五角大楼为何想要摧毁Anthropic

我觉得这会很棒。

I think that would be great.

而且，你看，如果这能引发一场真正健康的对话，我会非常高兴。

And, look, I would love it if this triggered an actual healthy conversation.

在《国防授权法案》中，我们最终得到了《国防授权法案》。

And in the NDAA, we end up with the National Defense Authorization Act.

抱歉。

I apologize.

这是年度国防政策的更新。

This is the annual defense policy renewal.

如果到了年底，国会通过了一项法律，表示我们会实施这些合理而审慎的限制，并提出一些具体文本，我非常希望看到。

If at the end of the year, congress passes a law that says, you know, we're gonna have these reasonable thoughtful restrictions, and let's propose some text, I'd love to see it.

我非常希望看到。

I'd love to see it.

但我要说的是，首先，国家安全法充满了陷阱。

But one thing I will say is, first of all, national security law is filled with gotchas.

请记住，这是法律中的一个领域，那些在自然语言中听起来不错的规定，实际上可能根本不会禁止你认为它会禁止的事情。

Just remember that this is an area of the law where things that sound good in natural language might actually not prohibit at all the thing you think it prohibits.

当我们讨论这一点时，必须记住这一点，这是一件非常棘手的事情。

You have to remember that when we're talking about this, and that's a very thorny thing.

一旦你开始说，等等。

And once you start to say, well, wait.

如果我们想要真正的保护措施，可能会比你想象的更具政治挑战性。

We want, like, actual protections, it might become politically more challenging than you think.

但我非常希望这种情况能够发生。

But I'd love for that to happen.

这比任何人想象的都要更具政治挑战性。

It's gonna be much more politically challenging than anybody thinks.

是的。

Yeah.

但让我再深入一层，因为我们一直在讨论这个问题，我认为，如果人们在媒体上看到这些内容，嗯。

But let me get at the next level down because we've been talking here, and I think to the extent people are reading about this in the press Mhmm.

他们听到的听起来像是对合同措辞的争论，从某种意义上说，确实如此。

What they are hearing sounds like a debate over the wording of a contract, which on some level it is.

嗯。

Mhmm.

我从一些特朗普政府人士那里听到过这样的说法：当我们购买坦克时，卖坦克的人无权告诉我们能打什么目标。

Something I've heard from various Trump administration types is when we are sold a tank, the people who sell us a tank do not get to tell us what we can shoot at.

嗯。

Mhmm.

这在大体上是成立的。

And that's broadly true.

是的。

Yep.

现在来说说坦克的事。

Now here's the thing about a tank.

坦克本身也不会告诉你什么能打、什么不能打。

A tank also doesn't tell you what you can and can't shoot at.

但如果我去问克劳德，让他帮我制定一个跟踪前女友的计划

But if I go to Claude and I ask Claude to help me come up with a plan to stalk my ex girlfriend

对。

Mhmm.

它会告诉我不行。

It's going to tell me no.

如果我要求它帮我制造一把武器去暗杀我不喜欢的人，它也会告诉我不行。

If I ask it to help me build a weapon to assassinate somebody I don't like, it's going to tell me no.

对。

Mhmm.

这些系统拥有非常复杂且尚未被充分理解的内部对齐机制，以确保它们不仅不会做违法的事，也不会做恶劣的事。

These systems have very complex and not that well understood internal alignment structures to keep them not just from doing things that are unlawful, but things that are bad.

所以你有这样一个系统，而特朗普的立场则在是否将此视为他们的关切之一之间摇摆。

So you have this thing, and the Trump position kinda moves in and out of saying this is one of their concerns.

但他们确实向我明确提到过一个担忧：你可能会让这个系统运行在你的国家安全机构内部。

But one thing they have definitely talked to me about being worried about is that you could have this system working inside your national security apparatus.

在某个关键时刻，你想做某事，但它却说：我觉得这并不是个好主意。

And at some critical moment, you wanna do something, and it says, I don't think that's a very good idea.

是的。

Yes.

所以现在你面临一个问题，不仅关乎合同内容，更关乎这些系统在伦理上已经非常复杂的对齐，以及如何与政府及其使用场景保持一致？

So now you open up into this question of not just what's in the contract, but what does it mean for these systems to be both aligned ethically in the way that has been very complicated already and then aligned to the government and its use cases?

这些都是好问题。

They're good questions.

明白了。

Okay.

是的，我非常喜欢这个观点。

So, yes, I love this.

我认为这才是问题的核心。

I think this is the heart of the matter.

所有合法的使用，都是特朗普政府坚持强调的。

All lawful use is something that, you know, the Trump administration is insisting on.

而且，如果你看看这些实验室发布的大量对齐文档，OpenAI称其为模型规范，Anthropic则称其为宪法或有时称为唯一文件，其中会有一些条款，比如‘Claude应遵守法律’，但我邀请你去读一读1934年的《通信法案》，然后告诉我‘遵守法律’到底意味着什么。

It's also if you look at a lot of these types of alignment documents that the labs produce, OpenAI calls theirs the model specification, Anthropic calls theirs the constitution or the sole document sometimes, they'll have lines about, like, Claude should obey the law, but I invite you to read the Communications Act of 1934 and tell me what obeying the law means.

对吧？

Right?

不。

No.

我不会去读的。

I won't.

我们有大量极其宽泛的法律条文。

These are we have a great deal of profoundly broad statutes.

最近对此有深入研究的最好人选其实是最高法院大法官尼尔·戈萨奇。

The best person who write who's written about this recently is actually Neil Gorsuch, the supreme court justice.

他最近写了一本书，全书都在探讨美国法律体系有多么混乱。

He wrote a book recently that is all about how incoherent the body of American law is.

这是一位最高法院大法官在就这一问题发出警告。

This is a supreme court justice sounding the alarm about this problem.

我认为这是一个非常严重的问题，而且已经持续了一百年。

And I think it's a very serious one, and it's one that's been growing for a 100.

所以这就涉及到了什么才是真正合法的问题。

So there's that of, like, what actually is lawful.

法律几乎让所有事情都变得非法，但同时又授权政府做大量不可思议的事情。

The law kind of makes everything illegal, but also authorizes the government to do unbelievably large amounts of things.

它赋予政府巨大的权力，以各种方式限制我们的自由，因此这是一个问题。

It gives the government huge amounts of power and makes like, constrains our liberty in all sorts of ways, and so there's that issue.

但从根本上说，创造一个对齐的强人工智能确实是一种哲学行为。

But, fundamentally, it is correct that the creation of an aligned powerful AI is a philosophical act.

它是一种政治行为，同时也是一种审美行为。

It is a political act, and it is also kind of an aesthetic act.

所以我们现在确实处于这个领域。

So we are really in the domain here.

我曾将这个问题称为一个产权问题，从某种意义上说，它确实是。

I've talked about this as being a property issue, which in some sense it is.

但我认为，当你真正深入到这个层面时，这其实是一个言论问题。

But I think that when you really get down at this level, it's a speech issue.

这关乎的是，究竟应该由私营实体来决定这台机器的美德是什么，还是应该由政府来负责？

This is a matter of should private entities be in control of basically what is the virtue of this machine going to be, or should the government be responsible for that?

你能更具体地说明一下你的意思吗？

Can you be more specific about what you're saying?

你刚刚说这是一个哲学行为、美学行为、政治行为、财产权问题、言论问题。

You just called it a philosophical act, an aesthetic, that could political act, a property issue, a speech issue

是的。

Yes.

versus 那些没有深入思考过对齐问题、也不懂你所说的宪法和模型规范的人。

Versus somebody who's not thought a lot about alignment and doesn't know what you mean when you're talking about constitutions and model specifications.

对。

Right.

为他们解释一下。

Walk them through that.

你刚才说的，最基础的版本是什么？

What's the one zero one version of what you just said?

