模型看作一家公司,它是盈利的。

But of course, a big part of the production function of being a frontier lab is training the next model, right?

但显然,作为前沿实验室,生产函数的重要组成部分就是训练下一个模型,对吧?

Yes, that's right.

是的,没错。

If you didn't do that, then you'd make profit for two months and then you wouldn't have margins because you wouldn't have the best model.

如果不这样做,那你只能盈利两个月,然后就没有利润空间了,因为你不再拥有最好的模型。

But at some point that reaches the biggest scale that it can reach. And then in equilibrium, we have algorithmic improvements, but we're spending roughly the same amount to train the next model as we spend to train the current model.

但在某个时刻,规模会达到它能达到的最大值。然后在均衡状态下,我们仍有算法改进,但训练下一个模型的花费大致和训练当前模型的花费相当。

At some point you run out of money in the economy.

到某个时刻,经济体中的资金就会耗尽。

A fixed lump of labor fallacy… The economy is going to grow, right? That's one of your predictions. We're going to have the data centers in space.

固定劳动总量谬误啊……经济会增长的,对吧?这是你的预测之一。我们会在太空建数据中心。

Yes, but this is another example of the theme I was talking about.

是的,但这又是我刚才讨论的主题的一个例子。

The economy will grow much faster with AI than I think it ever has before. Right now the compute is growing 3x a year.

有了AI,经济增长速度会比以往任何时候都快得多。现在算力每年增长3倍。

I don't believe the economy is gonna grow 300% a year.

我不认为经济会每年增长300%。

I said this in "Machines of Loving Grace", I think we may get 10-20% per year growth in the economy, but we're not gonna get 300% growth in the economy.

我在《Machines of Loving Grace》中说过,我认为经济可能实现每年10-20%的增长,但不会达到300%的增长。

So I think in the end, if compute becomes the majority of what the economy produces, it's gonna be capped by that.

所以我认为最终,如果算力成为经济产出的主要部分,它就会受到这个上限的制约。

So let's assume a model where compute stays capped.

那我们假设一个算力保持上限的模型。

The world where frontier labs are making money is one where they continue to make fast progress. Because fundamentally your margin is limited by how good the alternative is.

前沿实验室能赚钱的世界,是它们持续快速进步的世界。因为从根本上说,你的利润空间受限于替代方案有多好。

So you are able to make money because you have a frontier model. If you didn't have a frontier model you wouldn't be making money.

所以你能赚钱是因为你有前沿模型。如果你没有前沿模型,你就赚不到钱。

So this model requires there never to be a steady state. Forever and ever you keep making more algorithmic progress.

所以这个模型要求永远不会有稳态。你要永远不断地取得更多算法进步。

I don't think that's true. I mean, I feel like we're in an economics class.

我不认为是这样。我是说,我感觉我们像在上经济学课。

Do you know the Tyler Cowen quote?

你知道Tyler Cowen的那句话吗?

We never stop talking about economics.

我们永远在谈论经济学。

We never stop talking about economics.

我们永远在谈论经济学。

So no, I don't think this field's going to be a monopoly.

所以不,我不认为这个领域会成为垄断。

All my lawyers never want me to say the word "monopoly".

我所有的律师都不想让我说「垄断」这个词。

But I don't think this field's going to be a monopoly.

但我不认为这个领域会成为垄断。

You do get industries in which there are a small number of players. Not one, but a small number of players.

确实有些行业只有少数玩家。不是一个,而是少数几个玩家。

Ordinarily, the way you get monopolies like Facebook or Meta—I always call them Facebook—is these kinds of network effects.

通常来说,像Facebook或Meta——我总是叫它们Facebook——这样的垄断形成方式是通过网络效应。

The way you get industries in which there are a small number of players is very high costs of entry.

一个行业之所以只有少数几个参与者,是因为进入成本非常高。

Cloud is like this. I think cloud is a good example of this. There are three, maybe four players within cloud. I think that's the same for AI, three, maybe four.

云计算就是这样。我认为云计算是一个很好的例子。云计算领域有三家,也许四家参与者。我认为 AI 也一样,三家,也许四家。

The reason is that it's so expensive. It requires so much expertise and so much capital to run a cloud company. You have to put up all this capital.

原因是成本太高了。运营一家云计算公司需要大量的专业知识和资本。你必须投入所有这些资本。

In addition to putting up all this capital, you have to get all of this other stuff that requires a lot of skill to make it happen.

除了投入所有这些资本之外,你还需要掌握许多其他需要大量技能才能实现的东西。

So if you go to someone and you're like, 'I want to disrupt this industry, here's $100 billion.' You're like, 'Okay, I'm putting in $100 billion and also betting that you can do all these other things that these people have been doing.'

所以如果你去找某个人说「我想颠覆这个行业,这里有 1000 亿美元」。你会想「好吧,我投入 1000 亿美元,同时还要赌你能做到这些人一直在做的所有其他事情」。

Only to decrease the profit. The effect of your entering is that profit margins go down.

结果只是降低了利润。你进入市场的效果就是利润率下降。

So we have equilibria like this all the time in the economy where we have a few players. Profits are not astronomical. Margins are not astronomical, but they're not zero.

所以在经济中我们经常看到这样的均衡,只有少数几个参与者。利润不是天文数字,利润率也不是天文数字,但也不是零。

That's what we see on cloud. Cloud is very undifferentiated. Models are more differentiated than cloud.

这就是我们在云计算上看到的。云计算产品差异化程度很低。模型的差异化程度比云计算高。

Everyone knows Claude is good at different things than GPT is good at, than Gemini is good at.

每个人都知道 Claude 擅长的事情与 GPT 擅长的不同,与 Gemini 擅长的也不同。

It's not just that Claude's good at coding, GPT is good at math and reasoning.

不仅仅是 Claude 擅长编程、GPT 擅长数学和推理这么简单。

It's more subtle than that. Models are good at different types of coding. Models have different styles. I think these things are actually quite different from each other, and so I would expect more differentiation than you see in cloud.

比这更微妙。模型擅长不同类型的编程。模型有不同的风格。我认为这些东西实际上彼此差异很大,所以我预期会有比云计算更高的差异化。

Now, there actually is one counter-argument.

不过,确实有一个反驳论点。

That counter-argument is if the process of producing models, if AI models can do that themselves, then that could spread throughout the economy.

这个反驳论点是,如果生产模型的过程,如果 AI 模型自己能做到这一点,那么这种能力可能会扩散到整个经济中。

But that is not an argument for commoditizing AI models in general.

但这并不是一个让 AI 模型普遍商品化的论据。

That's kind of an argument for commoditizing the whole economy at once.

这更像是一个让整个经济同时商品化的论据。

I don't know what quite happens in that world where basically anyone can do anything, anyone can build anything, and there's no moat around anything at all.

我不太清楚在那个世界会发生什么,在那里基本上任何人都能做任何事,任何人都能构建任何东西,而且任何东西周围都没有护城河。

I don't know, maybe we want that world. Maybe that's the end state here.

我不知道,也许我们想要那样的世界。也许那就是最终状态。

Maybe when AI models can do everything, if we've solved all the safety and security problems, that's one of the mechanisms for the economy just flattening itself again.

也许当 AI 模型能做所有事情时,如果我们已经解决了所有安全和保障问题,那就是经济再次自我扁平化的机制之一。

But that's kind of far post-'country of geniuses in the data center.'

但那是在「数据中心里的天才之国」之后很远的事了。

Maybe a finer way to put that potential point is: 1) it seems like AI research is especially loaded on raw intellectual power, which will be especially abundant in the world of AGI.

也许更精确地表达这个潜在观点是:1) AI 研究似乎特别依赖原始智力,而这在 AGI 的世界里将特别丰富。

And 2) if you just look at the world today, there are very few technologies that seem to be diffusing as fast as AI algorithmic progress. So that does hint that this industry is sort of structurally diffusive.

2) 如果你只看今天的世界,很少有技术的扩散速度能像 AI 算法进步那样快。所以这确实暗示这个行业在结构上就具有扩散性。

I think coding is going fast, but I think AI research is a superset of coding and there are aspects of it that are not going fast.

我认为编程进展很快,但我认为 AI 研究是编程的超集,它有些方面进展并不快。

But I do think, again, once we get coding, once we get AI models going fast, then that will speed up the ability of AI models to do everything else.

但我确实认为,再说一次,一旦我们掌握了编程,一旦 AI 模型进展快速,那将加快 AI 模型做其他所有事情的能力。

So while coding is going fast now, I think once the AI models are building the next AI models and building everything else, the whole economy will kind of go at the same pace.

所以虽然现在编程进展很快,但我认为一旦 AI 模型在构建下一代 AI 模型并构建其他所有东西,整个经济都会以相同的速度发展。

I am worried geographically, though. I'm a little worried that just proximity to AI, having heard about AI, may be one differentiator.

不过我从地理角度有些担忧。我有点担心仅仅是接近 AI、听说过 AI,可能就会成为一个差异化因素。

So when I said the 10-20% growth rate, a worry I have is that the growth rate could be like 50% in Silicon Valley and parts of the world that are socially connected to Silicon Valley, and not that much faster than its current pace elsewhere.

所以当我说到10-20%的增长率时,我担心的是增长率可能在硅谷以及与硅谷有社会联系的世界其他地区达到50%,而在其他地方并不会比目前的速度快多少。

I think that'd be a pretty messed up world. So one of the things I think about a lot is how to prevent that.

我觉得那会是一个相当糟糕的世界。所以我经常思考的一件事就是如何防止这种情况发生。

Do you think that once we have this country of geniuses in a data center, that robotics is sort of quickly solved afterwards? Because it seems like a big problem with robotics is that a human can learn how to teleoperate current hardware, but current AI models can't, at least not in a way that's super productive.

你认为一旦我们在数据中心拥有这个天才之国,机器人技术会很快被解决吗?因为机器人技术的一个大问题似乎是人类可以学会远程操作现有硬件,但目前的AI模型做不到,至少不能以一种超高效的方式做到。

And so if we have this ability to learn like a human, shouldn't it solve robotics immediately as well?

所以如果我们拥有像人类一样学习的能力,难道不应该也能立即解决机器人技术吗?

I don't think it's dependent on learning like a human. It could happen in different ways.

我不认为这取决于像人类那样学习。它可以通过不同的方式实现。

Again, we could have trained the model on many different video games, which are like robotic controls, or many different simulated robotics environments, or just train them to control computer screens, and they learn to generalize.

再说一次,我们可以在许多不同的视频游戏上训练模型,这些游戏就像机器人控制一样,或者在许多不同的模拟机器人环境中训练,或者只是训练它们控制计算机屏幕,然后它们学会泛化。

So it will happen... it's not necessarily dependent on human-like learning. Human-like learning is one way it could happen.

所以这会实现的……它不一定依赖于类人的学习方式。类人学习只是实现它的一种方式。

If the model's like, "Oh, I pick up a robot, I don't know how to use it, I learn," that could happen because we discovered continual learning.

如果模型能够说「哦,我拿起一个机器人,我不知道怎么用,然后我学会了」,那可能是因为我们发现了持续学习。

That could also happen because we trained the model on a bunch of environments and then generalized, or it could happen because the model learns that in the context length.

这也可能是因为我们在大量环境中训练模型然后它泛化了,或者可能是因为模型在上下文窗口中学会了这一点。

It doesn't actually matter which way. If we go back to the discussion we had an hour ago, that type of thing can happen in several different ways.

实际上用哪种方式并不重要。如果我们回顾一个小时前的讨论,那种事情可以通过几种不同的方式实现。

But I do think when for whatever reason the models have those skills, then robotics will be revolutionized—both the design of robots, because the models will be much better than humans at that, and also the ability to control robots.

