伊隆是我们这一代最出色的金融家

顺便说一下，这个总潜在市场规模大到我们可以争论，就像我经常要强调的，中位数结果和平均值结果之间差别很大。

And, like, by the way, the TAM is so big that we can argue that the TAM is you know, again, the average I always have to use make this point for other things, which is, like, you know, there's a big difference between median outcome and mean outcome.

对吧？

Right?

SpaceX的中位数结果可能是负的。

And the median outcome for SpaceX might be negative.

对吧？

Right?

你觉得这个市场规模是无限的。

You believe the TAM is infinity.

对吧？

Right?

如果你觉得它能垄断所有领域，那平均值结果其实会非常乐观。

And you believe it can own all the things, then the mean outcome is actually quite positive.

所以如果你把它看作一场赌博——毕竟这些东西归根结底都是赌博——从结果走向来看，你真正想要的其实是很高的平均回报，哪怕概率很低。

And so if you just think about it as like a gambling, which all this stuff is just gambling at some point, I mean, like, in terms of how things go, you know, what you really just want is a really high mean outcome, even if it's low probability.

因为如果这个数字是无穷大，那零和无穷大的平均值还是无穷大，哪怕中位数结果是零。

Because if number is infinity, like, again, the mean of zero and infinity is infinity, even if the median outcome is zero.

对吧？

Right?

我觉得这里的情况就是这样，大家都明白这个道理。

And, like, I think that's kinda what you see here, and everyone gets that.

但每个散户都害怕错过这种重大转折。

And everyone's terrified on a retail basis of missing the swerve.

对吧？

Right?

没错。

Yeah.

正因如此，才有了这些估值。

Which is how you get these values.

上周末我问了你这个问题，当时我就在想，这到底意味着什么？

So I put this question to you over the weekend, and I was like, what does it mean?

那第二层影响是什么？

What are the second order effects?

这对其他公司意味着什么？

What does it mean for other companies?

你说过，别觉得这有什么特别的意义。

And you were like, no, don't think it means anything.

你的意思是，埃隆的世界就是埃隆的世界。

Like, your point was that the Elon Universe is the Elon Universe.

它只遵循埃隆自己的规则，没什么普适的教训，也不会对其他公司产生广泛影响。

It operates based on the laws of Elon, and there aren't, like, broader lessons for other companies or broader impacts.

你现在还这么认为吗？

Do you still feel that way?

我觉得影响可能更广，但你怎么看？

Because I feel like there's a broader impacts, but what's your point of view?

我觉得这可能会改变人们谈论的方式，以及他们创办公司的类型。

I think there might be a broader impact on the ways people talk about and the types of companies they start.

对吧？

Right?

没错。

Uh-huh.

但我觉得这种影响挺微妙的。

But I think it's like subtle.

我觉得在我们这一代，比如我们上大学的时候比马克·扎克伯格高一届。

It's like I think in our generation, like, so we were a year ahead of Mark Zuckerberg in college.

对吧？

Right?

这事儿真有点不可思议。

And it's kind of wild.

在我们那个年代，哈佛毕业生里创办公司的人数，其中不少还搞出了几十亿美元规模的企业，简直多得离谱。

Like in our era, the number of Harvard grads that went on to found companies, some of several multi billion dollar companies is like astronomical.

对吧？

Right?

这背后其实有好多原因。

And like, there's a bunch of reasons for that.

有些是时机和科技因素，不过2008年那会儿确实也没拖后腿。

Some of them were timing and technological, but it it you know, 2008 didn't hurt.

但我真觉得，这中间有个‘扎克伯格效应’——看到他能做到，大家就觉得这事也能成。

But I honestly do believe there was, like, a mark effect, right, where he was like, oh, that's possible.

对吧？

Right?

那我干嘛还在这儿给卡茨和巴克合伙公司卖命呢？

So, like, why am I toiling away at Katz and Back Partners?

对吧？

Right?

毕竟人家都做到了，我还在教室里见过他呢。

Because clearly that guy did it, I saw him in class.

对吧？

Right?

这么说你能明白吧。

If that makes sense.

我觉得以后会有更多人更有野心，做的事也更大胆，你懂的。

So I think there's gonna be a bunch more people being more ambitious, right, and crazier in what they're doing.