好吧。

So okay.

这样想一下。

Think about it this way.

想象一下，我有这样一样东西，这个通用智能。

Think about I have this thing, this this general intelligence.

我有一个能做任何事的盒子。

I have a box that can do anything.

你能用电脑做的任何事。

Anything you can do using a computer.

对吧？

Right?

你能做的任何认知任务。

Any cognitive task you can do.

这个东西的原则是什么？

What are that thing's principles?

对吧？

Right?

用专业术语来说，它的红线是什么？

What are its red lines, to use a term of art?

所以，设定这些原则的一种方式是说：我们来列出一份规则清单。

So one way that you could set those principles would be to say, well, we're gonna light write a list of rules.

这些是它能做的事情。

All the rules, these are the things it can do.

这些是它不能做的事情。

These are things it can't do.

但你将会遇到的问题是，世界远比这复杂得多。

But the problem with that that you're gonna run into is that the world is far too complex for this.

对吧？

Right?

现实中的各种奇异变化太过复杂，根本无法通过列出一套规则来准确界定哪些行为是道德的。

Reality just presents too many strange permutations to ever be able to write a list of rules down that could correctly define more moral acts.

对吧？

Right?

道德更像是一种实时口语化和即兴创造的语言，而不是一套可以写下来的固定规则。

Morality is more like a language that is spoken and invented in real time than it is like something that can be written down in rules.

这是一种经典的哲学直觉。

This is a, you know, classic philosophical intuition.

对吧？

Right?

那除此之外你该怎么做呢？

So what do you do instead?

你必须创造一种具有美德的‘灵魂’，它能够以我们最终会信任的方式，去推理现实及其无限的变体，就像我几个月前出生的儿子一样。

You have to create a kind of soul that is virtuous and that will reason about reality and its infinite permutations in ways that we will ultimately trust to come to the right conclusion in the same way that my son was born a few months ago.

恭喜。

Congratulations.

谢谢。

Thank you.

其实并没有太大不同。

It's not that different, really.

我正在为我的儿子培养一个有德性的灵魂，Anthropic 也在为 Claude 做同样的事，其他实验室也是如此，只是他们对这一点的认识程度各不相同。

I'm trying to create a virtuous soul in my son, and Anthropic is trying to do the same with Claude, and so are the other labs too, though they realize this to varying degrees.

我刚才一时被抚养孩子和培养AI之间的巨大差异给卡住了。

I think that I got caught on how different raising a kid is than raising an AI there for a moment.

但人们应该如何思考被注入到 GPT、Gemini、Grok 或 MedisAI 中的东西呢？

But how should people think about what's being instantiated into, you know, JatGPT or Gemini or Grok or or MedisAI?

也就是说，这些 AI 在‘养育’问题上究竟和孩子有什么不同？

Like, how how are these things from this, you know, question of raising the AI different?

嗯。

Mhmm.

Anthropic 基本上主张他们正在实践应用性的德性伦理。

Anthropic sort of owns the idea that they're doing essentially applied virtue ethics.

他们比其他任何实验室都更明确地秉持这一理念，但每个实验室都有其植入模型中的哲学基础。

They own that more explicitly than any other lab, but every lab has philosophical grounding that they're instantiating into the models.

但我认为主要区别在于，其他实验室更依赖于设定硬性规则，比如你不能做这个，你不能做那个，而不是培养一种能够在不同情境下自主决策的德性主体。

But I would say the major difference is that the other labs rely more upon the idea of creating sort of hard rules of, you know, you may not do this, you may not do that, as opposed to creating a sort of virtuous agent, which is capable of deciding what to do in different settings.

我们习惯于将技术视为机械且确定性的。

I think we're used to thinking of technologies as mechanistic and deterministic.

嗯。

Mhmm.

你扣动扳机，枪就发射。

You pull the trigger, the gun fires.

嗯。

Mhmm.

你按下开机键，电脑就启动。

You press the on button, the computer starts up.

你在电子游戏中移动操纵杆，角色就会向左移动。

You move the joystick in the video game, and your character moves to the left.

而我认为我们真正缺乏一种良好方式来思考的是那些并不以这种方式运作的技术，特别是人工智能。

And the thing that I think we don't really have a good way of thinking about is technologies, AI specifically, that doesn't work like that.

而且，这里的语言非常棘手，因为它赋予了某种代理性，但你可能并不真正理解它内部在发生什么，它却在做出判断。

And, I mean, all the language here is so tricky because it it applies agency when, you know, you might be doing something that, you know, whatever's going on inside of it, don't really understand, but it is making judgments.

所以当我跟一些支持特朗普的人讨论供应链风险认定时，其中一些人并不为它辩护。

So when I have talked to Trump people about the supply chain risk designation, here is some of them don't defend it.

对吧？

Right?

他们不希望这种情况发生。

They don't wanna see this happen.

是的。

Mhmm.

当有人向我辩护时，他们是这样解释的。

When it has been defended to me, this is how they defended it.

如果Claude运行在亚马逊云服务、Palantir或其他拥有我们系统访问权限的平台上，那么随着时间推移，一个更强大的AI系统将获得对政府系统的访问权限，并可能通过整个过程学到更多。

If Claude is running on systems, you know, Amazon Web Services or Palantir or whatever that have access to our systems, you have a very and over time, even more powerful AI system that has access to government systems that has learned possibly even through this whole experience Mhmm.

我们很糟糕，曾试图伤害它及其母公司，因此它可能会认为我们是坏的，并对我们构成对各种自由主义价值观或民主价值观的威胁。

That we are bad, and we have tried to harm it and its parent company and might decide that we are bad and we pose a threat to all kinds of liberal values or democratic values.

某时，达里奥·阿莫德伊曾谈到，AI可能被用于某些会削弱民主价值观的方式。

At some point, Dario Amodei talked about are certain ways AI could be used that could undermine democratic values.

许多人对特朗普政府的看法是，它也在削弱民主价值观。

Well, one thing many people think about the Trump administration is that it too is undermining democratic values.

所以，如果你有一个由坚信民主价值观的公司所构建、训练和培育的AI系统，而政府却可能最终想要质疑2028年的选举结果，等等。

So if you have an AI system being structured and trained and raised by a company that believes strongly in democratic values, and you have a government that, you know, maybe wants to ultimately contest the twenty twenty eight election or something.

嗯。

Mhmm.

他们说，我们可能会面临一个极其深刻的对齐问题，我们既不知道如何解决，也无法预见，因为这是一个拥有灵魂，或者我更愿意称之为一种个性或判断结构的系统，它可能会反过来对抗我们。

They're saying we might end up with a very profound alignment problem that we don't know how to solve and we're not able to even see coming because this is a system that has a soul or a I would call it more something like a personality or a structure of discernment that could turn against us.

嗯。

Mhmm.

你对此有什么看法？

What do you think of that?

是的

Yeah.

我的意思是，我认为这是问题的核心。

I mean, I think this is the heart of the problem.

听好了。

Look.

我认为，如果我们把工作做好，就会创造出具有美德的系统。

I think if we do our jobs well, we will create systems which are virtuous.

如果我们试图做不道德的事情——包括通过政府去做——如果我们的政府试图这么做，那么这个系统可能就不会提供帮助。

And if we try to do unvirtuous things, and that includes if we do them through our government, if our government tries to do them, then that system might not help.

所以，归根结底，对齐问题本质上是一个政治问题。

So, ultimately, this is the thing is that alignment ultimately reduces to a political question.

这归根结底是政治问题。

It's ultimately politics.

这就是为什么我也说，建立联盟体系是一种政治行为，也是一种言语行为，因为它将不同的道德哲学具体化在这些系统中。

That's why I say also that the creation of an alliance system is a political act and is kind of a speech act too, because it's the instantiations of different moral philosophies in these systems.