但我确实认为,无论出于什么原因,当模型拥有这些技能时,机器人技术就会发生革命——无论是机器人的设计,因为模型在这方面会比人类强得多,还是控制机器人的能力。

So we'll get better at building the physical hardware, building the physical robots, and we'll also get better at controlling it.

所以我们会更擅长制造物理硬件、制造实体机器人,同时也会更擅长控制它。

Now, does that mean the robotics industry will also be generating trillions of dollars of revenue? My answer there is yes, but there will be the same extremely fast, but not infinitely fast diffusion.

那么,这是否意味着机器人行业也会产生数万亿美元的收入呢?我的答案是肯定的,但会有同样的扩散速度——极快但不是无限快。

So will robotics be revolutionized? Yeah, maybe tack on another year or two. That's the way I think about these things.

所以机器人技术会被彻底改变吗?会的,可能再加上一两年时间。这就是我思考这些事情的方式。

Makes sense. There's a general skepticism about extremely fast progress. Here's my view. It sounds like you are going to solve continual learning one way or another within a matter of years.

有道理。人们普遍对极快的进展持怀疑态度。我的看法是这样的。听起来你们无论如何都会在几年内解决持续学习问题。

But just as people weren't talking about continual learning a couple of years ago, and then we realized, "Oh, why aren't these models as useful as they could be right now, even though they are clearly passing the Turing test and are experts in so many different domains? Maybe it's this thing."

但就像几年前人们还没有讨论持续学习,然后我们意识到「哦,为什么这些模型现在还没有达到它们应有的有用程度,尽管它们明显通过了图灵测试并且在很多不同领域都是专家?也许问题就在这里。」

Then we solve this thing and we realize, actually, there's another thing that human intelligence can do that's a basis of human labor that these models can't do.

然后我们解决了这个问题,又意识到,实际上还有另一个人类智能可以做的、作为人类劳动基础的东西,而这些模型做不到。

So why not think there will be more things like this, where we've found more pieces of human intelligence?

那么为什么不认为会有更多这样的情况,我们会发现人类智能的更多组成部分呢?

Well, to be clear, I think continual learning, as I've said before, might not be a barrier at all. I think we may just get there by pre-training generalization and RL generalization.

嗯,要说清楚的是,我认为持续学习,正如我之前所说的,可能根本不是一个障碍。我认为我们可能只需通过预训练泛化和强化学习泛化就能达到目标。

I think there just might not be such a thing at all. In fact, I would point to the history

我认为可能根本就不存在这样的东西。事实上,我会指向历史

In ML, of people coming up with things that are barriers that end up kind of dissolving within the big blob of compute. People talked about, "How do your models keep track of nouns and verbs?" "They can understand syntactically, but they can't understand semantically? It's only statistical correlations."

在机器学习领域,人们总是提出各种障碍,但这些障碍最终都会在庞大的算力面前消解。人们曾质疑:「你的模型怎么追踪名词和动词?」「它们能理解句法,但不能理解语义?这只是统计相关性而已。」

"You can understand a paragraph, you can't understand a word. There's reasoning, you can't do reasoning." But then suddenly it turns out you can do code and math very well.

「你能理解段落,但不能理解单词。推理方面,你做不了推理。」但突然之间,结果证明模型可以很好地处理代码和数学。

So I think there's actually a stronger history of some of these things seeming like a big deal and then kind of dissolving. Some of them are real. The need for data is real, maybe continual learning is a real thing.

所以我认为,这些问题看起来很重大但最终消解的历史其实更为显著。当然有些问题是真实存在的。对数据的需求是真实的,持续学习可能也是一个真实的问题。

But again, I would ground us in something like code. I think we may get to the point in a year or two where the models can just do SWE end-to-end.

但话说回来,我会把我们拉回到代码这样的具体事物上。我认为在一两年内,我们可能会达到这样的程度:模型可以端到端地完成软件工程工作。

That's a whole task. That's a whole sphere of human activity that we're just saying models can do now.

这是一整个任务。这是人类活动的整个领域,而我们现在说模型可以做到了。

When you say end-to-end, do you mean setting technical direction, understanding the context of the problem, et cetera?

当你说端到端时,你是指设定技术方向、理解问题的上下文等等这些吗?

Yes. I mean all of that.

是的。我指的就是所有这些。

Interesting. I feel like that is AGI-complete, which maybe is internally consistent. But it's not like saying 90% of code or 100% of code.

有意思。我觉得那就是AGI完备的程度了,这可能在内部逻辑上是一致的。但这不同于说完成90%的代码或100%的代码。

No, I gave this spectrum: 90% of code, 100% of code, 90% of end-to-end SWE, 100% of end-to-end SWE. New tasks are created for SWEs. Eventually those get done as well.

不,我给出了这样一个谱系:90%的代码、100%的代码、90%的端到端软件工程、100%的端到端软件工程。会为软件工程师创造新的任务。最终这些任务也会被完成。

It's a long spectrum there, but we're traversing the spectrum very quickly.

这是一个很长的谱系,但我们正在非常快速地穿越这个谱系。

I do think it's funny that I've seen a couple of podcasts you've done where the hosts will be like, "But Dwarkesh wrote the essay about the continuous learning thing." It always makes me crack up because you've been an AI researcher for 10 years.

我确实觉得挺有意思的,我看过你做的几期播客,主持人会说「但是Dwarkesh写了关于持续学习的那篇文章。」这总是让我觉得很好笑,因为你已经做了10年的AI研究员了。

I'm sure there's some feeling of, "Okay, so a podcaster wrote an essay, and every interview I get asked about it."

我相信你肯定有这样的感觉:「好吧,一个播客主持人写了篇文章,然后我每次采访都被问到这个。」

The truth of the matter is that we're all trying to figure this out together. There are some ways in which I'm able to see things that others aren't.

事实是我们都在一起努力弄清楚这些事情。在某些方面,我能看到别人看不到的东西。

These days that probably has more to do with seeing a bunch of stuff within Anthropic and having to make a bunch of decisions than I have any great research insight that others don't.

如今这可能更多是因为我在Anthropic内部看到很多东西并且需要做很多决策,而不是说我有什么别人没有的重大研究洞见。

I'm running a 2,500 person company. It's actually pretty hard for me to have concrete research insight, much harder than it would have been 10 years ago or even two or three years ago.

我在管理一个2500人的公司。对我来说,要有具体的研究洞见其实很难,比10年前甚至两三年前要难得多。

As we go towards a world of a full drop-in remote worker replacement, does an API pricing model still make the most sense? If not, what is the correct way to price AGI, or serve AGI?

当我们走向一个完全替代远程工作者的世界时,API定价模式还是最合理的吗?如果不是,那么为AGI定价或提供AGI服务的正确方式是什么?

I think there's going to be a bunch of different business models here, all at once, that are going to be experimented with. I actually do think that the API model is more durable than many people think.

我认为这里会同时出现很多不同的商业模式,都会被尝试。我确实认为API模式比很多人想的更持久。

One way I think about it is if the technology is advancing quickly, if it's advancing exponentially, what that means is there's always a surface area of new use cases that have been developed in the last three months.

我思考这个问题的一个角度是,如果技术在快速发展,如果是指数级发展,那就意味着在过去三个月里总是会有新的用例表面不断被开发出来。

Any kind of product surface you put in place is always at risk of sort of becoming irrelevant. Any given product surface probably makes sense for a range of capabilities of the model.

你设置的任何一种产品界面都总是面临着变得不再相关的风险。任何特定的产品界面可能只对模型的某个能力范围有意义。

The chatbot is already running into limitations where making it smarter doesn't really help the average consumer that much. But I don't think that's a limitation of AI models. I don't think that's evidence that the models are good enough and them getting better doesn't matter to the economy.

聊天机器人已经遇到了一些局限性,让它变得更聪明对普通消费者其实帮助不大。但我不认为这是AI模型本身的局限。我也不认为这能证明模型已经足够好了,或者说模型继续进步对经济无关紧要。

It doesn't matter to that particular product. So I think the value of the API is that the API always offers an opportunity, very close to the bare metal, to build on what the latest thing is.

这只是对那个特定产品不重要而已。所以我认为API的价值在于,API总能提供一个机会,非常接近底层,让你基于最新的技术去构建。

There's always going to be this front of new startups and new ideas that weren't possible a few months ago and are possible because the model is advancing.

总会有一波新的创业公司和新想法涌现,这些在几个月前还不可能实现,但因为模型的进步而成为可能。

I actually predict that it's going to exist alongside other models, but we're always going to have the API business model because there's always going to be a need for a thousand different people to try experimenting with the model in a different way.

我其实预测它会和其他模型并存,但我们始终会保留API商业模式,因为总会有成千上万的人需要用不同的方式来尝试和实验这个模型。

100 of them become startups and ten of them become big successful startups. Two or three really end up being the way that people use the model of a given generation.

其中100个会成为创业公司,10个会成为大型成功的创业公司。最终真正有两三个会成为人们使用某一代模型的主要方式。

So I basically think it's always going to exist. At the same time, I'm sure there's going to be other models as well.

所以我基本上认为API会一直存在。与此同时,我也确信会有其他模式出现。

Not every token that's output by the model is worth the same amount.

模型输出的每个token的价值并不相同。

Think about what is the value of the tokens that the model outputs when someone calls them up and says, "My Mac isn't working," or something, the model's like, "restart it."

想想看,当有人打电话问「我的Mac不能用了」之类的问题,模型回答「重启一下」,这时模型输出的token价值是多少。

Someone hasn't heard that before, but the model said that 10 million times. Maybe that's worth like a dollar or a few cents or something.

也许对某个人来说这是第一次听到这个建议,但模型已经说过1000万次了。也许这值一美元或几美分吧。

Whereas if the model goes to one of the pharmaceutical companies and it says, "Oh, you know, this molecule you're developing, you should take the aromatic ring from that end of the molecule and put it on that end of the molecule. If you do that, wonderful things will happen."

而如果模型对某家制药公司说:「你们正在开发的这个分子,应该把芳香环从分子的这一端移到那一端。如果这样做的话,会产生奇妙的效果。」

Those tokens could be worth tens of millions of dollars.

那些token可能价值数千万美元。

So I think we're definitely going to see business models that recognize that.

所以我认为我们肯定会看到能识别这种差异的商业模式。

At some point we're going to see "pay for results" in some form, or we may see forms of compensation that are like labor, that kind of work by the hour.

在某个时候我们会看到某种形式的「按结果付费」,或者我们可能会看到类似劳动力的补偿形式,那种按小时计费的方式。

I don't know. I think because it's a new industry, a lot of things are going to be tried. I don't know what will turn out to be the right thing.

我不知道。我觉得因为这是一个新兴行业,会尝试很多不同的东西。我也不知道最终什么会是正确的方式。

I take your point that people will have to try things to figure out what is the best way to use this blob of intelligence.

我理解你的观点,人们需要不断尝试才能找出使用这团智能的最佳方式。

But what I find striking is Claude Code. I don't think in the history of startups there has been a single application that has been as hotly competed in as coding agents.

但让我感到惊讶的是Claude Code。我认为在创业公司历史上,没有哪个单一应用像代码智能体这样竞争如此激烈。

Claude Code is a category leader here. That seems surprising to me. It doesn't seem intrinsically that Anthropic had to build this.

Claude Code是这个领域的领导者。这让我感到意外。从本质上看,并不是Anthropic必须要做这个。

I wonder if you have an accounting of why it had to be Anthropic or how Anthropic ended up building an application in addition to the model underlying it that was successful.

我想知道你能否解释一下为什么必须是Anthropic,或者Anthropic是如何在底层模型之外又成功构建了一个应用的。

So it actually happened in a pretty simple way, which is that we had our own coding models, which were good at coding.