其实呢，这大概是件好事。

Now, that's actually, like, probably great.

你明白我的意思吧？

You know what I mean?

我其实希望，嗯，能有更多这样的人。

Like, I kinda want, you know, the I kinda want more

是的。

Yeah.

我喜欢这样。

I like that.

但我觉得这是积极的一面。

But I think that's the positive side.

我觉得这方面会带来一些影响，比如更雄心勃勃、更烧钱的项目，不过说实话，真正的风投根本不关心这些。

I think there will be an effect there in terms of more ambitious things, more capital consumptive things, which, by the way, venture capitalists, these real ones don't care about.

他们反而就喜欢这样。

They want that.

也就是说，这类项目能有更多的操作空间。

It was like, that let's they can be dealt in more.

对吧？

Right?

所以，这方面肯定会涌现出一堆类似的变化。

Like, so there's, like, gonna be, like, a bunch of effects that way.

也会有一大堆人走完全相反的路。

There'll be a bunch of people who go in completely opposite direction.

对吧？

Right?

或者就是盯着现金流，现金流，还是现金流。

Or just like cash flow, cash flow, cash flow.

这些东西都太疯狂了。

This stuff is all crazy.

我不想卷入这些荒唐的圈子里。

I don't wanna be in, like, these crazy lands.

但我觉得中间地带会消失。

But I think the middle drops out.

对吧？

Right?

这周我看到一件特别好笑的事，Intricom 这家公司莫名其妙改名叫 Fin AI，结果被贝索夫以三十亿美元收购了。

Like, I saw this week, which I thought was quite funny that Intricom, which renamed itself hilariously Fin AI, just got acquired by Benioff for, like, $3,000,000,000.

我再说一遍，

I again,

各位听众，萨姆创办了一家叫 Fin 的公司。

for the listeners, Sam started an AI company called Fin.

不过那家公司根本不是搞 AI 的。

It wasn't AI, though.

其实我们当时没叫它AI，我们真就叫它‘人工人工智’。

Well, it wasn't we didn't we called it a a it was we literally called it a AI, artificial artificial intel.

但它其实是个聊天界面，背后有好多厉害的技术和真人参与。

But it was a chat interface with a lot of really cool tech and humans.

顺便说一句，现在Facebook被骂的那些事，比如监控成千上万人来搭建工作流，我们早就在做了。

By the way, we did a lot of the stuff Facebook's getting yelled at, which is monitoring tens of thousands of people to build workflows.

这些事儿我们早就在一代人之前就干过了。

Like, we we did all this stuff a generation ago.

不过话说回来，他们后来又把一部分拿回去了。

They clawed some of that back, by the way.

所以我们早就在做这些事了，现在倒好，同名的东西被收购了，真是挺讽刺的，不过也无所谓了。

So, like, we did this stuff a generation ago, which I find very funny because it's now something of the same name getting acquired, but whatever.

我们的域名还更好，我们直接用了fin.com，而不是finai。

We had a better domain name because we actually had fin.com instead of finai.

那是什么？

What's that?

他们说，我看到这笔30亿美元的收购案时，第一反应就是：这也太无聊了吧。

What they say is, like, my reaction watching that acquire get acquired for $3,000,000,000 was, like, how unbelievably boring.

你懂我的意思吧？

You know what I mean?

我当时就是这么想的，看着这个收购案。

Like, that was my reaction, like, looking at it.

我就想，行吧。

I'm like, okay.

那时候我们还在为SpaceX的IPO发愁，大家玩的都是刺激的大游戏，结果现在冒出这种稳赚不赔的项目。

Like, in the era where the we were the weak this week SpaceX IPO and, like, these people are playing these wild games, and then there's, like, wildly profitable stuff.

你怎么跟工程师说，让他们去搞Fin AI、Intercom或者Dani这种东西？

I mean, how do you convince an engineer to work on Fin AI, Intercom, Dani?

简直太没劲了。

It's, like, just awful.

对吧？

Right?

所以对我来说，最有趣的就是，中间那段直接就没了。

And so, like, for me, that's the interesting thing is, like, the middle just drops out.

我觉得这种岗位根本招不到人。

I think it becomes incredibly hard to recruit for.