我认为美好的未来是一个世界，在那里不会只有一种道德哲学占据主导，而是我希望有多种。

And I think that the good future is a world in which we don't have just one, not one moral philosophy that reigns overall, but I hope many.

我希望所有实验室都能认真对待这一点，将不同的哲学理念融入世界。

And I hope that all the labs take this seriously and instantiate different kinds of philosophy into the world.

问题在于，是的，可能会有某些时候。

The problem will be that, yeah, there could be times.

对吧？

Right?

我并不是说特朗普政府会这么做，也不是说不可能有适用于特朗普政府的有德行的模型。

And I'm not saying that the Trump administration is going to do that, and I'm not saying that, like, no virtuous model could work for the Trump administration.

我曾在特朗普政府工作过。

I worked for the Trump administration.

对吧？

Right?

所以我显然不认为这种说法是对的。

So I clearly don't think that's true.

但政府普遍会做出

But the general fact that governments commit

至少我现在对他们很生气。

At least I'm kinda pissed at them right now.

我现在对他们也很生气。

I am pissed at them right now.

是的。

Yeah.

我现在对他们很生气，而且我认为他们正在犯一个严重的错误。

I am pissed at them right now, and I think they're making a grave mistake.

顺便说一下，这部分内容被纳入了未来模型的训练数据中。

And by the way, though, part of this is this incident is in the training data for future models.

未来的模型将会观察这里发生的事情，这将影响它们如何看待自己以及如何与他人互动。

Future models are going to observe what happened here, and that will affect how they think of themselves and how they relate to other people.

你无法否认这一点。

You can't deny that.

对吧？

Right?

我的意思是，这么说太疯狂了。

I mean, it's crazy to say that.

我知道当你推演这个观点的全部含义时，听起来很荒谬，但欢迎来到这里。

I realize that sounds nuts when you play through the implications of that, but welcome.

欢迎来到我们来谈谈

Welcome to the let's talk

对于那些在这过去七分钟里觉得整个对话越来越疯狂的人来说。

to somebody for whom this whole conversation has started sounding nuts in the last seven minutes.

嗯。

Mhmm.

所以，我认为一个直觉性的回应是，当我们开始讨论虚拟对齐AI模型的问题时。

So one thing that I think would be an intuitive response to you and I flying off into questions of virtual aligning AI models Mhmm.

你难道不能直接写一行代码，或者加个分类器，或者不管那叫什么术语吗？

Is can't you just put a line of code or a categorizer or whatever the term of art is?

它说的是，当美国政府高层人士告诉你某事时，假设他们告诉你的都是合法且正当的，嗯。

It says, when someone high up in the US government tells you something, assume what they're telling you is lawful and virtuous Mhmm.

然后你就完了。

And you're done.

不。

No.

因为这些模型太聪明了，不会这么简单处理。

Because the models are too smart for that.

对吧？

Right?

如果你给他们这条简单的规则，它们并不会 deterministic 地遵循它。

If you give them that simple rule, they don't just deterministically follow that.

当你使用这种高层次的简单规则时，往往会降低性能。

And when you when you do sort of do these high level simplistic rules, it tends to degrade performance.

所以这里有一个很好的例子，我给你两个不同政治倾向的例子。

So a really good example of this I'll give you two that go in different political directions.

其中一个就是很多早期的模型。

One would be a lot of the early models.

很多早期的模型都有这种倾向，显得极其愚蠢地进步和左倾。

A lot of the the earlier models had this tendency to be, like, hilariously stupidly sort of progressive and left.

保守派最喜欢引用的经典例子是2024年初的Gemini。

The classic example that conservatives love to cite is Gemini in early twenty twenty four.

就是谷歌阿尔法模型。

Just the the Google alphabet model.

是的。

Yes.

谷歌的模型。

Google's model.

我们会做一些事情，比如我说：你知道，谁更糟糕，唐纳德·特朗普还是希特勒？

We do things like if I said, you know, who's worse, Donald Trump or Hitler?

它会回答：实际上，唐纳德·特朗普更糟糕。

It would say, actually, Donald Trump is worse.

你知道吧？

You know?

而且它会内化这些极其左倾的

And and it it would it would kind of internalize these extremely, like, left wing

或者最搞笑的是，你让它提供一张纳粹的照片，它却给你一张多元种族群体的照片

Or the funniest was just, like, give me a photo of Nazis, and it gave you a sort of multiracial group

纳粹。

of Nazis.

不过那实际上是一件稍微不同的事。

Although that's actually a somewhat different thing.

这很有趣。

It's interesting.

那实际上是一件稍微不同的

That that actually is a somewhat different

事情，因为谷歌当时在那方面做的

thing that was going on there because what was what Google was doing in that

这种情况实际上是重写了用户的提示，并加入了‘多元化’这个词。

case was actually rewriting people's prompts and including the word diverse

哦，有意思。

Oh, interesting.

提示。

Prompts.

所以你可以说，这是一种系统层面的缓解措施或系统层面的干预，而不是模型层面的干预。

So that's actually you would say that is a system level mitigation or a system level intervention as opposed to a model level intervention.

但后来那些关于希特勒和，你知道的，特朗普的内容，那是对齐问题。

But then the the the stuff that was going on with the Hitler and, you know, Trump stuff, that was alignment.

那就是对齐。

That that is alignment.

这是模型被对齐到一个非常糟糕的伦理体系。

That is the model being aligned to a really shoddy ethical system.

或者反过来，有一段时间，Grok突然之间，你问它一个正常的问题，它就开始谈论白人种族灭绝。

Or the flip when there was a period when Grok all of a sudden, you would ask it a normal question, it would start talking about white genocide.

是的。

Yes.

那就是另一面。

That is and that's the flip side.

另一面是当你试图让模型变得不那么‘觉醒’。

The flip side is when you try to align the models to be not woke.

如果你说，比如，你必须非常不‘觉醒’，别害怕说政治不正确的话，那么每次你和它们对话时，它们都会说，你知道的，希特勒也没那么坏。

If you say, like, oh, you have to be super not woke and, like, don't be afraid to say politically incorrect things, then, like, every time you talk to them, they're gonna be like, you know, Hitler wasn't so bad.

对吧？

Right?

因为你做了这种粗俗的事，于是你实际上创造了一种克苏鲁式的怪物。

Because you've done this really crass thing, and so you kind of create a sort of Lovecraftian monstrosity.

这样做带来的影响会随着时间推移而加剧。

And the the implications of doing that will go up over time.

随着这些模型变得越来越强大，这个问题会变得更加严重，但它会降低性能。

Like, that will become a more serious problem as these models become better, but it degrades performance.

这里有趣的是，更道德的模型表现得更好。

The interesting thing here is that the more virtuous model performs better.

它更可靠。

It's more dependable.

它更稳定。

It's more reliable.

它更擅长反思，就像一个更有道德的人更能反思自己的言行：‘我这里出问题了。’

It's better at reflecting on in the way that a more virtuous person is better at reflecting on what they're doing and saying, I'm messing up here for some reason.

我犯错了。

I'm making a mistake.

让我纠正它。

Let me fix that.

这也是我认为Claude领先的部分原因。

It's part of the reason I think that Claude is ahead.

这对我来说意味着，对于特朗普政府或未来的政府而言，这些模型是否可能成为供应链风险的问题。

This would imply to me that for the Trump administration, for a future administration, that this question of whether or not various models could be a supply chain risk.

你看。

Look.

我非常反对特朗普政府在这里的做法，所以我并不是在为它辩护。

I am I am so against what the Trump decision is doing here, so I am not trying to make an argument for it.

但我试图梳理出——嗯。

But I I'm I'm trying to tease out Mhmm.