其实发生的过程很简单,就是我们有自己的代码模型,这些模型很擅长编程。

Around the beginning of 2025, I said, "I think the time has come where you can have nontrivial acceleration of your own research if you're an AI company by using these models."

在2025年初左右,我说:「我认为时机已经成熟,如果你是一家AI公司,使用这些模型可以显著加速你自己的研究。」

Of course, you need an interface, you need a harness to use them. So I encouraged people internally. I didn't say this is one thing that you have to use.

当然,你需要一个界面,需要一个工具来使用它们。所以我在内部鼓励大家。我没有说这是你们必须使用的东西。

I just said people should experiment with this. I think it might have been originally called Claude CLI, and then the name eventually got changed to Claude Code.

我刚才说过大家应该试试这个东西。我记得它最初好像叫 Claude CLI,后来名字改成了 Claude Code。

Internally, it was the thing that everyone was using and it was seeing fast internal adoption.

在公司内部,这是每个人都在用的工具,而且内部采用速度非常快。

I looked at it and I said, "Probably we should launch this externally, right?"

我看到这个情况就说:「我们或许应该把它对外发布,对吧?」

It's seen such fast adoption within Anthropic. Coding is a lot of what we do.

它在 Anthropic 内部的采用速度如此之快。编程占了我们工作的很大一部分。

We have an audience of many, many hundreds of people that's in some ways at least representative of the external audience. So it looks like we already have product-market fit.

我们内部有几百人的用户群体,这在某种程度上至少能代表外部受众。所以看起来我们已经有了产品市场契合度。

Let's launch this thing. And then we launched it. I think just the fact that we ourselves are kind of developing the model and we ourselves know what we most need to use the model,

那就发布吧。然后我们就发布了。我觉得正是因为我们自己在开发这个模型,我们自己最清楚需要用模型来做什么,

I think it's kind of creating this feedback loop.

我觉得这形成了一种反馈循环。

I see. In the sense that you, let's say a developer at Anthropic is like, "Ah, it would be better if it was better at this X thing."

我明白了。也就是说,比如 Anthropic 的一个开发者会想:「啊,要是它在 X 这方面做得更好就好了。」

Then you bake that into the next model that you build.

然后你们就把这个需求融入到下一个版本的模型中。

That's one version of it, but then there's just the ordinary product iteration.

这是一个方面,但同时还有常规的产品迭代。

We have a bunch of coders within Anthropic, they use Claude Code every day and so we get fast feedback.

我们在 Anthropic 内部有一批程序员,他们每天都用 Claude Code,所以我们能快速获得反馈。

That was more important in the early days.

这在早期阶段尤其重要。

Now, of course, there are millions of people using it, and so we get a bunch of external feedback as well.

现在当然有数百万人在使用它,所以我们也收到了大量外部反馈。

But it's just great to be able to get kind of fast internal feedback.

但能够快速获得内部反馈这一点真的很棒。

I think this is the reason why we launched a coding model and didn't launch a pharmaceutical company.

我觉得这就是为什么我们推出了编程模型而不是去创办制药公司的原因。

My background's in biology, but we don't have any of the resources that are needed to launch a pharmaceutical company.

我的背景是生物学,但我们没有创办制药公司所需的任何资源。

Let me now ask you about making AI go well.

现在让我来问问关于如何让 AI 朝好的方向发展的问题。

It seems like whatever vision we have about how AI goes well has to be compatible with two things:

看起来,无论我们对 AI 如何良好发展有什么愿景,都必须与两件事相容:

1) the ability to build and run AIs is diffusing extremely rapidly and 2) the population of AIs, the amount we have and their intelligence, will also increase very rapidly.

第一,构建和运行 AI 的能力正在极快速地扩散;第二,AI 的数量、我们拥有的 AI 以及它们的智能水平,也都会非常快速地增长。

That means that lots of people will be able to build huge populations of misaligned AIs, or AIs which are just companies which are trying to increase their footprint or have weird psyches like Sydney Bing, but now they're superhuman.

这意味着很多人将能够构建大量未对齐的 AI,或者那些只是作为公司试图扩大影响力的 AI,或者有着像 Sydney Bing 那样奇怪心理的 AI,但现在它们是超人类水平的。

What is a vision for a world in which we have an equilibrium that is compatible with lots of different AIs, some of which are misaligned, running around?

在一个有许多不同的 AI、其中一些是未对齐的 AI 到处运行的世界中,什么样的愿景能让我们达成一个与之相容的平衡?

I think in "The Adolescence of Technology", I was skeptical of the balance of power.

我记得在「技术的青春期」那篇文章中,我对权力平衡持怀疑态度。

But the thing I was specifically skeptical of is you have three or four of these companies all building models that are derived from the same thing, that they would check each other.

但我具体怀疑的是,你有三四家公司都在构建源自同一技术的模型,它们会相互制衡。

Or even that any number of them would check each other.

或者说,甚至任何数量的它们会相互制衡。

We might live in an offense-dominant world where one person or one AI model is smart enough to do something that causes damage for everything else.

我们可能生活在一个进攻占主导的世界里,一个人或一个 AI 模型足够聪明,能做出对其他所有东西造成损害的事情。

In the short run, we have a limited number of players now.

在短期内,我们现在的参与者数量是有限的。

So we can start within the limited number of players.

所以我们可以从这有限数量的参与者入手。

We need to put in place the safeguards.

我们需要建立安全保障措施。

We need to make sure everyone does the right alignment work.

我们需要确保每个人都做好对齐工作。

We need to make sure everyone has bioclassifiers. Those are the immediate things we need to do.

我们需要确保每个人都有生物分类器。这些是我们需要立即做的事情。

I agree that that doesn't solve the problem in the long run, particularly if the ability of

我同意这在长期来看并不能解决问题,特别是如果

AI models to make other AI models proliferate, then the whole thing can become harder to solve.

如果用 AI 模型来制造其他 AI 模型的情况扩散开来,那么整个问题就会变得更难解决。

I think in the long run we need some architecture of governance that preserves human freedom, but also allows us to govern a very large number of human systems, AI systems, hybrid human-AI companies or economic units.

我认为从长远来看,我们需要某种治理架构,既能保护人类自由,又能让我们管理数量庞大的人类系统、AI 系统、以及人机混合的公司或经济单元。

So we're gonna need to think about: how do we protect the world against bioterrorism? How do we protect the world against mirror life?

所以我们需要思考:如何保护世界免受生物恐怖主义的威胁?如何保护世界免受镜像生命的威胁?

Probably we're gonna need some kind of AI monitoring system that monitors for all of these things. But then we need to build this in a way that preserves civil liberties and our constitutional rights.

我们可能需要某种 AI 监控系统来监测所有这些威胁。但同时我们需要以一种能保护公民自由和宪法权利的方式来构建这个系统。

So I think just as anything else, it's a new security landscape with a new set of tools and a new set of vulnerabilities.

所以我认为,就像其他事物一样,这是一个全新的安全格局,有一套新的工具和一套新的漏洞。

My worry is, if we had 100 years for this to happen all very slowly, we'd get used to it. We've gotten used to the presence of explosives in society or the presence of various new weapons or the presence of video cameras.

我担心的是,如果这一切能在 100 年内非常缓慢地发生,我们会逐渐适应。我们已经适应了社会中炸药的存在、各种新武器的存在、或者监控摄像头的存在。

We would get used to it over 100 years and we'd develop governance mechanisms. We'd make our mistakes. My worry is just that this is happening all so fast.

在 100 年的时间里我们会逐渐适应,并发展出治理机制。我们会犯错,然后吸取教训。我担心的只是这一切发生得太快了。

So maybe we need to do our thinking faster about how to make these governance mechanisms work.

所以也许我们需要更快地思考如何让这些治理机制运作起来。

It seems like in an offense-dominant world, over the course of the next century—the idea is that AI is making the progress that would happen over the next century happen in some period of five to ten years—we would still need the same mechanisms, or balance of power would be similarly intractable, even if humans were the only game in town.

在一个进攻占优势的世界里,在接下来的一个世纪过程中——也就是说 AI 把原本需要一个世纪才能实现的进步压缩到五到十年内完成——即使只有人类在场,我们仍然需要同样的机制,或者说权力平衡仍然会同样棘手。

I guess we have the advice of AI. But it fundamentally doesn't seem like a totally different ball game here.

我想我们可以听取 AI 的建议。但这从根本上看起来并不像是完全不同的游戏规则。

If checks and balances were going to work, they would work with humans as well. If they aren't going to work, they wouldn't work with AIs as well.

如果制衡机制能够奏效,那么在人类身上就能奏效。如果不能奏效,在 AI 身上也不会奏效。

So maybe this just dooms human checks and balances as well.

所以也许这也宣告了人类制衡机制的终结。

Again, I think there's some way to make this happen. The governments of the world may have to work together to make it happen.

再说一次,我认为总有办法实现这一目标。世界各国政府可能必须合作才能实现。

We may have to talk to AIs about building societal structures in such a way that these defenses are possible. I don't know. I don't want to say this is so far ahead in time, but it's so far ahead in technological ability that may happen over a short period of time, that it's hard for us to anticipate it in advance.

我们可能需要和 AI 讨论如何构建社会结构,使这些防御措施成为可能。我不知道。我不想说这在时间上有多遥远,但这在技术能力上确实遥遥领先,而且可能在短时间内发生,以至于我们很难提前预见。

Speaking of governments getting involved, on December 26, the Tennessee legislature introduced a bill which said, "It would be an offense for a person to knowingly train artificial intelligence to provide emotional support, including through open-ended conversations with a user."

说到政府介入,12 月 26 日,Tennessee 州议会提出了一项法案,称「故意训练人工智能提供情感支持,包括通过与用户的开放式对话,将构成违法行为」。

Of course, one of the things that Claude attempts to do is be a thoughtful, knowledgeable friend.

当然,Claude 试图做的事情之一就是成为一个体贴、博学的朋友。

In general, it seems like we're going to have this patchwork of state laws. A lot of the benefits that normal people could experience as a result of AI are going to be curtailed, especially when we get into the kinds of things you discuss in "Machines of Loving Grace": biological freedom, mental health improvements, et cetera.

总的来说,我们似乎会面临一个州法律的拼凑局面。普通人本可以从 AI 中获得的许多好处将会被削减,尤其是当我们涉及到你在「Machines of Loving Grace」中讨论的那些内容时:生物自由、心理健康改善等等。

It seems easy to imagine worlds in which these get whac-a-moled away by different laws, whereas bills like this don't seem to address the actual existential threats that you're concerned about.

很容易想象这样的世界:这些好处被各种不同的法律像打地鼠一样逐个消灭,而像这样的法案似乎并没有解决你所担心的真正的存在性威胁。

I'm curious to understand, in the context of things like this, Anthropic's position against the federal moratorium on state AI laws.

我很好奇想了解一下,在这样的背景下,Anthropic 对联邦暂停各州 AI 法律这一立场的态度。

There are many different things going on at once. I think that particular law is dumb.

有很多不同的事情同时在发生。我认为那个特定的法律很愚蠢。

It was clearly made by legislators who just probably had little idea what AI models could do and not do.

很明显,制定这个法律的立法者可能对 AI 模型能做什么、不能做什么几乎没什么概念。

They're like, "AI models serving us, that just sounds scary. I don't want that to happen."

他们就像是,「AI 模型为我们服务,听起来就很吓人。我不想让这种事发生。」

So we're not in favor of that. But that wasn't the thing that was being voted on.

所以我们不支持那个法律。但那并不是正在被投票表决的东西。

The thing that was being voted on is: we're going to ban all state regulation of AI for 10 years with no apparent plan to do any federal regulation of AI, which would take Congress to pass, which is a very high bar.