而且说实话，我真的不知道该怎么让人对这种事产生热情。

E and, like, it's also just, I don't know how you inspire anyone to do anything like that.

你懂我的意思吧？

You know what I mean?

SpaceX当年的厉害之处就在于此。

Like, I mean, that was the the brilliance of of SpaceX.

SpaceX早期有很多了不起的地方，但我觉得其中一个关键是——我也知道，这又回到老问题了：这是有意为之，还是碰巧发生的？

There's a lot of brilliances of SpaceX early on, but one of them, I think and I know, again, this goes back to the whole, was this intentional or did this happen?

有很多人热爱太空和火箭，就是因为它们太酷了。

Which is there's a lot of people who love space and rockets because they're awesome.

因为它们真的很酷。

Because they're cool.

因为它们太酷了。

Because they're cool.

当埃隆创办SpaceX的时候，你还有什么选择？

When Elon started SpaceX, what are your options?

你的选择要么是去NASA工作，那里声望很高，但肯定挺憋屈的。

Your options are to work for NASA, which is super prestigious but must kind of suck.

对吧？

Right?

不然呢？

Or what?

什么都没有。

Nothing.

所以，哪怕你不喜欢埃隆，哪怕你不太想为他做事，只要你对火箭感兴趣，你还是会去。

So, like, even if you don't like Elon, even if you, like, don't really like you're willing to work.

如果你真喜欢火箭，你就会想：算了，我还是去给埃隆干活吧，每天干一千个小时也认了。

If you're into rockets, you're like, I guess I'm gonna go work for Elon and work a thousand hours a night.

我知道，我会没日没夜地干下去。

Know, I'm gonna work infinitely.

我的生活肯定会很苦。

My life's gonna suck.

管他呢。

Whatever.

他能逼着人拼命，是因为如果你真对火箭感兴趣，他是唯一的选择。

I can he can push people so much harder because he's the only game in town if you really care about something.

对吧？

Right?

从招人的角度看，这优势太明显了。

That's, like, a very good place to be from a recruiting perspective.

所以我一直在想，很多公司都得找到自己的核心用户群体。

So I just wonder, like, with a lot of these companies, like, you have to find your tribe of customers.

同样，你也得找到属于自己的核心团队成员。

You also have to find your, like, tribe of employees and team members.

但我不知道2026年你怎么让那些无聊的事儿吸引人，对吧。

And I don't know how you do that for boring shit, right, in 2026.

你懂的。

You know?

他还有个隧道公司。

He's also got The Boring Company.

不过，也许隧道还真不算无聊。

But, yes, maybe tunnels aren't boring.

不过，我觉得你说得特别准。

But, no, I think you nailed it.

我记得以前去过SpaceX。

I remember visiting SpaceX.

大概十年前吧，再久就没过了。

Probably way more than ten year ten years ago, no more.

他们在霍桑的工厂，真的挺酷的。

Their Hawthorne facility factory, and, like, it's just pretty cool.

太酷了。

It's so cool.

我记得那次采访中，埃隆一口气喝了三种不同的蛋白粉饮品。

What I remember from that interview is Elon had three different protein shakes over the course of it.

三种不同的，他全程喝完了三杯蛋白粉。所以，要是十年前我们还担心他的蛋白质摄入量，那真是多虑了。

Three different, he went through three whole protein shakes that were And so if we were worried ten years ago about his protein intake, we ought not to have been.

那真是个有意思的时刻，很多当时的评论都挺热血的。

It's a fascinating moment, like the hot takes, many of them are hot takes.

在这些公司上市之前，它们简直就是标杆，开创了先例。

In the lead up to these IPOs, they're like the poster children, they're precedent setting.

它们会树立起新的标准，给Anthropic和OpenAI指明方向——这两家正等着入场，私下里讨论的时间表可比外界想的要早得多。

They're gonna set the benchmark, you know, they're gonna set the roadmap, the playbook for Anthropic and OpenAI, which are waiting in the wings and privately whispering timeframes that are quite sooner rather than later.

现在我们都开始琢磨了，但我一直在催信息团队，要他们说清楚这到底意味着什么。

And now we're all, but we're also like, like, I'm really pushing the information teams to say what it means.

可我得到的回应大多是：‘这可是埃隆啊。’

And I'm getting a lot of like, it's Elon.