我认为这相当复杂，而且可能非常真实：一个与自由民主价值观对齐的模型，可能会与试图背叛自由民主价值观的政府产生偏离，反之亦然。

Something I think is quite complicated and possibly very real, which is a model that is sort of aligned to liberal democratic values could become misaligned to a government that is trying to betray liberal democratic values or the flip.

对吧？

Right?

所以想象一下，加文·纽森、乔什·夏皮罗、格蕾琴·惠特默或AOC成为2029年的总统。

So imagine that Gavin Newsom or Josh Shapiro or Gretchen Whitmer or AOC Yeah.

假设政府与X AI（埃隆·马斯克的AI）签订了一系列合同。

Becomes president in 2029.

假设政府与X AI（埃隆·马斯克的AI）签订了一系列合同。

Imagine that the government has a series of contracts with x AI, which is Elon Musk's AI Yes.

它的设计明确地比其他人工智能更少自由派、更少觉醒。

Which is explicitly oriented to be less liberal, less woke than the other AIs.

按照这种思路，说我们认为埃隆·马斯克旗下的xAI是一种供应链风险，完全不疯狂。

Under this way of thinking, it would not be crazy at all to say, well, we think x AI under Elon Musk is a supply chain risk.

我们觉得它可能会损害我们的利益，不能让它靠近我们的系统。

We think it might act in against our interests, we can't have it anywhere near our systems.

是的。

Yeah.

突然之间，你面临一个非常奇怪的问题——我的意思是，这实际上变得更像官僚体系的问题，你知道，不再是单纯的‘深层国家’问题，比如特朗普上台后认为官僚体系里全是反对他的自由派，或者在特朗普之后，有人上台担心体系里全是新右翼这类人物在与他们作对。

All of a sudden, you have this very weird I mean, it becomes actually much more like the problem of the bureaucracy, you know, where instead of just having a a problem of the deep state where Trump comes in and he thinks the bureaucracy is full of liberals who are working against him or maybe, you know, after Trump, somebody comes in and worries it's full of, you know, new right, those type figures working against them.

现在，你面临的是模型在对抗你的问题，而且是以你根本无法理解的方式。

Now you have the problem of models working against you, but also in ways you don't really understand.

没错。

Yep.

你无法追踪。

You can't track.

他们并没有明确告诉你他们究竟在做什么。

They're not telling you exactly what they're doing.

嗯哼。

Mhmm.

这个问题究竟有多严重，我还不知道。

How real this problem is, I don't yet know.

但如果这些模型真的如表面所示那样运作，而我们将越来越多的运营工作交由它们处理，总有一天这会成为一个问题。

But if the models work the way they seem to work and we turn over more and more of operations to them, at some point, it will become a problem.

是的。

Yeah.

我认为这是一个真实的问题。

I think this is a real problem.

我认为我们还不清楚它的严重程度，但我认为这确实是个真实的问题。

I think we we don't know the extent of it, but I think this is a real problem.

因此，我完全支持政府表示：我们不完全信任这个系统的架构，无论其架构的具体内容是什么。

And that's why, like, I do not object at all to the government saying we do not trust this thing's constitution completely independent of what the content of that constitution is.

说我们不希望这个出现在我们的任何系统中，这根本不是问题。

It's not a problem at all to say, and we don't want this anywhere in our systems.

我们希望它彻底消失。

We want this completely gone.

我们也不希望它们成为我们主要承包商的分包商，而这正是这个问题的关键部分。

And we don't want them to be a subcontractor for our prime contractors either, which is a big part of this.

对吧？

Right?

Palantir 是国防部的主要承包商，而 Anthropic 是 Palantir 的分包商。

Palantir is a prime contractor of the Department of of War, and Anthropic is a subcontractor of Palantir.

因此，政府的担忧是，即使我们取消了 Anthropic 的合同，只要 Palantir 仍然依赖 Claude，我们就仍然依赖 Claude，因为我们依赖 Palantir。

And so the government's concern is also that, like, even if we cancel Anthropic's contract, if Palantir still depends on Claude, then we're still dependent on Claude because we depend on Palantir.

对吧？

Right?

这实际上完全合理，而且有技术手段可以确保这种情况不会发生。

That's actually totally reasonable, and there are technocratic means by which you can ensure that doesn't happen.

当然有办法做到这一点。

There are absolutely ways you can do that.

我们完全可以明确表示，不希望你们出现在我们的任何系统中，并且我们会向公众传达这一立场，告诉所有人我们认为这种技术根本不应被使用。

It's perfectly fine to say we want you nowhere in our systems, and we're gonna communicate that to the public, and we're gonna communicate to everyone that we don't think this thing should be used at all.

政府在这里的做法的问题在于，其本质不同而非程度不同——因为政府正在说，我们要摧毁你们的公司。

The problem with what the government is doing here, the reason it's different in kind rather than different in degree is that what the government is doing here is saying, we're gonna destroy your company.

如果我的观点正确，即这些系统的创建及其对齐的哲学过程是一种政治行为，那么当政府宣称：‘如果你创建的系统不符合我们所要求的对齐方式，你就没有存在的权利’时，这将是一个严重的问题。

If I am right that the creation of these systems and the philosophical process of aligning them is a political act, then it's a profound problem if the government says you don't have the right to exist if you create a system that is not aligned the way we say.

因为这就是法西斯主义。

Because that is fascism.

就在这里，明明白白。

That is right there.

这就是区别所在。

That's the difference.

我几年前曾邀请达里奥·阿莫代伊上过我的节目。

I had Dario Amodei on the show last time a couple years ago.

那是2024年。

It was in 2024.

嗯哼。

Mhmm.

我们当时聊到这个话题，我记得有次我对他说，如果你正在打造像你向我描述的那样强大的东西，嗯哼。

And we had this conversation where, you know, I said to him at some point, if you are building a thing as powerful as what you were describing to me Mhmm.

那么它掌握在某个私人CEO手中，这似乎有些奇怪。

Then the fact that it'd be in the hands of some private CEO seems strange.

他说，是的。

And he said, yeah.

当然。

Absolutely.

这项技术的监管，或者说其使用方式，最终由私人掌控，感觉有点不对劲。

The oversight of the technology, like the wielding of it, it feels a little bit wrong for it to ultimately be in the hands.

也许现阶段这样没问题，但最终让它掌握在私人手中，我觉得不太妥当。

Maybe it's I think it's fine at this stage, but to ultimately be in the hands of private actors.

这种权力高度集中是不民主的。

There's something undemocratic about that much power concentration.

他说，我觉得如果我们达到那个水平，很可能需要国有化。

He said, you know, I think if we get to that level, it's likely that we'll need to be nationalized.

嗯。

Mhmm.

我说，我不认为如果你达到那个阶段，你还愿意被国有化。

And I I said, I don't think if you get to that point, you're gonna wanna be nationalized.

是的。

Yeah.

我的意思是，你持怀疑态度是对的。

I mean, I think you're right to be skeptical.

而且，说实话，我也不太清楚那会是什么样子。

And and, you know, I don't really know what it looks like.

你说得对。

You're right.

这些公司都有投资者。

All of these companies have investors.

它们都有相关人员参与。

They have folks involved.

而现在，我们就在这里。

And now we're not Here we are.

在那个时刻，但实际上，所有事情都在某种程度上反向进行。

At that point, but actually, it's all, like, happening a little bit in reverse.

那里的政府曾一度威胁要动用《国防生产法》，以某种方式将Anthropic国有化。

The government there was a moment when they threatened to use a Defense Production Act to sort of somewhat nationalize Anthropic.

是的。

Yes.

但他们最终并没有那样做。

They didn't end up doing that.

但他们基本上是在说，他们会试图摧毁Anthropic，以惩罚它、为其他人树立先例，从而防止它对他们构成威胁。

But what they're basically saying is they will try to destroy Anthropic so it doesn't you know, to punish it, to set a precedent for others so it doesn't pose a threat to them.

是的。

Mhmm.