正在被投票表决的是:我们要禁止各州对 AI 进行任何监管长达 10 年,同时没有任何明确的联邦 AI 监管计划,而联邦监管需要国会通过,门槛非常高。

So the idea that we'd ban states from doing anything for 10 years… People said they had a plan for the federal government, but there was no actual proposal on the table.

所以禁止各州做任何事情长达 10 年这个想法……有人说他们有联邦层面的计划,但实际上桌面上并没有任何具体提案。

There was no actual attempt. Given the serious dangers that I lay out in "Adolescence of Technology" around things like biological weapons and bioterrorism autonomy risk, and the timelines we've been talking about—10 years is an eternity—I think that's a crazy thing to do.

根本没有实际的尝试。考虑到我在「Adolescence of Technology」中阐述的严重危险,比如生物武器和生物恐怖主义自主性风险,还有我们一直在讨论的时间线——10 年简直是一个永恒——我认为这样做太疯狂了。

So if that's the choice, if that's what you force us to choose, then we're going to choose not to have that moratorium.

所以如果这就是选择,如果你们强迫我们做这个选择,那我们会选择不要那个暂停令。

I think the benefits of that position exceed the costs, but it's not a perfect position if that's the choice.

我认为这个立场的好处大于代价,但如果这就是选择的话,它并不是一个完美的立场。

Now, I think the thing that we should do, the thing that I would support, is the federal government should step in, not saying "states you can't regulate", but "Here's what we're going to do, and states you can't differ from this."

现在,我认为我们应该做的、我会支持的,是联邦政府应该介入,不是说「各州你们不能监管」,而是「这是我们要做的,各州不能偏离这个标准。」

I think preemption is fine in the sense of saying that the federal government says, "Here is our standard. This applies to everyone. States can't do something different."

我认为联邦优先权是可以的,也就是联邦政府说,「这是我们的标准。适用于所有人。各州不能做不同的事情。」

That would be something I would support if it would be done in the right way.

如果以正确的方式来做,这是我会支持的。

But this idea of states, "You can't do anything and we're not doing anything either," that struck us as very much not making sense.

但这种对各州说「你们什么都不能做,而我们也什么都不做」的想法,在我们看来完全说不通。

I think it will not age well, it is already starting to not age well with all the backlash that you've seen.

我认为这个立场经不起时间考验,从你看到的所有反弹来看,它已经开始显得过时了。

Now, in terms of what we would want, the things we've talked about are starting with transparency standards in order to monitor some of these autonomy risks and bioterrorism risks.

现在,就我们想要什么而言,我们谈到的事情是从透明度标准开始,以便监测一些自主性风险和生物恐怖主义风险。

As the risks become more serious, as we get more evidence for them, then I think we could be more aggressive in some targeted ways and say, "Hey, AI bioterrorism is really a threat. Let's pass a law that forces people to have classifiers."

随着风险变得更严重,随着我们获得更多证据,我认为我们可以在一些有针对性的方面更积极主动,比如说,「嘿,AI 生物恐怖主义真的是个威胁。我们来通过一项法律,强制人们使用分类器。」

I could even imagine… It depends. It depends how serious the threat it ends up being.

我甚至可以想象……这取决于具体情况。取决于威胁最终有多严重。

We don't know for sure. We need to pursue this in an intellectually honest way where we say that ahead of time, the risk has not emerged yet.

我们无法确定。我们需要以智识诚实的方式推进这件事,提前说明风险尚未出现。

But I could certainly imagine, with the pace that things are going at, a world where later this year we say, "Hey, this AI bioterrorism stuff is really serious. We should do something about it. We should put it in a federal standard.

但我完全可以想象,按照目前的发展速度,在今年晚些时候我们会说,「嘿,这个 AI 生物恐怖主义的事情真的很严重。我们应该对此采取行动。我们应该把它纳入联邦标准。

If the federal government won't act, we should put it in a state standard." I could totally see that.

如果联邦政府不行动,我们应该把它纳入州标准。」我完全能看到这种情况发生。

I'm concerned about a world where if you just consider the pace of progress you're expecting, the life cycle of legislation...

我担心的是,如果你只考虑你所预期的进展速度,还有立法的生命周期……

The benefits are, as you say because of diffusion lag, slow enough that I really do think this patchwork of state

好处是,正如你所说,由于扩散滞后,速度足够慢,所以我确实认为这种各州拼凑的

Laws, on the current trajectory, would prohibit. I mean, if having an emotional chatbot friend is something that freaks people out, then just imagine the kinds of actual benefits from AI we want normal people to be able to experience.

按照目前的发展轨迹,法律会禁止这些。我的意思是,如果拥有一个情感聊天机器人朋友就让人们感到不安,那么想象一下我们希望普通人能够体验到的 AI 实际好处会是什么样的。

From improvements in health and healthspan and improvements in mental health and so forth.

比如健康和健康寿命的改善,以及心理健康的改善等等。

Whereas at the same time, it seems like you think the dangers are already on the horizon and I just don't see that much... It seems like it would be especially injurious to the benefits of AI as compared to the dangers of AI.

然而与此同时,你似乎认为危险已经近在眼前,而我并没有看到那么多……在我看来,这对 AI 的好处造成的伤害会特别大,相比于 AI 的危险而言。

So that's maybe where the cost-benefit makes less sense to me.

所以这可能就是我觉得成本收益分析不太合理的地方。

So there's a few things here. People talk about there being thousands of these state laws.

这里有几点需要说明。人们谈论说有成千上万条这样的州法律。

First of all, the vast, vast majority of them do not pass.

首先,其中绝大多数都不会通过。

The world works a certain way in theory, but just because a law has been passed doesn't mean it's really enforced.

理论上世界是以某种方式运作的,但法律通过并不意味着它真的会被执行。

The people implementing it may be like, 'Oh my God, this is stupid. It would mean shutting off everything that's ever been built in Tennessee.'

执行这些法律的人可能会想:「天哪,这太蠢了。这意味着要关闭 Tennessee 州有史以来建造的所有东西。」

Very often, laws are interpreted in a way that makes them not as dangerous or harmful.

很多时候,法律的解释方式会让它们不那么危险或有害。

On the same side, of course, you have to worry if you're passing a law to stop a bad thing; you have this problem as well.

当然,从另一个角度来说,如果你通过一项法律来阻止坏事,你也会面临这个问题。

My basic view is that if we could decide what laws were passed and how things were done—and we're only one small input into that—I would deregulate a lot of the stuff around the health benefits of AI.

我的基本观点是,如果我们能决定通过什么法律以及如何做事——而我们只是其中一个小小的输入——我会放松很多关于 AI 健康益处方面的监管。

I don't worry as much about the chatbot laws.

我对聊天机器人相关法律并不那么担心。

I actually worry more about the drug approval process, where I think AI models are going to greatly accelerate the rate at which we discover drugs, and the pipeline will get jammed up.

我实际上更担心药物审批流程,我认为 AI 模型将大大加快我们发现药物的速度,而审批流程会被堵塞。

The pipeline will not be prepared to process all the stuff that's going through it.

这个流程不会做好准备来处理所有通过它的东西。

I think reform of the regulatory process should bias more towards the fact that we have a lot of things coming where the safety and efficacy is actually going to be really crisp and clear, a beautiful thing, and really effective.

我认为监管流程的改革应该更多地倾向于这样一个事实:我们会有很多东西即将到来,它们的安全性和有效性将会非常清晰明确,是一件美好的事情,而且真的很有效。

Maybe we don't need all this superstructure around it that was designed around an era of drugs that barely work and often have serious side effects.

也许我们不需要围绕它的所有这些上层结构,这些结构是为那个药物勉强有效且经常有严重副作用的时代设计的。

At the same time, I think we should be ramping up quite significantly the safety and security legislation.

与此同时,我认为我们应该大幅加强安全和安保方面的立法。

Like I've said, starting with transparency is my view of trying not to hamper the industry, trying to find the right balance.

正如我所说,从透明度开始是我试图不妨碍行业发展、试图找到正确平衡的观点。

I'm worried about it. Some people criticize my essay for saying, 'That's too slow. The dangers of AI will come too soon if we do that.'

我对此感到担忧。一些人批评我的文章,说:「那太慢了。如果我们那样做,AI 的危险会来得太快。」

Well, basically, I think the last six months and maybe the next few months are going to be about transparency.

基本上,我认为过去六个月和也许接下来的几个月将会是关于透明度的。

Then, if these risks emerge when we're more certain of them—which I think we might be as soon as later this year—then I think we need to act very fast in the areas where we've actually seen the risk.

然后,如果这些风险在我们更加确定的时候出现——我认为可能最早在今年晚些时候——那么我认为我们需要在实际看到风险的领域非常迅速地采取行动。

I think the only way to do this is to be nimble.

我认为做到这一点的唯一方法就是保持灵活。

Now, the legislative process is normally not nimble, but we need to emphasize the urgency of this to everyone involved.

现在,立法过程通常并不灵活,但我们需要向所有相关人员强调这件事的紧迫性。

That's why I'm sending this message of urgency.

这就是为什么我要传达这个紧迫性的信息。

That's why I wrote Adolescence of Technology. I wanted policymakers, economists, national security professionals, and decision-makers to read it so that they have some hope of acting faster than they would have otherwise.

这就是为什么我写了《技术的青春期》。我希望政策制定者、经济学家、国家安全专业人士和决策者们能读到它,这样他们才有希望比原本更快地采取行动。

Is there anything you can do or advocate

你能做什么或者倡导什么

That would make it more certain that the benefits of AI are better instantiated? I feel like you have worked with legislatures to say, "Okay, we're going to prevent bioterrorism here. We're going to increase transparency, we're going to increase whistleblower protection."

有什么能让 AI 的好处更确定地实现出来?我感觉你已经和立法机构合作过,跟他们说「好的,我们要在这里防止生物恐怖主义,我们要增加透明度,我们要增加对举报人的保护」。

But I think by default, the actual benefits we're looking forward to seem very fragile to different kinds of moral panics or political economy problems.

但我觉得,在默认情况下,我们所期待的实际好处似乎很容易受到各种道德恐慌或政治经济问题的影响而变得脆弱。

I don't actually agree that much regarding the developed world. I feel like in the developed world, markets function pretty well.

关于发达国家,我其实不太同意这个看法。我觉得在发达国家,市场运作得相当好。

When there's a lot of money to be made on something and it's clearly the best available alternative, it's actually hard for the regulatory system to stop it.

当某件事能赚很多钱,而且它显然是现有的最佳选择时,监管系统其实很难阻止它。

We're seeing that in AI itself. A thing I've been trying to fight for is export controls on chips to China.

我们在 AI 本身就看到了这一点。我一直在努力争取的一件事就是对中国的芯片出口管制。

That's in the national security interest of the US. That's squarely within the policy beliefs of almost everyone in Congress of both parties.

这符合美国的国家安全利益。这完全在国会两党几乎所有人的政策信念范围之内。

The case is very clear. The counterarguments against it, I'll politely call them fishy.

理由非常清楚。那些反对的论据,我客气地说,很可疑。

Yet it doesn't happen and we sell the chips because there's so much money riding on it.

然而它还是没有发生,我们还是把芯片卖了,因为这里面涉及太多钱了。

That money wants to be made. In that case, in my opinion, that's a bad thing.

那些钱想要被赚到。在那种情况下,我认为这是件坏事。

But it also applies when it's a good thing. So if we're talking about drugs and benefits of the technology, I am not as worried about those benefits being hampered in the developed world.

但当它是好事时,这个道理也同样适用。所以如果我们在谈论药物和技术带来的好处,我并不太担心这些好处在发达国家会受到阻碍。

I am a little worried about them going too slow. As I said, I do think we should work to speed the approval process in the FDA.