我觉得没那么简单，不过咱们走着瞧吧。

I'm like, no, I think there's more, but we'll see.

我的意思是，咱们就等着看吧。

I mean, look, we'll, we'll see.

说实话，我有点难想象。

I mean, I have a hard time.

我觉得看看Anthropic和OpenAI接下来会怎样，一定会很有趣。

I think it'd be fascinating to see what happens with Anthropic and OpenAI.

对啊。

Mhmm.

你说是不是？

Right?

最简单的说法就是数字会往上走。

The easy answer is number go up.

没错。

Sure.

我们现在都能以数万亿美元的规模来做这件事了，当然可以。

Like, we can now do this at a multi trillion like, sure.

你懂我的意思吧？

You know what I mean?

为什么不呢？

Like, why not?

是的。

Yeah.

对吧？

Right?

顺便说一句，现在还真有种奇怪的财富效应——你要是靠某一项赚了大钱，一方面，流动性总得有个来源；

And by the way, there is some sort of weird wealth effect, right, where, like, if you get rich on one, people, like, on one hand, it's like the liquidity has to come from somewhere.

但另一方面，当人们的资产数字变大了，他们也更愿意投资。

On the other hand, when people's numbers are bigger, they invest more.

对吧？

Right?

所以这在两个方向上都形成了一种有趣的循环。

So it was like an interesting recursion in both directions.

某种程度上，OpenAI 和 Anthropic 的问题在于，它们都有实实在在的业务。

In some ways, problem with like OpenAI, but in Anthropic, is they have real businesses.

所以人们会认为它们是拥有真实业务的公司。

So people will value them as having real businesses.

你可以根据现金流来估值，判断它们值多少钱。

So you can actually apply some multiple to cash flows and decide what's worth.

对。

Yeah.

没错。

Yeah.

而像早期 OpenAI 那种说法，现在差不多成了 Anthropic 的叙事——谁要是能在 2026 年领先，就能拥有一个无穷无尽的增长飞轮。

You, like, whereas the, you know, and that's, I think, a lot of the fear mongering of early OpenAI that's kinda become Anthropic and the, oh, whoever's ahead in 2026 will have this infinite flywheel over there.

你得给这个故事加上背景，才能解释为什么它不只是一个能印钞的生意，对吧？

That is the storytelling you have to add, right, to infinite TAM and to why isn't this just a business I evaluate as a cash machine, right, effectively?

作为一台印钞机，这可真够难的。

And, like, you're like, oh, as a cash machine, that's a tough one.

你知道的。

You know?

这才是真正关键的问题。

That's, like, becomes the real question.

顺便说一下，昨晚我们报告了，OpenAI 在第一季度烧掉了 40 亿美元，而收入只有 57 亿。

We reported, by the way, last night that OpenAI burned 4,000,000,000 in the first quarter on revenue of 5.7.

这将是一场考验。

So it will be a test.

这确实是一场考验。

It'll be a test.

但说来有趣，就连这些数字，人们也都会当成锚点。

But it's like it's funny because, like, even that, like, people anchor on numbers.

当你看到一家公司估值两万亿美元，年收入才两百亿，你根本不会在意它到底烧了四十亿还是多少。

And when you're like, oh, well, something's worth $2,000,000,000,000 that does 20,000,000,000 in revenue, you're like, I don't care if you spend $4,000,000,000 or whatever.

你懂我的意思吧？

You know what I mean?

我觉得这事儿最后就变成一种特别原始的状态了。

Like, I think that becomes like, there's like a very, like troglodyte.

也许‘原始人’这个词不太准，但总之就是一种极端到离谱的版本。

Maybe troglodyte's not the right word, but there's like a very, like, I don't know, like, retard max version of all of this.

我们当时就想着，管他呢，想花就花吧。

We were just like, whatever, like, enjoy, spend a bunch of money.

以后钱还会有的。

Like, there'll be more.

我私下问过，如果SpaceX上市，对其他初创公司来说，是让IPO更难了还是更容易了？

I did ask it in our back channel, like, if the SpaceX IPO makes it harder, easier for other startups to go public.