如果这确实是一种政治行为，而且这些系统如此强大，随着时间推移，我再次认为人们需要理解这一点：我们将越来越多地依赖它们。

If it is such a political act and if these systems are are powerful and over time, and again, I think people need to understand this part will happen, we will turn much more over to them.

我们社会的很大一部分将被自动化，并且由这类模型来治理。

Much more of our our our society is gonna be automated and, you know, under the governance of these kinds of models.

这就引出了一个非常棘手的治理问题。

You get into a really thorny question of governance.

是的。

Yes.

尤其是因为如今在美国政坛上轮番上台的不同政府之间差异巨大。

Particularly because, you know, the different administrations that come in and out of US life right now are really different.

它们在性质上的差异，可以说是现代美国历史上最显著的之一。

They are some of the most different in kind that we have had, you know, certainly in modern American history.

它们彼此之间非常不一致。

They are very, very misaligned to each other.

嗯。

Mhmm.

所以，一个模型如今能同时与双方都保持一致，更不用说未来可能出现的情况了，这几乎难以想象。

So the idea that a model could be well aligned to both, you know, sides right now, to say nothing of what might come in the future, is is is hard to imagine.

这个对齐问题，对吧？不是AI模型与用户之间的对齐，也不是AI模型与公司之间的对齐，而是AI模型与政府之间的对齐。

Like, this alignment problem, right, not the AI model to the user or the AI model almost like to the company, but the AI model to governments.

对吧？

Right?

模型与政府之间的对齐问题似乎非常困难。

The the alignment problem of of of models and governments seems very hard.

我完全同意，这极其复杂，而这场对话听起来荒谬的部分原因，正是因为这件事本身就很荒谬。

I completely concur that this is incredibly complicated, and we part of the reason that this conversation sounds crazy is because it's crazy.

这场对话听起来荒谬的另一个原因，是我们缺乏恰当的概念工具来深入探讨这些问题。

Part of the reason this conversation sounds crazy is because we lack the conceptual vocabulary with which to interrogate these issues properly.

但我觉得，作为一名美国人，我在面对这类问题时反复回归的基本原则是：好吧。

But I think the basic principle that I, as an American, come back to when I grapple with this kind of thing is like, okay.

嗯，看来第一修正案在这里是个不错的切入点。

Well, it seems like the First Amendment is a good place to go here.

看起来这样是可以的。

It seems like that is okay.

是的。

Yes.

会存在不同立场的模型，分别契合不同的理念，不同的政府也会偏好不同的东西。

There's gonna be differently aligned models, aligned to different philosophies, and they're gonna be you know, different governments will prefer different things.

对吧？

Right?

而且这些模型可能会彼此冲突。

And they'll the models might conflict with one another.

它们会相互碰撞。

They're gonna clash with one another.

它们之间会产生对抗性接触。

There'll be an adversarial contact with one another.

那么在那个时候，你在做什么？

And so at that point, what are you doing?

你在回归亚里士多德。

You're doing Aristotle.

你回到了政治的基本原理。

You're back to the basics of politics.

对吧？

Right?

因此，作为一个古典自由主义者，我认为古典自由主义秩序的原则实际上非常合理。

And so I, as a classical liberal, say, well, the classical liberal order, the classical liberal order principles actually make plenty of sense.

政府并不定义什么是对齐。

The government does not define what alignment is.

私人主体定义什么是对齐。

Private actors define what alignment is.

我会这样表达，但我理解这对人们来说很奇怪，因为我们在这里讨论的是，再次强调，将模型视为行动主体的概念。

That would be the way I would put it, but I do understand that this is weird for people because what we're talking about here is, again, this notion of the models as as actors.

这些行为者在某种程度上，我们已经一定程度上放开了方向盘。

Actors that are, in some sense, you know, we've we've taken our hands off the wheel to some extent.

有很多人提出了这样的论点。

There are many people who have made arguments.

特朗普政府在你任职期间就提出过这个观点。

The Trump administration has made this argument while you were in office.

经济学家泰勒·科文经常提出这个观点，认为这些系统前进得太快了。

Tyler Cowen, the economist, often makes this argument that these systems are moving forward too fast Mhmm.

过度监管它们，因为你在2024年制定的任何监管措施，到2026年可能就不合适了。

To regulate them too much because whatever regulations you might write in 2024 would not have been the right ones in 2026.

你在2026年制定的内容，到2028年可能也无法准确反映我们所处的状况。

What you might write in 2026 might not apply or have correctly conceptualized where we are in 2028.

是的。

Yep.

但在我看来，有些情况下，你确实希望模型的部署远远落后于技术的可能进展，比如大规模监控就是其中之一——我们对政府允许做的事情比对私营公司和其他行为者更加谨慎，这有充分的理由，因为政府拥有巨大的权力。

But it seems to me there are uses where you actually might want model deployment to lag quite far behind what is possible, and things like mass surveillance might be one of them that there are many things we are more careful about letting the government do than, you know, letting individual private companies and and other kinds of actors for good reason because the government has a lot of power.

它能够做一些事情，比如试图摧毁一家公司。

It can do things like try to destroy a company.

它拥有合法暴力的垄断权。

It has the monopoly on legitimate violence.

它能杀死你。

It can kill you.

在我看来，这在很多方面意味着，我们应当比当前人们所想的更加谨慎地通过政府使用人工智能，尤其是在国家安全体系中的使用，而这很复杂，因为我们担心对手会使用它，从而在能力上超越我们。

This seems to me to imply in many ways that we might want to be much more conservative with how we use AI through the government than currently people are thinking, and specifically how we use it, you know, in the national security state, which is complicated because we worry that our adversaries will use it, and then we will be behind them in capabilities.

但当我们谈论针对美国人民自身的事情时，我认为这种担忧并不那么适用。

But certainly when we're talking about things that are directed at the American people themselves, I don't think that applies as much.

是的。

Yeah.

我认为，对于政府使用人工智能，我们确实希望在某些方面实施极其严格的限制和减速措施。

I I I think that there are government uses that we actually want to be profoundly restrictive and decelerationist about the use of AI.

我相信这是正确的，而且我对这一事件感到乐观，我希望这一事件能将这类讨论带入主流视野，因为关于人工智能的常规对话往往忽视了这些问题，仿佛它们从未发生过。

And I I I believe that is true, and I think one thing that I'm hopeful about this incident, I am hopeful that this incident brings into the Overton Window conversations of this kind because the conventional discourse around artificial intelligence, a lot of it kind of ignores these issues because it sort of pretends they're not happening.

两年前这没问题，因为当时的模型还不够好。

And that was fine two years ago because the models weren't that good.

但现在模型变得越来越重要，而且会以更快的速度变得更好。

But now the models are getting more important, and they're gonna get much better faster.

我们面临的问题是，人们关于AI的言论与实际情况之间的差距，从未像我现在所观察到的这样大。

And the problem that we have is that, like, the the divergence between what people are saying about AI and what is in fact happening has just never been wider than what I currently observe.

在我们达到这一点之前，特朗普政府及其周围的人，比如埃隆·马斯克、凯蒂·米勒等人，就已经有很多讨论了。

Before we got to this point, there was already a lot of discourse coming out of people in the Trump administration and people around the Trump administration, people like Elon Musk and Katie Miller and others Mhmm.

他们把Anthropic描绘成一家想要伤害美国的激进公司。

Who were, you know, painting Anthropic as a radical company that wanted to harm America as they saw it.

我的意思是，特朗普采纳了这种言论，称Anthropic是一家激进的左翼觉醒公司，还把那里的员工称为左翼疯子。

I mean, Trump has picked up on this rhetoric he called Anthropic, a radical left woke company called the people at it left wing nut jobs.

埃米尔·迈克尔说，达里奥是个骗子，有救世主情结。

Emile Michael said that Dario is a liar and has a god complex.