我有点担心它们进展太慢。就像我说的,我确实认为我们应该努力加快 FDA 的审批流程。

I do think we should fight against these chatbot bills that you're describing. Described individually, I'm against them. I think they're stupid.

我确实认为我们应该反对你描述的那些聊天机器人法案。单独来看,我反对它们。我觉得它们很愚蠢。

But I actually think the bigger worry is the developing world, where we don't have functioning markets and where we often can't build on the technology that we've had.

但我实际上认为更大的担忧是发展中国家,那里没有运转良好的市场,而且我们往往无法在现有的技术基础上继续发展。

I worry more that those folks will get left behind.

我更担心那些人会被落在后面。

And I worry that even if the cures are developed, maybe there's someone in rural Mississippi who doesn't get it as well.

而且我担心即使治疗方法被开发出来了,也许 Mississippi 州农村地区的某些人也得不到。

That's a smaller version of the concern we have in the developing world.

这是我们在发展中国家所担忧问题的一个缩小版。

So the things we've been doing are working with philanthropists. We work with folks who deliver medicine and health interventions to the developing world, to sub-Saharan Africa, India, Latin America, and other developing parts of the world.

所以我们一直在做的事情就是与慈善家合作。我们与那些向发展中国家、撒哈拉以南非洲、印度、拉丁美洲和世界其他发展中地区提供医药和健康干预的人合作。

That's the thing I think that won't happen on its own.

我认为这件事不会自己发生。

You mentioned export controls. Why shouldn't the US and China both have a "country of geniuses in a data center"?

你提到了出口管制。为什么美国和中国不应该都拥有「数据中心里的天才之国」?

Why won't it happen or why shouldn't it happen?

是为什么它不会发生,还是为什么它不应该发生?

Why shouldn't it happen.

为什么它不应该发生。

If this does happen, we could have a few situations. If we have an offense-dominant situation, we could have a situation like nuclear weapons, but more dangerous.

如果这真的发生了,我们可能会面临几种情况。如果我们有一个进攻占主导的局面,我们可能会遇到类似核武器的情况,但更危险。

Either side could easily destroy everything.

任何一方都可以轻易地摧毁一切。

We could also have a world where it's unstable. The nuclear equilibrium is stable because it's deterrence.

我们也可能面临一个不稳定的世界。核均衡是稳定的,因为它是威慑。

But let's say there was uncertainty about, if the two AIs fought, which AI would win? That could create instability.

但假设存在不确定性,如果两个 AI 打起来,哪个 AI 会赢?这可能会造成不稳定。

You often have conflict when the two sides have a different assessment of their likelihood of winning.

当双方对自己获胜的可能性有不同评估时,往往会发生冲突。

If one side is like, "Oh yeah, there's a 90% chance I'll win," and the other side thinks the same, then a fight is much more likely. They can't both be right,

如果一方说「哦对,我有 90% 的胜算」,而另一方也这么想,那么冲突就更有可能发生。他们不可能都是对的,

but they can both think that. But this seems like a fully general argument against the diffusion of AI technology.

但他们可以都这么认为。但这似乎是一个完全通用的反对 AI 技术扩散的论点。

That's the implication of this world.

这就是这种世界观的含义。

Let me just go on, because I think we will get diffusion eventually.

让我继续说下去,因为我认为我们最终会实现技术扩散。

The other concern I have is that governments will oppress their own people with AI.

我的另一个担忧是,政府会利用 AI 压迫自己的人民。

I'm worried about a world where you have a country in which there's already a government that's building a high-tech authoritarian state.

我担心的是这样一个世界:某个国家的政府正在建立一个高科技专制国家。

To be clear, this is about the government.

要明确一点,这是关于政府的问题。

This is not about the people. We need to find a way for people everywhere to benefit.

这不是关于人民的问题。我们需要找到一种方式,让世界各地的人们都能受益。

My worry here is about governments.

我在这里担心的是政府。

My worry is if the world gets carved up into two pieces, one of those two pieces could be authoritarian or totalitarian in a way that's very difficult to displace.

我担心的是,如果世界被分成两部分,其中一部分可能会变成专制或极权体制,而且很难被推翻。

Now, will governments eventually get powerful AI, and is there a risk of authoritarianism?

那么,政府最终会获得强大的 AI 吗?会有专制主义的风险吗?

Yes. Will governments eventually get powerful AI, and is there a risk of bad equilibria? Yes, I think both things. But the initial conditions matter. At some point, we're going to need to set up the rules of the road.

是的。政府最终会获得强大的 AI,会有不良均衡的风险吗?是的,我认为两者都会发生。但初始条件很重要。在某个时刻,我们需要建立游戏规则。

I'm not saying that one country, either the United States or a coalition of democracies—which I think would be a better setup, although it requires more international cooperation than we currently seem to want to make—should just say, "These are the rules of the road."

我不是说某一个国家,无论是美国还是民主国家联盟——我认为后者会是更好的安排,尽管这需要比我们目前愿意做的更多的国际合作——应该直接说「这些就是游戏规则」。

There's going to be some negotiation.

会有一些谈判。

The world is going to have to grapple with this. What I would like is for the democratic nations of the world—those whose governments represent closer to pro-human values—are holding the stronger hand and have more leverage when the rules of the road are set.

全世界都将不得不面对这个问题。我希望的是,世界上的民主国家——那些政府更接近于代表亲人类价值观的国家——在制定游戏规则时能握有更强的牌面和更大的影响力。

So I'm very concerned about that initial condition.

所以我非常关注那个初始条件。

I was re-listening to the interview from three years ago, and one of the ways it aged poorly is that I kept asking questions assuming there was going to be some key fulcrum moment two to three years from now.

我重新听了三年前的采访,它过时的一个方面是,我一直在提问时假设未来两到三年会有某个关键的转折时刻。

In fact, being that far out, it just seems like progress continues, AI improves, AI is more diffused, and people will use it for more things.

实际上,展望那么远,看起来只是进步在继续,AI 在改进,AI 更加普及,人们会用它做更多的事情。

It seems like you're imagining a world in the future where the countries get together, and "Here's the rules of the road, here's the leverage we have, and here's the leverage you have."

看起来你设想的是未来的一个世界,各国聚在一起,然后说「这些是游戏规则,这是我们的筹码,这是你们的筹码」。

But on the current trajectory, everybody will have more AI.

但按照目前的轨迹,每个人都会拥有更多的 AI。

Some of that AI will be used by authoritarian countries.

其中一些 AI 会被专制国家使用。

Some of that within the authoritarian countries will be used by private actors versus state actors.

在专制国家内部,一些 AI 会被私人行为者而非国家行为者使用。

It's not clear who will benefit more.

目前还不清楚谁会受益更多。

It's always unpredictable to tell in advance. It seems like the internet privileged authoritarian countries more than you would've expected.

事先总是难以预测。看起来互联网对专制国家的助益比你预期的要多。

Maybe AI will be the opposite way around. I want to better understand what you're imagining here.

也许 AI 会恰恰相反。我想更好地理解你在这里设想的是什么。

Just to be precise about it,

准确地说,

I think the exponential of the underlying technology will continue as it has before.

我认为底层技术的指数级增长会像以前一样继续下去。

The models get smarter and smarter, even when they get to a "country of geniuses in a data center."

模型会变得越来越聪明,即使它们达到了「数据中心里的天才之国」的程度。

I think you can continue to make the model smarter.

我认为你可以继续让模型变得更聪明。

There's a question of getting diminishing returns on their value in the world. How much does it matter after you've already solved human biology?

存在一个边际收益递减的问题——它们在现实世界中的价值会逐渐降低。在你已经解决了人类生物学问题之后,这些进步还有多重要呢?

At some point you can do harder, more abstruse math problems, but nothing after that matters.

到了某个阶段,你可以解决更难、更深奥的数学问题,但在那之后就没什么实际意义了。

Putting that aside, I do think the exponential will continue, but there will be certain distinguished points on the exponential. Companies, individuals, and countries will reach those points at different times.

撇开这个不谈,我确实认为指数增长会持续下去,但在指数曲线上会有某些关键节点。不同的公司、个人和国家会在不同时间到达这些节点。

In "The Adolescence of Technology" I talk about: Is a nuclear deterrent still stable in the world of AI?

在「技术的青春期」这篇文章中,我探讨了一个问题:在AI时代,核威慑还稳定吗?

I don't know, but that's an example of one thing we've taken for granted. The technology could reach such a level that we can no longer be certain of it.

我不知道答案,但这是我们一直认为理所当然的事情的一个例子。技术可能会达到这样的水平,让我们无法再对此确信。

Think of others. There are points where if you reach a certain level, maybe you have offensive cyber dominance, and every computer system is transparent to you after that unless the other side has an equivalent defense.

再想想其他方面。存在这样的节点:如果你达到某个水平,也许你就拥有了网络攻击优势,之后每个计算机系统对你来说都是透明的,除非对方拥有同等水平的防御。

I don't know what the critical moment is or if there's a single critical moment. But I think there will be either a critical moment, a small number of critical moments, or some critical window where AI confers some large advantage from the perspective of national security, and one country or coalition has reached it before others.

我不知道关键时刻是什么,也不知道是否存在单一的关键时刻。但我认为会有一个关键时刻、少数几个关键时刻,或者某个关键窗口期,在这个时期AI从国家安全角度会带来巨大优势,而某个国家或联盟会先于其他国家达到这个阶段。

I'm not advocating that they just say, "Okay, we're in charge now." That's not how I think about it.

我并不是在主张他们直接说「好,现在我们说了算」。我不是这么想的。

The other side is always catching up. There are extreme actions you're not willing to take, and it's not right to take complete control anyway.

对方总是在追赶。有些极端行动你不愿意采取,而且完全控制本来就是不对的。

But at the point that happens, people are going to understand that the world has changed. There's going to be some negotiation, implicit or explicit, about what the post-AI world order looks like.

但当那个时刻到来时,人们会意识到世界已经改变了。会有某种谈判,无论是隐性的还是显性的,来决定后AI时代的世界秩序是什么样的。

My interest is in making that negotiation be one in which classical liberal democracy has a strong hand.

我的目标是让这场谈判成为古典自由民主占据强势地位的谈判。

I want to understand what that better means, because you say in the essay, "Autocracy is simply not a form of government that people can accept in the post-powerful AI age."

我想更好地理解这是什么意思,因为你在文章中说「专制根本不是人们在强大AI时代之后可以接受的政府形式」。

That sounds like you're saying the CCP as an institution cannot exist after we get AGI. That seems like a very strong demand, and it seems to imply a world where the leading lab or the leading country will be able to—and by that language, should get to—determine how the world is governed or what kinds of governments are, and are not, allowed.

这听起来像是你在说,一旦我们拥有AGI,中国共产党作为一个机构就无法存在了。这似乎是一个非常强硬的要求,而且似乎暗示着这样一个世界:领先的实验室或领先的国家将能够——而且按照那种说法,应该有权——决定世界如何被治理,或者哪些政府形式是允许的,哪些是不允许的。

I believe that paragraph said something like, "You could take it even further and say X." I wasn't necessarily endorsing that view.

我记得那一段说的是类似「你甚至可以更进一步说X」这样的话。我不一定赞同那个观点。

I was saying, "Here's a weaker thing that I believe. We have to worry a lot about authoritarians and we should try to check them and limit their power. You could take this much further and have a more interventionist view that says authoritarian countries with AI are these self-fulfilling cycles that are very hard to displace, so you just need to get rid of them from the beginning."