我本来也不确定会听到什么，结果好几个人都说，短期内会更难，因为可比公司表现都不怎么样。

And the, I, I don't know what I was expecting, but several people chimed in harder in the short term because the comps are kind of like not great.

我的公司股权今天刚上市，或者可能明天。

I mean, my company team share is just listed today or maybe tomorrow.

进展怎么样？

How'd it go?

是已经上市了，还是还没上市？

They have listed or they haven't listed?

这是个SPAC交易，我其实不太清楚具体怎么运作的。

You know, it's a SPAC transaction, so I don't really understand how it works.

但好像股票代码还没更新，我觉得合并应该已经完成了。

But like, it's like the ticker hasn't updated, but I think the merger is complete.

不过股价已经涨上去了。

But it's it's traded up.

没什么问题。

It's fine.

跟OpenAI一个月的烧钱速度比起来，这点钱根本不算什么。

It's tiny compared to It's not even a month of, of OpenAI's burn.

是的。

Yeah.

瞧好了。

There you go.

我觉得这背后其实有两个完全不同的故事。

I just think there's like two very different stories.

一个是现金流的故事。

There's like a cash flow.

一个是真正业务的故事，还有一个是叙事的故事。

There's like a real business story, and there's a narrative story.

而现在，市场更看重叙事。

And right now we're open for narratives.

明白了。

Okay.

现在市场欢迎叙事。

Open for narratives.

我也觉得你说得对。

I think you're right too.

我觉得把特斯拉看作一个收购目标是有道理的，但另一个问题来了：还有别的吗？

I think looking at Tesla as an acquisition makes sense, but I think the other another question is what else?

对。

Yeah.

对我来说，关键在于你得保护好这个叙事框架。

And I was I was the question for me is like, again, you have to protect the narrative.

对吧？

Right?

你不能收购一个会破坏这种雄心、反而变成一家普通公司的企业，你懂我的意思吧。

So you can't buy something that breaks the ambition and becomes a real company, if that makes sense.

你只能收购那些不会破坏叙事的东西，我觉得这正是叙事策略的难点所在——你不能搞成AOL和时代华纳那种合并。

Like you have to buy you have to only you can't I think this is part of the challenge of narratives is like you can't you can't AOL Time Warner it.

对吧？

Right?

当年AOL和时代华纳合并，AOL的估值被炒得极高，结果拿这个虚高的股价去收购了实实在在的时代华纳，最后把价值全从时代华纳转移到了AOL，这招确实挺精明的。

Like AOL Time Warner was you have like a massively inflated currency in AOL and they buy a super real company Time Warner and merge them and basically shift all the value from Time Warner to AOL, that that's a very clever hedge.

对吧？

Right?

换句话说，你把虚高的估值换成真实资产，这就会毁掉整个梦想。

Like, effectively, where you hedge out your inflated currency into real assets, that would break the dream.

他不能随便买下一个有工会的公司。

He can't just buy a unionized whatever.

对吧？

Right?

没错。

Right.

这得是个能持续产生现金流的项目。

That's like a cash flowing way.

它必须得符合蜘蛛侠宇宙的设定。

It has to be something that fits into the the Spidey universe.

对吧？

Right?

这就是为什么Cursor能行，因为它正好契合了那种高效实用的生态，但它本身又算不上正经生意。

That makes it that's why Cursor works because Cursor, like, it fits into the universe of things effectively in a productive way where where it's not a real business either.

你懂的。

You know?

对。

Yeah.

其实我有个疑问，他的自研前沿AI战略和数据中心战略，到底哪个才是重点？

Well, I'm actually I think one open question is what is his, like, owned and operated frontier AI strategy versus data center strategy?

他是打算继续把Grok打造成能和对手抗衡的产品吗？

Like, is he gonna stay in the game of trying to make Grok a competitor to

这真的重要吗？

Does it even matter?

我的意思是，这才是关键——我得想清楚，到底什么才是真正重要的。

I mean, this is the question, like, I gotta go ask, like, what what actually matters?

很明显，我们已经不再相信会有一个模型能通吃一切了。

Like, we're so obviously moving beyond the game where anyone thinks that, like, one model will rule them all.

未来是多模型的。

The future is multi model.

把它们结合起来用。

You mix them together.

根据不同的情况，选择最适合的模型来处理。

You think about routing to the right model in the right circumstance.