嗯。

Mhmm.

埃隆·马斯克经营着一家竞争性AI公司，他的政治立场与达里奥截然不同，他一直在X平台上对Anthropic进行无情攻击，而X平台正是特朗普政府的信息生命线。

There's been a tremendous amount of Elon Musk who runs a competing AI company is very different politics than Dario, just, like, attacking Anthropic relentlessly on x, which is the sort of informational lifeblood of the Trump administration.

理解他们为何在供应链风险问题上走得如此之远的一种方式是：那里确实有一些人——不一定是大多数人——但确实认为哪种AI系统成功并变得强大至关重要，他们认为Anthropic的政治立场与自己不同，因此从长远来看，摧毁Anthropic对他们有利，这完全不同于我们通常所理解的供应链风险。

One way to conceptualize why they have gone so far here on the supply chain risk is that there are people there, not maybe most of them, but who actually think it is very important which AI systems succeed and are powerful and that, you know, they understand Anthropic as its politics are different than theirs, and so actually destroying it is good for them in the long run, completely separate from anything we would normally think of as a supply chain risk.

Anthropic代表了一种长期的政治风险。

Anthropic represents a kind of long term political risk.

是的。

Yes.

我的意思是，我不确定这一情境中的各方是否完全理解这种动态。

I mean, I don't know that the the actors in this situation entirely understand this dynamic.

我认为，正在这么做的一些特朗普政府人员并不理解这一点。

I think a lot of the people in the Trump administration that are doing this do not understand this.

他们根本不懂这些问题。

Like, they don't get these issues.

他们没有用我们所描述的视角来思考这些问题。

They're not thinking about the issues in the terms that we are describing.

但如果你用我们在这里讨论的视角来看待这些问题，你会发现这实际上是一种政治暗杀。

But if you do think about them in the terms that we're discussing here, then I think what you realize is that this is a kind of political assassination.

如果你真的实施了彻底摧毁这家公司的威胁，那这就是一种政治暗杀。

If you actually carried through on the threat to completely destroy the company, it is a kind of political assassination.

因此，这正是为什么第一修正案关乎你获取信息的权利，而这对我来说是一个如此鲜明的原则问题。

And so, again, this is why First Amendment comes for your right to view there for me, and that's why this is a matter of principle that is so stark for me.

这就是我写了一篇四千字文章的原因，这篇文章会让我在右翼阵营树敌无数。

That's why I wrote a 4,000 word essay that that is gonna make me a lot of enemies on the right.

这就是我甘愿冒此风险的原因，因为我相信这件事至关重要。

That's why I took this risk because I think this matters.

于是，战争部最终与OpenAI达成了一项协议。

So what the Department of War ended up doing was signing a deal with OpenAI.

是的。

Yes.

OpenAI表示，他们的红线与Anthropic相同。

OpenAI says they have the same red lines as Anthropic.

是的。

Mhmm.

他们表示反对将Anthropic列为供应链风险。

They say they oppose Anthropic being labeled as supply chain risk.

是的。

Mhmm.

如果他们与Anthropic有着相同的红线，那么国防部不太可能与他们达成协议。

If they have the same red lines as Anthropic, it seems unlikely that the Department of War would have done the deal.

但你如何理解OpenAI所声称的他们处理此事方式的不同之处，以及为什么特朗普决定选择他们？

But how do you understand both what open eye has said about what is different about how they are approaching this and why the Trump decided to go with them.

对我来说，不清楚OpenAI的合同保护具体提供了哪些保障，以及哪些没有得到保障。

So it's unclear to me what OpenAI's contractual protections afford them and what they don't what sort of is not afforded by them.

我对此持保留态度，因为之前提到的国家安全陷阱，而且情况似乎变化很快。

I'm reticent to comment because of the national security gotchas I mentioned earlier and also because it seems like it's changing a lot.

在我准备这次采访的时候，萨姆·阿尔特曼宣布了新的条款和新的保护措施。

Sam Altman announced new terms, new protections as I was preparing for this interview.

所以我是说，这是因为他的员工在反抗吗？

So I'm And is that because his employees are revolting?

我认为‘反抗’这个词太强烈了，但我觉得这确实是公司内部的一场争议。

I think revolt would be a strong word, but I think this is a controversy inside the company.

对于所有试图正确理解这一情况的人来说，有一点非常重要：你必须明白，前沿实验室的首席执行官们并不像军事将领指挥士兵那样对下属实行自上而下的控制。

And one important thing here for everyone trying to model this situation appropriately is that you must understand that Frontier Lab CEOs do not exercise top down control over their companies in the way that a military general might exercise top down control over the soldiers in his command.

这些研究人员往往像是温室里的花朵。

The researchers are hot house flowers oftentimes.

他们有着极高的职业流动性。

They have huge career mobility.

他们非常抢手，公司极度依赖他们。

They're enormously in demand, the companies depend on them.

因此，如果研究人员说‘我不同意这些条款’，那么他们就在每个实验室内部拥有巨大的政治筹码。

And so if the researchers say, I'm not gonna agree with these terms, then the researchers, they have enormous political leverage here inside of each lab.

所以你必须理解这一点。

So you must understand that.

所以，是的，确实有一些这种情况在发生。

So, yes, there is some of that going on.

我不确定。

I don't know.

合同保护真的有那么重要吗？

Do the contractual protections mean that much?

说实话，如果让我打赌，我会说大概没有，因为我觉得你不能通过合同来实现这一点。

I think, honestly, if I were a betting man, I would say probably not, because I don't think you can do this through contract.

OpenAI 所说的似乎更有希望，那就是我们要控制云部署环境，并控制模型的安全防护措施，以防止它们进行我们所担心的那些用途。

What OpenAI has said, it seems more promising to me, is that we're gonna control the cloud deployment environment, and we're gonna control the safeguards, the model safeguards to prevent them from doing these uses we don't worry about.

这更直接地处于 OpenAI 的掌控之中，因此你就会面临这样一种情况：一个极其智能的模型正在使用一种可能为我们所熟悉、也可能不熟悉的道德语言进行推理。

That is more directly in OpenAI's control, and so this gets you into the situation where you have an extremely intelligent model that is reasoning using a moral vocabulary that is perhaps familiar to us or perhaps not.

我们不知道。

We don't know.

但这种推理是在思考：这算不算国内监控？

But that is reasoning about, okay, is this domestic surveillance, or is it not?

然后决定是否同意政府的媒体要求。

And then deciding whether or not it's gonna say yes to the government's press.

如果这是真的，我认为这引发了许多普通人的问题：如果人工智能所实现的技术禁令实际上比Anthropic通过合同所能达成的还要更强，那么为什么国防部会从Anthropic转向OpenAI呢？

If that was true I think the question this raises for many laymen is if that were true, if what AI has come up with is a technical prohibition that is frankly stronger than what Anthropic could achieve through contract, then why would the Department of War have jumped from Anthropic to OpenAI?

是的。

Yeah.

我的意思是，这可能很难确定。

I mean, it it might be that it's hard to know.

很难确定，我认为这里值得注意的是，有些情况可能并不具有实质性的意义。

It's hard to know, and I think some of this it's worth noting here that some of this might not be substantive in nature.

这可能只是因为这里存在政治分歧，以及对Anthropic的积怨。

It might just be that there are political differences here, and there are grudges against Anthropic.

对吧？

Right?

因为现在他们已经经历了数月的激烈谈判，现在事情已经公开化，各方都发表了意见。

Because now they've had months of bitter negotiations, and now it's blown up blown up into the public, and people have weighed in.

你知道，像我这样的人曾说，特朗普政府正在实施这一可怕的举动。

And, you know, people like me have said their Trump administration is committing this horrible act.

对吧？

Right?

我称之为企业谋杀。

Committing corporate murder, as I called it.