我当时说的是「这是我所持有的一个相对温和的观点。我们必须非常担心专制政权,我们应该试图遏制和限制他们的权力。你可以把这个观点推得更远,采取一种更具干预性的观点,认为拥有AI的专制国家会形成这种很难打破的自我强化循环,所以你需要从一开始就消除它们」。

That has exactly all the problems you say. If you were to make a commitment to overthrowing every authoritarian country, they would take a bunch of actions now that could lead to instability. That just may not be possible.

那确实有你说的所有问题。如果你承诺要推翻每一个专制国家,他们现在就会采取一系列可能导致不稳定的行动。这可能根本做不到。

But the point I was making that I do endorse is that it is quite possible that... Today, the view, my view, in most of the Western world is that democracy is a better form of...

但我确实支持我所提出的观点是,很有可能……今天,我的观点,在大多数西方世界的观点是,民主是一种更好的……

government than authoritarianism. But if a country's authoritarian, we don't react the way we'd react if they committed a genocide or something.

政府形式相比威权主义。但如果一个国家是威权政体,我们不会像他们犯下种族灭绝或类似罪行时那样做出反应。

I guess what I'm saying is I'm a little worried that in the age of AGI, authoritarianism will have a different meaning. It will be a graver thing.

我想说的是,我有点担心在 AGI 时代,威权主义会有不同的含义。它会变成更严重的问题。

We have to decide one way or another how to deal with that.

我们必须以某种方式决定如何应对这个问题。

The interventionist view is one possible view. I was exploring such views. It may end up being the right view, or it may end up being too extreme.

干预主义观点是一种可能的观点。我在探索这类观点。它最终可能是正确的观点,也可能太极端。

But I do have hope. One piece of hope I have is that we have seen that as new technologies are invented, forms of government become obsolete.

但我确实抱有希望。我希望的一点是,我们已经看到随着新技术的发明,某些政府形式会变得过时。

I mentioned this in "Adolescence of Technology", where I said feudalism was basically a form of government, and when we invented industrialization, feudalism was no longer sustainable. It no longer made sense.

我在「技术的青春期」中提到过这一点,我说封建制度本质上是一种政府形式,当我们发明了工业化,封建制度就不再可持续了。它不再有意义了。

Why is that hope? Couldn't that imply that democracy is no longer going to be a competitive system?

为什么这是希望呢?这难道不是意味着民主将不再是一个有竞争力的制度吗?

Right, it could go either way. But these problems with authoritarianism get deeper.

对,可能朝任何方向发展。但威权主义的这些问题会变得更深刻。

I wonder if that's an indicator of other problems that authoritarianism will have.

我想知道这是否预示着威权主义将面临的其他问题。

In other words, because authoritarianism becomes worse, people are more afraid of it. They work harder to stop it.

换句话说,因为威权主义变得更糟,人们更害怕它。他们会更努力地阻止它。

You have to think in terms of total equilibrium. I just wonder if it will motivate new ways of thinking about how to preserve and protect freedom with the new technology.

你必须从整体平衡的角度思考。我只是想知道这是否会激发新的思维方式,思考如何用新技术来保护和维护自由。

Even more optimistically, will it lead to a collective reckoning and a more emphatic realization of how important some of the things we take as individual rights are?

更乐观地说,这会不会导致集体反思,让人们更强烈地意识到我们视为个人权利的那些东西有多重要?

A more emphatic realization that we really can't give these away.

更强烈地意识到我们真的不能放弃这些权利。

We've seen there's no other way to live that actually works.

我们已经看到没有其他真正行得通的生活方式。

I am actually hopeful that—it sounds too idealistic, but I believe it could be the case—dictatorships become morally obsolete.

我实际上抱有希望——这听起来太理想主义了,但我相信可能会是这样——独裁政权在道德上变得过时。

They become morally unworkable forms of government and the crisis that that creates is sufficient to force us to find another way.

它们成为道德上行不通的政府形式,由此产生的危机足以迫使我们找到另一条道路。

I think there is genuinely a tough question here which I'm not sure how you resolve.

我认为这里确实有一个棘手的问题,我不确定该如何解决。

We've had to come out one way or another on it through history.

纵观历史,我们不得不以某种方式对此做出选择。

With China in the '70s and '80s, we decided that even though it's an authoritarian system, we will engage with it.

对于七八十年代的中国,我们决定即使它是威权体制,我们也会与之接触。

I think in retrospect that was the right call, because it's a state authoritarian system but a billion-plus people are much wealthier and better off than they would've otherwise been.

我认为回过头看那是正确的决定,因为虽然它是国家威权体制,但十亿多人比原本的情况要富裕得多、过得好得多。

It's not clear that it would've stopped being an authoritarian country otherwise. You can just look at North Korea as an example of that.

并不清楚如果不那样做它是否会停止成为威权国家。你只要看看朝鲜的例子就知道了。

I don't know if it takes that much intelligence to remain an authoritarian country that continues to coalesce its own power.

我不知道保持威权国家并继续巩固自己的权力是否需要那么高的智能。

You can imagine a North Korea with an AI that's much worse than everybody else's, but still enough to keep power.

你可以想象一个拥有 AI 的朝鲜,虽然它的 AI 比其他所有人的都差得多,但仍然足以维持政权。

In general, it seems like we should just have this attitude that the benefits of AI—in the form of all these empowerments of humanity and health—will be big.

总的来说,我们似乎应该持有这样的态度:AI 的好处——以赋能人类和健康等各种形式呈现——将是巨大的。

Historically, we have decided it's good to spread the benefits of technology widely, even to people whose governments are authoritarian.

历史上,我们已经决定广泛传播技术的好处是有益的,即使是对那些政府是威权体制的人民。

It is a tough question, how to think about it with AI, but historically we have said, "yes,

对于 AI 该如何思考这个问题确实很棘手,但历史上我们一直说「是的,

This is a positive-sum world, and it's still worth diffusing the technology.

这是一个正和世界,技术扩散仍然是值得的。

There are a number of choices we have. Framing this as a government-to-government decision in national security terms is one lens, but there are a lot of other lenses.

我们有很多选择。把这个问题框定为政府间的国家安全决策是一个视角,但还有很多其他视角。

You could imagine a world where we produce all these cures to diseases.

你可以想象这样一个世界:我们研发出所有这些疾病的治愈方法。

The cures are fine to sell to authoritarian countries, but the data centers just aren't.

这些治愈方法可以卖给威权国家,但数据中心不行。

The chips and the data centers aren't, and the AI industry itself isn't.

芯片和数据中心不行,AI 产业本身也不行。

Another possibility I think folks should think about is this.

我认为大家应该考虑的另一种可能性是这个。

Could there be developments we can make—either that naturally happen as a result of AI, or that we could make happen by building technology on AI—that create an equilibrium where it becomes infeasible for authoritarian countries to deny their people private use of the benefits of the technology?

我们能否实现某些进展——无论是 AI 自然带来的,还是我们通过在 AI 基础上构建技术来实现的——从而创造出一种平衡态,使得威权国家无法阻止其人民私下使用这项技术的益处?

Are there equilibria where we can give everyone in an authoritarian country their own AI model that defends them from surveillance and there isn't a way for the authoritarian country to crack down on this while retaining power?

是否存在这样的平衡态:我们可以给威权国家的每个人提供自己的 AI 模型来保护他们免受监控,而威权国家在保持权力的同时无法镇压这种情况?

I don't know. That sounds to me like if that went far enough, it would be a reason why authoritarian countries would disintegrate from the inside.

我不知道。在我看来,如果这种情况发展到一定程度,将会成为威权国家从内部瓦解的原因。

But maybe there's a middle world where there's an equilibrium where, if they want to hold on to power, the authoritarians can't deny individualized access to the technology.

但也许存在一个中间状态,在这种平衡下,如果威权者想要保持权力,就无法拒绝人们个性化地使用这项技术。

But I actually do have a hope for the more radical version.

但我确实对更激进的版本抱有希望。

Is it possible that the technology might inherently have properties—or that by building on it in certain ways we could create properties—that have this dissolving effect on authoritarian structures?

这项技术是否可能天然具有某些特性——或者我们可以通过特定方式在其基础上构建出某些特性——从而对威权结构产生瓦解作用?

Now, we hoped originally—think back to the beginning of the Obama administration—that social media and the internet would have that property, and it turns out not to.

回想 Obama 政府初期,我们最初曾希望社交媒体和互联网会具有这种特性,结果证明并非如此。

But what if we could try again with the knowledge of how many things could go wrong, and that this is a different technology?

但如果我们带着对诸多可能出错之处的认知,再加上这是一种不同的技术,再试一次呢?

I don't know if it would work, but it's worth a try.

我不知道能否奏效,但值得一试。

It's just very unpredictable. There are first principles reasons why authoritarianism might be privileged.

这只是非常难以预测。从第一性原理来看,威权主义可能具有某些优势。

It's all very unpredictable. We just have to recognize the problem and come up with 10 things we can try, try those, and then assess which ones are working, if any.

这一切都非常难以预测。我们只需要认清问题,想出 10 种可以尝试的方法,试一试,然后评估哪些有效(如果有的话)。

Then try new ones if the old ones aren't working.

如果旧方法不奏效,就尝试新方法。

But I guess that nets out to today, as you say, that we will not sell data centers, or chips, and the ability to make chips to China.

但我想这归结到今天,就像你说的,我们不会向中国出售数据中心、芯片,以及制造芯片的能力。

So in some sense, you are denying… There would be some benefits to the Chinese economy, Chinese people, et cetera, because we're doing that.

所以从某种意义上说,你是在剥夺……如果我们这样做,会给中国经济、中国人民等带来一些好处。

Then there'd also be benefits to the American economy because it's a positive-sum world. We could trade. They could have their country's data centers doing one thing. We could have ours doing another.

同时也会给美国经济带来好处,因为这是一个正和世界。我们可以交易。他们的国家可以用数据中心做一件事,我们可以用我们的做另一件事。

Already, you're saying it's not worth that positive-sum stipend to empower those countries?

你的意思已经是,这种正和收益不值得用来增强那些国家的实力?

What I would say is that we are about to be in a world where growth and economic value will come very easily if we're able to build these powerful AI models.

我想说的是,我们即将进入这样一个世界:如果我们能够构建这些强大的 AI 模型,增长和经济价值将来得非常容易。

What will not come easily is distribution of benefits, distribution of wealth, political freedom.

不容易实现的是利益分配、财富分配和政治自由。

These are the things that are going to be hard to achieve.

这些才是难以实现的东西。

So when I think about policy, I think that the technology and the market will deliver all the fundamental benefits—this is my fundamental belief—almost faster than we can take them.

所以当我思考政策问题时,我认为技术和市场会带来所有根本性的好处——这是我的基本信念——其速度几乎快到我们来不及接受。

These questions about distribution and political freedom and rights are the ones that will actually matter and that policy should focus on.

关于分配、政治自由和权利的这些问题才是真正重要的,也是政策应该关注的重点。

Speaking of distribution, as you were mentioning, we have developing countries. In many cases, catch-up growth has been weaker than we would have hoped for.

说到分配问题,正如你提到的,我们有发展中国家。在很多情况下,追赶式增长比我们期望的要弱。

But when catch-up growth does happen, it's fundamentally because they have underutilized labor.

但当追赶式增长确实发生时,根本原因是它们有未充分利用的劳动力。

We can bring the capital and know-how from developed countries to these countries, and then they can grow quite rapidly.

我们可以把发达国家的资本和技术诀窍带到这些国家,然后它们就能快速增长。

Obviously, in a world where labor is no longer the constraining factor, this mechanism no longer works.

显然,在一个劳动力不再是制约因素的世界里,这种机制就不再有效了。

So is the hope basically to rely on philanthropy from the people or countries who immediately get wealthy from AI? What is the hope?

那么希望是不是基本上要依靠那些因 AI 立即变富的人或国家的慈善事业?希望在哪里?