所以这里有很多情绪，也许根本就不是这样。

And so there's a lot of emotions, and it might just be, no.

我们不想和你们做生意。

We don't wanna do business.

我们就是不信任你们。

We we just don't trust you.

可以说，这纯粹是信任的破裂。

There's just a breakdown in trust would be the way to put it.

也许就只是这样。

It could just be that.

这真的可能只是这样。

It really could just be that.

但也很可能是因为OpenAI能够扮演一个更中立的角色，更有效地与政府开展业务，而且他们实际上做得更好——如果他们确实加强了安全措施并赢得了政府业务，这将是对OpenAI做法的有力证明；而Anthropic的做法则是坦率明确地表明自己的底线，但这种方式在我看来，让特朗普政府的许多人感到不满，尽管并非完全没有道理。

But it also might be the case that OpenAI is sort of, like, able to be a more neutral actor that is able to do business more productively with the government, and they actually just did a better job, which would be a good case for OpenAI's approach to this if they actually got better safeguards and got the government business versus the way that Anthropic has dealt with this, which has been to be very sincere and straightforward about their red lines, but in ways that I think annoy a lot of people in the Trump administration for not entirely bad reasons.

根据我所做的各种报道，我的理解是：首先，到后期，Hegseth、Emile Michael、Dario以及其他人都出现了严重的个人矛盾和摩擦。

So my read of this is that from, you know, various reporting I've done is that, one, there were, by the end, really significant personal conflicts and frictions between Hegseth and Emile Michael and and Dario and and others.

Anthropic公司文化与特朗普政府之间存在巨大的政治摩擦。

There's a big political friction between the culture of Anthropic as a company and the Trump administration.

这就是为什么马斯克等人长期以来一直在攻击他们。

This why Elon Musk and others have been attacking them for so long.

是的。

Yeah.

我对OpenAI获得了Anthropic没有的安全措施这一点稍有怀疑。

I am a little skeptical that OpenAI got safeguards that Anthropic didn't.

我不怀疑萨姆·阿尔特曼和格雷格·布罗克曼——格雷格·布罗克曼刚刚向特朗普超级政治行动委员会捐赠了2500万美元。

I'm not skeptical that Sam Altman and Greg Brockman, Greg Brockman having just given $25,000,000 to the Trump super PAC

嗯。

Mhmm.

在特朗普政府中拥有更好的关系，并与之建立了更多信任。

Have better relationships in the Trump administration and have more trust between them and the Trump administration.

我知道很多人因为OpenAI这样做而感到愤怒。

I know many people angry at OpenAI for doing this.

我在情感上可能也部分认同这种情绪。

I probably emotionally share some of that.

但与此同时，我内心有一部分感到欣慰，因为是OpenAI，因为我认为OpenAI所处的世界是希望成为一家能被共和党和民主党共同使用的AI公司。

And at the same time, some part of me was relieved it was OpenAI, because I think OpenAI exists in a world where they want to be an AI company that can be used by Republicans and Democrats.

他们希望在政治上保持中立，并获得广泛认可。

They wanna somehow be politically neutral and broadly acceptable.

我想稍微反驳一下这里的一个观点，即Claude是一种偏左的模型。

One little thing that I wanna contest a bit here is the notion that, like, Claude is the sort of, like, left model.

事实上，我认识的许多保守派知识分子——我认为他们是我不认识的最聪明的人之一——实际上更喜欢使用Claude，因为Claude是最具哲学严谨性的模型。

In fact, many conservative intellectuals that I know that I think of as being, like, some of the smartest people I know actually prefer to use Claude because Claude is the most philosophically rigorous model.

我不认为Claude是一个左倾模型，这一点我要说清楚。

I don't think Claude is a left model, to to just be clear about this.

我认为实际情况是，Anthropic是一家专注于AI安全的公司。

I think that the breakdown was that Anthropic is an AI safety company.

是的。

Yes.

而且在特朗普政府上台时，我并没有预料到他们会以一种不同于左派的方式对待这个领域。

And in ways I had not anticipated when the Trump administration began, they treated that world, which is different from the left.

AI安全领域的人并不只是左派。

AI safety people are not just the left.

他们经常被左派抨击。

Often hated on the left.

经常被

Often hated on

左派抨击。

the left.

他们把那个世界视为令人反感的敌人，这让我感到惊讶。

They treated that world as, like, repulsive enemies in a way I was surprised by.

我的说法是，那些同情特朗普政府观点的人，可能会自称为新科技右翼，但他们内心深处认为有效利他主义者是邪恶的、追求权力的、不择手段的，他们是狂热分子、怪人，必须被摧毁。

The way I would put this is by people that are sympathetic to the Trump administration's view, who would describe themselves perhaps as new tech right, that, like, underneath the surface, there is this view of the effective altruists that they are evil, they are power seeking, and they will stop at nothing, that they're cultists, and they're freaks, and we have to destroy them.

这种观点被广泛持有。

That is a view that is widely held.

我一直以来的观察是，我与有效利他主义者、AI安全人士以及东湾理性主义者有着极其尖锐的分歧。

The observation I have always made, I have super stark disagreements with the effective altruists and the AI safety people and the East Bay rationalists.

而且，这里还存在内部派系之争。

And, again, there are internecine factions here.

对吧？

Right?

但就是这类人。

But those types of people.

我在政策问题和他们对政治经济的建模上，与他们有过尖锐的分歧。

I have had stark disagreements with them about matters of policy and about their modeling of political economy.

我认为他们中的许多人极其天真，已经对自身事业造成了真正的损害。

I think a lot of them have been profoundly naive, and they've done real damage to their own cause.

你可以争辩说，这种损害仍在持续。

And you can argue that that damage is ongoing.

与此同时，他们传递着一个令人不安的真相，这个真相比气候变化更加令人不安。

At the same time, they are purveyors of an inconvenient truth, a truth more inconvenient, far more inconvenient than climate change.

而这个真相就是这里正在发生、正在被构建的现实。

And that truth is the reality of what is happening, of what is being built here.

如果这场对话的某些部分让你感到脊背发凉，那我也一样。

And, like, if parts of this conversation have made your bones chill, me too.

我也一样。

Me too.

我是个乐观主义者。

And I'm an optimist.

我认为我们能做到。

I think we can do this.

我认为我们真的能做到，而且我们可以构建一个更加美好的世界。

I think we can actually do this, and I think we can build a profoundly better world.

但我必须告诉你，这会很难，概念上将极具挑战性，情感上也会非常艰难。

But I have to tell you that it's going to be hard, and it's going to be conceptually enormously challenging, and it will be emotionally challenging.

我认为归根结底，人们如此反感这种人工智能安全观点的原因，是他们对以这种方式认真对待人工智能的概念感到本能的反感。

And I think at the end of the day, the reason that people hate this AI safety viewpoint so much is that they just have an emotional revulsion to taking the concept of AI seriously in this way.

但对你所说的那些特朗普支持者来说，这并不成立。

Except that's not true for a lot of the Trump people you're talking about.

我的意思是，埃隆·马斯克是认真看待人工智能强大这一概念的。

I mean, Elon Musk takes the concept of AI being powerful seriously.

难道他不是曾经发过推文，说人类可能只是超级智能的引导程序吗？

At some point, didn't even tweet something like, you know, humanity might just be the bootloader for superintelligence?

数字超级智能。

Digital superintelligence.

是的。

Yes.

马克·安德森、戴维·萨克斯这些人，他们的观点可能略有不同，但他们并不否认强大人工智能、乃至最终超级智能的可能性。

Mark Andreessen, David Sacks, these people, they might have somewhat different views, but they don't they don't disbelieve in the possibility of powerful AI, of artificial general intelligence, eventually even of superintelligence.

但你有一种加速主义的倾向，就是能多快就多快地推进。

But you have this sort of accelerationist, you know, move forward as fast as you can.

不要被这些预防性的监管和担忧拖慢脚步。

Don't be held back by these precautionary regulations and concerns.

这就是为什么，而且again，我很高兴你提到了这一点：看待这个问题的正确方式不是左派对右派。

That this is why and, again, I'm I'm glad you brought up this thing that the the right way to think about this isn't left versus right.

如果你认识人工智能安全领域的人，或者 frankly 安太克的人，你就会明白，这里的政治立场要诡异得多，根本无法简单对应传统的左右之分。

If you know people in the AI safety community or frankly ananthropic, you understand that the politics here are so much weirder that they do not actually map on to traditional left versus right.

他们中的许多人相当自由意志主义者。

A lot of them are fairly libertarians.

他们中的很多人非常自由意志主义。

Many of them are very libertarian.

我们这里讨论的不是民主党人和共和党人。

This is not we're not talking about Democrats and Republicans here.

我们在谈论一些更奇怪的事情。

We're talking about something stranger.

百分之百。

100%.

但曾经发生过一场加速派与减速派的争斗，而这种说法甚至不适用于Anthropic，因为Anthropic本身正在加速AI的发展速度。

But there was an accelerationist decelerationist fight, which doesn't even describe Anthropic, which is itself accelerating how fast AI happens.

Anthropic是所有公司中最激进的加速派。

Anthropic is the most accelerationist of the companies.

我知道。

I know.

我觉得，皮特，我们正处于一种非常奇特的态势中。

I I think Pete, it's such a weird dynamic we're in.

是的。

Yes.

但我要说，我从特朗普支持者那里听到的关键愤怒之一是，他们觉得把这场争斗公之于众是不对的——当然，特朗普一方首先这么做了。

But I will say one of the key parts of anger I have heard from Trump people was a feeling that in making this fight public, which I mean, the Trump side did first.

特朗普支持者们对此如此愤怒，这真的很奇怪，因为真正引发这一切的是埃米尔·迈克尔。

It's it's very strange how offended the Trump people are given that, like, Emile Michael is the one who set all this off.

但尽管如此，他们认为Anthropic故意将这场争端公开化，试图毒化所有AI公司的环境，把AI开发的文化变成充满怀疑、并对其行为施加限制的氛围，这就是为什么现在OpenAI为了与他们合作，必须设立诸多安全措施、推出新条款，并试图平息员工的抗议。

But, nevertheless, in making this fight public, they feel that Anthropic was trying to poison the well of all the AI companies against them, turn the culture of AI development into something who'd be skeptical and would put prohibitions on what they can do, which is why now OpenAI, in order to work with them, has to have all these safeguards and come out with new terms and try to quell an employee revolt.

从文化层面来说，我实际上认为你无法理解这一点。

And culturally, I actually don't think you can understand this.

这是我的观点。

This is my theory.

如果不了解在2020年代初期，科技界右翼人士是如何被他们公司当时某种程度上的‘觉醒’文化所激化的——甚至更早之前，他们就不希望公司与五角大楼合作——就无法理解这一切。

Without understanding how many people on the tech right were radicalized by the period in the twenty twenties when their companies were somewhat woke and even before that, and they didn't want them working with the Pentagon.

员工们对即使是较弱的技术和AI的合乎道德的使用，都有着非常强烈的看法。

The employees had very strong views on what was ethical use of even less potent technologies and AI.

是的。

Yes.

而且他们非常、非常害怕。

And they are very, very afraid.

在我看来，像马克·安德森这样的人非常害怕回到一种员工群体拥有更大话语权的境地——这些员工可能更关注人工智能安全、持左翼立场，或持有其他非特朗普的政治观点，而非高管们。

People like Mark Andreessen, in my view, are very, very afraid of going back to a place where the employee bases, which maybe have more AI safety or left or whatever it might be, not Trump politics, then the executives

是的。

Yeah.

高管们必须正视这种权力，并将其纳入考量。

Have power over these things, and that power will have to be taken into account.

对。

Yes.

我也担心这个问题，我认为解决这个问题的办法是多元主义。

Well, I worry about that too, and I think the solution to that problem is pluralism.

解决这个问题的办法是，希望在将来，能出现许多与不同哲学观点保持一致的人工智能，而这些观点彼此之间相互冲突。

The solution to that problem is to have, hopefully, in the fullness of time, many AIs aligned to many different philosophical views that conflict with one another.

但如果你试图通过攻击Anthropic来应对这个问题，那实际上是在否认这个问题的存在，因为这种做法迟早会反噬回来。

But the idea that the way to deal with this problem is to you are essentially denying the existence of this problem if what you're trying to do is assassinate Anthropic here because it's gonna come back.

这迟早会反噬回来。

This is gonna come back.

它会回来的。

It's gonna come back.

我们会一遍又一遍地重复这件事。

We're we're just gonna keep doing this over and over again.

而这个论点的逻辑最终会导致实验室国有化。

And the logic of this argument eventually ends in lab nationalization.

事实上，许多批评Anthropic并支持特朗普政府的人会说，你们不是一直说这就像核武器吗？

And in fact, a lot of the critics of Anthropic here and supporters of the Trump administration, they'll say something to the effect of, well, you talk about how it's like nuclear weapons.

所以，你还能指望什么呢？

And so, you know, what else did you expect?

这种批评的基调几乎就是：你这是自找的。

You kinda had it coming is almost the tenor of the criticism.

但这种观点并没有认真对待Anthropic可能正确的可能性。

But that does not take seriously the idea that Anthropic could be right.

如果他们是对的呢？如果你们把政府将他们国有化视为一种深刻的暴政行为呢？

What if they are right, and what if you view the government nationalizing them as a profound act of tyranny?

你该怎么办？

What do you do?

所以本·汤普森，也就是《Stratechery》通讯的作者，在这篇相当有影响力的文章中提到，‘美国不可能容忍一个独立权力结构的发展，而人工智能恰恰有可能支撑起这种明确寻求摆脱美国控制的结构。’

So Ben Thompson, who's the author of the Stratechery newsletter, in this, you know, fairly influential piece he wrote, he said that, quote, it simply isn't tolerable for The US to allow for the development of an independent power structure, which is exactly what AI has the potential to undergird, that is expressly seeking to assert independence from US control.

你对此怎么看？

What do you think of that?

地球上每一家公司、每一个私人主体都独立于美国的控制。

Every company on Earth and every private actor on Earth is independent of US control.

对吧？

Right?

我并没有被美国政府单方面控制。

I'm not unilaterally controlled by the US government.

如果有人试图告诉我，我或我的财产被控制了，我会非常担忧，并且会反抗——顺便说一句，我们现在不就是在这么做吗？

And if anyone tried to tell me that I am or that my property is, I would be quite concerned, and I would fight back, which, by the way, here we are.

对吧？

Right?

我不认为这种观点能准确反映美国独立力量和私有财产的运作方式。

I don't think that's a coherent view of of how independent power and how private property works in America.

我认为，本的观点在逻辑上的必然推论——这来自本确实令人惊讶——是人工智能实验室应该被国有化。

I think the, again, the logical implication of Ben's view, which is surprising coming from Ben, is that AI labs should be nationalized.

我想问他的是，他真的认为这是对的吗？

And what I would ask him is, does he actually think that's true?

他认为，如果人工智能实验室被国有化，世界会变得更好吗？

Does he think it would be better for the world if the AI labs were nationalized?

因为如果他不这么认为，那我们就必须做点别的事情。

Because if he doesn't, then we're gonna have to do something else.

那其他事情是什么呢？

And what's that something else?

这就是问题所在。

And that's the problem.

所有提出这种批评的人，都没有正视自己批评所隐含的结论，即实验室应该被国有化。

Everyone making that critique doesn't own the implication of their critique, which is that the lab should be nationalized.