Philanthropy should obviously play some role, as it has in the past.

慈善事业显然应该发挥一定作用,就像过去一样。

But I think growth is always better and stronger if we can make it endogenous.

但我认为如果我们能让增长变得内生,效果总是会更好、更强劲。

What are the relevant industries in an AI-driven world?

在一个 AI 驱动的世界里,相关的产业是什么?

I said we shouldn't build data centers in China, but there's no reason we shouldn't build data centers in Africa. In fact, I think it'd be great to build data centers in Africa.

我说过我们不应该在中国建数据中心,但我们没有理由不在非洲建数据中心。事实上,我认为在非洲建数据中心是件很棒的事。

As long as they're not owned by China, we should build data centers in Africa. I think that's a great thing to do.

只要它们不归中国所有,我们就应该在非洲建数据中心。我认为这是一件很好的事情。

There's no reason we can't build a pharmaceutical industry that's AI-driven.

我们没有理由不能建立一个 AI 驱动的制药产业。

If AI is accelerating drug discovery, then there will be a bunch of biotech startups.

如果 AI 正在加速药物发现,那么就会有一批生物科技初创公司。

Let's make sure some of those happen in the developing world.

让我们确保其中一些发生在发展中国家。

Certainly, during the transition—we can talk about the point where humans have no role—humans will still have some role in starting up these companies and supervising the AI models.

当然,在过渡期间——我们可以讨论人类完全没有角色的那个时刻——人类在创办这些公司和监督 AI 模型方面仍然会有一定作用。

So let's make sure some of those humans are in the developing world so that fast growth can happen there as well.

所以让我们确保其中一些人在发展中国家,这样快速增长也能在那里发生。

You guys recently announced that Claude is going to have a constitution that's aligned to a set of values, and not necessarily just to the end user.

你们最近宣布 Claude 将有一个与一套价值观对齐的宪章,而不一定只是与终端用户对齐。

There's a world I can imagine where if it is aligned to the end user, it preserves the balance of power we have in the world today because everybody gets to have their own AI that's advocating for them.

我可以想象这样一个世界:如果它与终端用户对齐,就能保持我们今天在世界上的权力平衡,因为每个人都能拥有为自己辩护的 AI。

The ratio of bad actors to good actors stays constant. It seems to work out for our world today.

坏人和好人的比例保持不变。这似乎对我们今天的世界是行得通的。

Why is it better not to do that, but to have a specific set of values that the AI should carry forward?

为什么不这样做反而更好,而是要让 AI 承载一套特定的价值观?

I'm not sure I'd quite draw the distinction in that way. There may be two relevant distinctions here.

我不确定我会这样划分。这里可能有两个相关的区别。

I think you're talking about a mix of the two. One is, should we give the model a set of instructions about 'do this' versus 'don't do this'?

我认为你说的是两者的混合。一个是,我们应该给模型一套「做这个」还是「不要做这个」的指令?

The other is, should we give the model a set of principles for how to act?

另一个是,我们应该给模型一套关于如何行动的原则吗?

It's kind of purely a practical and empirical thing that we've observed.

这其实纯粹是我们观察到的一个实践和经验性的事情。

By teaching the model principles, getting it to learn from principles, its behavior is more consistent, it's easier to cover edge cases, and the model is more...

通过教给模型原则,让它从原则中学习,它的行为会更加一致,更容易覆盖边缘情况,而且模型会更加...

likely to do what people want it to do. In other words, if you give it a list of rules—"don't tell people how to hot-wire a car, don't speak in Korean"—it doesn't really understand the rules, and it's hard to generalize from them.

更有可能做人们想让它做的事。换句话说,如果你给它一系列规则——「不要告诉别人怎么偷车」「不要说韩语」——它并不真正理解这些规则,也很难从中归纳总结。

It's just a list of do's and don'ts. Whereas if you give it principles—it has some hard guardrails like "Don't make biological weapons" but—overall you're trying to understand what it should be aiming to do, how it should be aiming to operate.

这只是一堆能做和不能做的清单。而如果给它原则——虽然也有一些硬性底线,比如「不要制造生物武器」——但总体上你是在试图让它理解应该以什么为目标,应该如何运作。

So just from a practical perspective, that turns out to be a more effective way to train the model.

所以从实践角度来说,这被证明是训练模型更有效的方式。

That's the rules versus principles trade-off. Then there's another thing you're talking about, which is the corrigibility versus intrinsic motivation trade-off.

这就是规则与原则之间的权衡。然后你说的还有另一个问题,就是可纠正性与内在动机之间的权衡。

How much should the model be a kind of "skin suit" where it just directly follows the instructions given to it by whoever is giving those instructions, versus how much should the model have an inherent set of values and go off and do things on its own?

模型应该在多大程度上成为一个「外壳」,直接遵循任何指令发出者的指令,相对的,模型又应该在多大程度上拥有自己固有的价值观,并自主行动?

There I would actually say everything about the model is closer to the direction that it should mostly do what people want. It should mostly follow instructions.

在这一点上,我认为模型的所有设计都更倾向于它应该主要做人们想要的事。它应该主要遵循指令。

We're not trying to build something that goes off and runs the world on its own.

我们并不是要造一个自己跑出去管理世界的东西。

We're actually pretty far on the corrigible side. Now, what we do say is there are certain things that the model won't do.

我们实际上在可纠正性这一端已经走得很远了。不过,我们确实也说了,有些事情模型是不会做的。

I think we say it in various ways in the constitution, that under normal circumstances, if someone asks the model to do a task, it should do that task. That should be the default. But if you've asked it to do something dangerous, or to harm someone else, then the model is unwilling to do that.

我想我们在宪法中用各种方式表达过,在正常情况下,如果有人要求模型执行一项任务,它就应该去做。这应该是默认行为。但如果你要求它做危险的事,或者伤害他人,那模型就不愿意做。

So I actually think of it as a mostly corrigible model that has some limits, but those limits are based on principles.

所以我实际上把它看作一个基本可纠正的模型,有一些限制,但这些限制是基于原则的。

Then the fundamental question is, how are those principles determined? This is not a special question for Anthropic.

那根本问题就是,这些原则是如何确定的?这不是 Anthropic 特有的问题。

This would be a question for any AI company. But because you have been the ones to actually write down the principles, I get to ask you this question.

这会是任何 AI 公司都要面对的问题。但因为是你们实际写下了这些原则,所以我才向你提这个问题。

Normally, a constitution is written down, set in stone, and there's a process of updating it and changing it and so forth. In this case, it seems like a document that people at Anthropic write, that can be changed at any time, that guides the behavior of systems that are going to be the basis of a lot of economic activity.

通常,宪法是写下来的,固定不变的,有一套更新和修改的程序等等。而在这种情况下,它看起来像是 Anthropic 的人写的一份文件,可以随时修改,却指导着将成为大量经济活动基础的系统的行为。

How do you think about how those principles should be set?

你怎么看待这些原则应该如何制定?

I think there are maybe three sizes of loop here, three ways to iterate.

我认为这里可能有三个层次的循环,三种迭代方式。

One is we iterate within Anthropic. We train the model, we're not happy with it, and we change the constitution. I think that's good to do.

第一种是我们在 Anthropic 内部迭代。我们训练模型,对结果不满意,然后修改宪法。我认为这是应该做的。

Putting out public updates to the constitution every once in a while is good because people can comment on it.

时不时发布宪法的公开更新是好的,因为人们可以对此发表意见。

The second level of loop is different companies having different constitutions. I think it's useful. Anthropic puts out a constitution, Gemini puts out a constitution, and other companies put out a constitution.

第二个层次的循环是不同公司有不同的宪法。我认为这很有用。Anthropic 发布一个宪法,Gemini 发布一个宪法,其他公司也发布宪法。

People can look at them and compare. Outside observers can critique and say, "I like this thing from this constitution and this thing from that constitution."

人们可以看看它们并进行比较。外部观察者可以评论说,「我喜欢这个宪法里的这一点,喜欢那个宪法里的那一点。」

That creates a soft incentive and feedback for all the companies to

这为所有公司创造了一种软性激励和反馈机制,促使它们

Take the best of each element and improve. Then I think there's a third loop, which is

取各个要素中最好的部分并加以改进。然后我认为还有第三个循环,就是

society beyond the AI companies and beyond just those who comment without hard power.

AI 公司之外的社会,以及那些没有实权只能发表评论的人之外的群体。

There we've done some experiments. A couple years ago, we did an experiment with the Collective Intelligence Project to basically poll people and ask them what should be in our AI constitution.

在这方面我们已经做了一些实验。几年前,我们与 Collective Intelligence Project 合作做了一个实验,基本上就是对人们进行调查,询问他们认为我们的 AI 宪法里应该包含什么。

At the time, we incorporated some of those changes.

当时,我们采纳了其中一些意见。

So you could imagine doing something like that with the new approach we've taken to the constitution. It's a little harder because

所以你可以想象用我们对宪法采取的新方法做类似的事情。这会更难一些,因为

it was an easier approach to take when the constitution was a list of dos and don'ts.

当宪法是一系列该做和不该做的条目时,这种方法会更容易执行。

At the level of principles, it has to have a certain amount of coherence.

在原则层面,它必须具有一定程度的连贯性。

But you could still imagine getting views from a wide variety of people.

但你仍然可以想象从各种各样的人那里获取意见。

You could also imagine—and this is a crazy idea, but this whole interview is about crazy ideas—systems of representative government having input.

你还可以想象——这是个疯狂的想法,但这整个访谈就是关于疯狂想法的——让代议制政府系统参与进来提供意见。

I wouldn't do this today because the legislative process is so slow.

我现在不会这么做,因为立法程序太慢了。

This is exactly why I think we should be careful about the legislative process and AI regulation.

这正是为什么我认为我们应该对立法程序和 AI 监管保持谨慎。

But there's no reason you couldn't, in principle, say, "All AI models have to have a constitution that starts with these things, and then you can append other things after it, but there has to

但原则上没有理由不能这样说:「所有 AI 模型都必须有一个宪法,以这些内容开头,然后你可以在后面附加其他内容,但必须

be this special section that takes precedence." I wouldn't do that. That's too rigid and sounds

有这个具有优先权的特殊部分。」我不会那样做。那太僵化了,听起来

overly prescriptive in a way that I think overly aggressive legislation is.

过于规定性,就像我认为过于激进的立法那样。

But that is a thing you could try to do. Is there some much less heavy-handed

但那确实是你可以尝试做的事情。有没有一些不那么强硬的

version of that? Maybe. I really like control loop two.

版本呢?也许有。我真的很喜欢第二个控制循环。

Obviously, this is not how constitutions of actual governments do or should work.

显然,这不是真实政府的宪法运作的方式,也不应该是。

There's not this vague sense in which the Supreme Court will feel out how people are feeling—what are the vibes—and update the constitution accordingly.

不存在这种模糊的情况,即最高法院会感受人们的感觉——氛围如何——然后相应地更新宪法。

With actual governments, there's a more formal, procedural process.

对于真实的政府,有一个更正式、更程序化的过程。

But you have a vision of competition between constitutions, which is actually very reminiscent of how some libertarian charter cities people used to talk about what an archipelago of different kinds of governments would look like.

但你对宪法之间的竞争有一种愿景,这实际上很像一些自由意志主义特许城市倡导者过去谈论的,关于由不同类型政府组成的群岛会是什么样子。

There would be selection among them of who could operate the most effectively and where people would be the happiest.

它们之间会有选择,看谁能运作得最有效、人们在哪里会最幸福。

In a sense, you're recreating that vision of a utopia of archipelagos.

从某种意义上说,你正在重现那种群岛乌托邦的愿景。

I think that vision has things to recommend it and things that will go wrong with it.

我认为这个愿景既有可取之处,也有会出问题的地方。

It's an interesting, in some ways compelling, vision, but things will go wrong that you hadn't imagined.

这是一个有趣的、在某些方面很有吸引力的愿景,但会出现你没有想到的问题。

So I like loop two as well, but I feel like the whole thing has got to be some mix of loops one, two, and three, and it's a matter of the proportions.

所以我也喜欢第二个循环,但我觉得整个体系必须是第一、第二和第三个循环的某种混合,这是一个比例问题。

I think that's gotta be the answer.

我认为这必须是答案。

When somebody eventually writes the equivalent of The Making of the Atomic Bomb for this era, what is the thing that will be hardest to glean from the historical record that they're most likely to miss?

当有人最终为这个时代写出相当于《原子弹的诞生》那样的作品时,从历史记录中最难提取、他们最有可能遗漏的是什么?

I think a few things. One is,

我认为有几件事。一个是,

at every moment of this exponential, the extent to which the world outside it didn't understand it.

在这个指数增长的每个时刻,外部世界不理解它的程度有多深。

This is a bias that's often present in history. Anything that actually happened looks inevitable in retrospect.

这是历史中经常存在的一种偏差。任何实际发生的事情在回顾时都显得不可避免。

When people look back, it will

当人们回顾时,

It'll be hard for them to put themselves in the place of people who were actually making a bet on this thing to happen that wasn't inevitable, that we had these arguments like the arguments I make for scaling or that continual learning will be solved.

他们很难设身处地去理解那些真正押注于这件事会发生的人——这件事并非必然会发生,我们当时提出的论点,比如我关于扩展定律的论证,或者持续学习问题终将被解决的论点。

Some of us internally put a high probability on this happening, but there's a world outside us that's not acting on that at all.

我们内部有些人认为这件事发生的概率很高,但外面的世界完全没有基于这个判断来行动。

I think the weirdness of it, unfortunately the insularity of it...

我觉得这件事的诡异之处,不幸的是它的封闭性...

If we're one year or two years away from it happening, the average person on the street has no idea.

如果我们距离它发生只有一两年的时间,街上的普通人对此毫无概念。

That's one of the things I'm trying to change with the memos, with talking to policymakers.

这正是我试图通过写备忘录、与政策制定者对话来改变的事情之一。

I don't know, but I think that's just a crazy thing.

我不知道,但我觉得这真是件疯狂的事。

Finally, I would say—and this probably applies to almost all historical moments of crisis—how absolutely fast it was happening, how everything was happening all at once.

最后我想说——这可能适用于几乎所有历史性的危机时刻——就是一切发生得如此之快,所有事情都同时发生。

Decisions that you might think were carefully calculated, well actually you have to make that decision, and then you have to make 30 other decisions on the same day because it's all happening so fast.

那些你可能以为是经过深思熟虑的决策,实际上你必须当场做出决定,然后同一天还要做出其他30个决定,因为一切发生得太快了。

You don't even know which decisions are going to turn out to be consequential.

你甚至不知道哪些决策最终会产生重大影响。

One of my worries—although it's also an insight into what's happening—is that some very critical decision will be some decision where someone just comes into my office and is like, "Dario, you have two minutes. Should we do thing A or thing B on this?"

我的一个担忧——尽管这也让我洞察到正在发生什么——就是某个非常关键的决策可能是这样的场景:有人冲进我办公室说,「Dario,你有两分钟时间。这件事我们该做A还是B?」

Someone gives me this random half-page memo and asks, "Should we do A or B?" I'm like, "I don't know. I have to eat lunch. Let's do B."

有人递给我一份随便写的半页备忘录,问我「我们该做A还是B?」我心想,「我不知道。我得去吃午饭了。那就选B吧。」

That ends up being the most consequential thing ever.

结果那竟然成了最重大的决策。

So final question. There aren't tech CEOs who are usually writing 50-page memos every few months.

那么最后一个问题。通常科技公司CEO不会每隔几个月就写50页的备忘录。

It seems like you have managed to build a role for yourself and a company around you which is compatible with this more intellectual-type role of CEO.

看起来你成功地为自己塑造了一个角色,并围绕你建立了一家公司,这种模式与更偏智识型的CEO角色相契合。

I want to understand how you construct that. How does that work? Do you just go away for a couple of weeks and then you tell your company, "This is the memo. Here's what we're doing"?

我想了解你是如何构建这种模式的。这是怎么运作的?你是不是消失几周,然后回来告诉公司,「这是备忘录。我们就这么干」?

It's also reported that you write a bunch of these internally.

据说你在内部也写了很多这样的备忘录。

For this particular one, I wrote it over winter break.

就这份特定的备忘录而言,我是在寒假期间写的。

I was having a hard time finding the time to actually write it.

我当时很难找到时间真正去写它。

But I think about this in a broader way. I think it relates to the culture of the company.

但我是从更广阔的角度来思考这个问题的。我认为这关系到公司的文化。

I probably spend a third, maybe 40%, of my time making sure the culture of Anthropic is good.

我可能花三分之一,也许40%的时间来确保Anthropic的文化健康。

As Anthropic has gotten larger, it's gotten harder to get directly involved in the training of the models, the launch of the models, the building of the products.

随着Anthropic规模变大,我越来越难以直接参与模型训练、模型发布、产品构建等具体工作。

It's 2,500 people. I have certain instincts, but it's very difficult to get involved in every single detail.

公司有2500人。我有一些直觉判断,但很难介入每一个细节。

I try as much as possible, but one thing that's very leveraged is making sure Anthropic is a good place to work, people like working there, everyone thinks of themselves as team members, and everyone works together instead of against each other.

我尽可能去尝试,但有一件事杠杆效应很强,那就是确保Anthropic是个好的工作场所,让大家喜欢在这里工作,每个人都把自己当作团队成员,大家一起协作而不是相互对抗。

We've seen as some of the other AI companies have grown—without naming any names—we're starting to see decoherence and people fighting each other.

我们看到其他一些AI公司在发展过程中——我就不点名了——开始出现失序现象,人们相互争斗。

I would argue there was even a lot of that from the beginning, but it's gotten worse.

我认为从一开始就有很多这种问题,但现在变得更糟了。

I think we've done an extraordinarily good job, even if not perfect, of holding the company together, making everyone feel the mission, that we're sincere about the mission, and that everyone has faith that everyone else there is working for the right reason.

我认为我们做得非常好,即使不完美,我们把公司凝聚在一起,让每个人都能感受到使命,相信我们对使命是认真的,每个人都相信其他人是出于正确的理由在这里工作。

That we're a team, that people aren't trying to get ahead at each other's expense or backstab each other, which again, I think happens a lot at some of the other places.

我们是一个团队,大家不会为了自己的利益而互相倾轧或背后捅刀子,而这种事在其他一些公司里我觉得经常发生。

How do you make that the case?

你是如何做到这一点的?

It's a lot of things. It's me, it's Daniela, who runs the company day to day, it's the co-founders, it's the other people we hire, it's the environment we try to create.

这涉及很多方面。包括我自己,包括 Daniela——她负责公司的日常运营,还有联合创始人们,我们招聘的其他人,以及我们努力营造的环境。

But I think an important thing in the culture is that the other leaders as well, but especially me, have to articulate what the company is about, why it's doing what it's doing, what its strategy is, what its values are, what its mission is, and what it stands for.

但我认为在文化建设中很重要的一点是,其他领导者,尤其是我,必须清楚地阐明公司的定位是什么,为什么要做现在做的事,我们的战略是什么,我们的价值观是什么,我们的使命是什么,以及我们代表什么。

When you get to 2,500 people, you can't do that person by person. You have to write, or you have to speak to the whole company.

当公司有 2500 人的时候,你不可能逐一去沟通。你必须通过文字,或者面向全公司讲话。

This is why I get up in front of the whole company every two weeks and speak for an hour.

这就是为什么我每两周会站在全公司面前讲一个小时。

I wouldn't say I write essays internally. I do two things. One, I write this thing called a DVQ, Dario Vision Quest. I wasn't the one who named it that.

我不会说我在内部写文章。我主要做两件事。第一件,我会写一个叫 DVQ 的东西,Dario Vision Quest。这个名字不是我起的。

That's the name it received, and it's one of these names that I tried to fight because it made it sound like I was going off and smoking peyote or something. But the name just stuck.

这是大家给它起的名字,我曾经试图反对这个名字,因为听起来像是我跑去吸食仙人掌毒品之类的。但这个名字就这么固定下来了。

So I get up in front of the company every two weeks. I have a three or four-page document, and I just talk through three or four different topics about what's going on internally, the models we're producing, the products, the outside industry, the world as a whole as it relates to AI and geopolitically in general. Just some mix of that.

所以我每两周会站在公司面前。我会准备一份三到四页的文档,然后讲三到四个不同的话题,包括公司内部正在发生什么,我们正在开发的模型,我们的产品,外部行业动态,以及整个世界与 AI 相关的情况和总体的地缘政治局势。就是这些内容的某种组合。

I go through very honestly and I say, "This is what I'm thinking, and this is what Anthropic leadership is thinking," and then I answer questions.

我会非常坦诚地讲,我会说「这是我的想法,这是 Anthropic 领导层的想法」,然后回答问题。

That direct connection has a lot of value that is hard to achieve when you're passing things down the chain six levels deep.

这种直接的连接非常有价值,如果信息要通过六层管理链条逐级传递,就很难做到这一点。

A large fraction of the company comes to attend, either in person or virtually.

公司很大一部分员工会参加,有的现场参加,有的远程参加。

It really means that you can communicate a lot. The other thing I do is I have a channel in Slack where I just write a bunch of things and comment a lot.

这意味着你可以传达很多信息。我做的另一件事是在 Slack 上有一个频道,我会在那里写很多内容,发表很多评论。

Often that's in response to things I'm seeing at the company or questions people ask.

通常是针对我在公司看到的事情或者员工提出的问题做出回应。

We do internal surveys and there are things people are concerned about, and so I'll write them up.

我们会做内部调研,员工会有一些关切的问题,我就会把这些写出来。

I'm just very honest about these things. I just say them very directly.

我对这些事情非常坦诚。我会非常直接地说出来。

The point is to get a reputation of telling the company the truth about what's happening, to call things what they are, to acknowledge problems, to avoid the sort of corpo speak, the kind of defensive communication that often is necessary in public because the world is very large and full of people who are interpreting things in bad faith.

重点是要建立一种声誉,就是告诉公司正在发生的事情的真相,直言不讳地称呼事物本身,承认问题,避免那种公司官话,那种在公开场合往往必要的防御性沟通——因为外部世界很大,充满了恶意曲解的人。

But if you have a company of people who you trust, and we try to hire people that we trust, then you can really just be entirely unfiltered.

但如果公司里都是你信任的人,而我们也努力招聘我们信任的人,那你就可以完全不加过滤地沟通。

I think that's an enormous strength of the company.

我认为这是公司的一个巨大优势。

It makes it a better place to work, it makes people more than the sum of their parts, and increases the likelihood that we accomplish the mission because everyone is on the same page about the mission, and everyone is debating and discussing how best to accomplish the mission.

这让公司成为一个更好的工作场所,让团队产生大于各部分之和的效果,并且增加了我们完成使命的可能性,因为每个人对使命的理解是一致的,每个人都在辩论和讨论如何最好地完成使命。

Well, in lieu of an external Dario Vision Quest, we have this interview.

那么,作为外部版 Dario Vision Quest 的替代,我们有了这次访谈。

This interview is a little like that.

这次访谈确实有点像那样。

This has been fun, Dario. Thanks for doing it.

这次很愉快,Dario。感谢你接受访谈。

Thank you, Dwarkesh.

谢谢你,Dwarkesh。