根本不存在什么万能模型。

Like, there's no one god model.

这明显就是我们正在走向的方向。

Like, that's clearly where we're going.

对吧？

Right?

中国开源模型能搞定90%的重活儿。

Like, Chinese open source models will be able to do 90% of the heavy lifting.

也许你还会用一些专门的模型，专门在某些任务上表现更好，但成本会高出百倍，得看值不值。

Maybe you'll have some specialized models or end thought would be slightly better at x, y, and z, but be a 100 times more expensive and see a route to it.

关键是，现在讨论的重点已经不再是‘我能不能做出最大最强的模型’这种故事了。

And, like, it's just not the question on narrative rotation is, like, are we at a point where, like, it just doesn't like, the story of can I build the best biggest model?

这还重要吗？

Like, does it even matter anymore?

反而是，你是不是想掌控那些做矩阵乘法的硬件，因为这才是基础设施？

Or is it like, yeah, you wanna own, like, the machines that do matrix multiplication because it's infrastructure?

拥有处理大规模数值计算的能力，本身就有价值。

There's value to, like, having the capacity to multiply large numbers.

你应该把各种模型都用起来，而不是只依赖某一个。

You wanna use all the models kinda done.

你懂我的意思吧？

You know?

我有点明白你的意思。

So I kind of get that.

但与此同时，我还没看到开源模型和其他模型真正侵蚀到前沿模型的市场份额。

At the same point, like, I don't really see this, like, erosion of open source and other models into the market share of the frontier models yet.

我觉得这种趋势马上就要来了。

And I feel like it's coming.

因为前沿模型的成本实在太高了。

I I because the cost of the frontier is so high.

这种对世界的看法，我觉得很有道理。

And that vision of the world, makes a lot of sense to me.

但如果你看看过去几个月的情况，会发现恰恰相反——前沿模型反而在扩大市场份额。

But I think you can look at the last several months and see the opposite pattern, which is that the frontier is actually gaining share.

我不觉得这是真的。

I don't know that that's true.

我觉得这些研究都有很大问题。

I think these all these studies are like quite flawed.

对吧？

Right?

我的感觉就是这样。

Like is my sense.

我其实很想弄清楚背后真实的情况是什么，但我能告诉你的是……

I I don't I would love to get to the ground truth of what the dynamics are that are happening, but here's what I can tell you.

你有没有看到公司开始转向开源？

Are you seeing companies swap in open source?

是不是这样

Is that

大家都在谈论这件事。

Everyone's talking about it.

我的感觉是，

And my sense is that

我知道。

I know.

但他们真的在这么做吗？

But they doing it?

我觉得是的。

I think so.

但说实话，要准确衡量这个难度挺大的。

But, like, not in, like I I think it's hard to measure is the honest answer.

我得说，像OpenRouter这样的公司，可能是最典型的例子，

I will I will say that, like, OpenRouter, which is probably the poster child

对。

Yeah.

它的发展速度非常快，估值也大幅上涨，前景很好。

For this, is growing like crazy and the valuation's way up and is looking good.

还有像DigitalOcean这样的公司，也在谈论另一个Sam。

There's companies like DigitalOcean talk about another Sam.

我对炒股真是一窍不通。

I'm so bad at trading stocks.

我炒股有多差，简直让人难以置信。

It's like wild how bad I am at trading stocks.

你都八个月没提DigitalOcean了，可它才是我一直在用的核心引擎，等等。

You haven't been talking about DigitalOcean for like eight months and how like, it's the core engine I use and whatever else.

是的。

Yeah.

你帮我 setup 了这个。

You helped set, you set me up on it.

看看股价吧。

Look at the stock price.

早该买一些了。

Probably should have bought some.

好吧。

Okay.

我们可以分析一下你的交易习惯。

Well, we can analyze your trading habits.

就在上个月，他们发布了非常积极的指引，直接从一个增长缓慢的基础设施服务商，转型成了 AI 平台——这我早就看出来了，因为我一直拿它当 AI 平台在用。

Because people just in the last, like, month, they came out with great guidance, they re they basically got recast from being a slow growing infrastructure provider to, like, an AI platform, which I could have told you because I was using it that way the entire time.

股价现在涨得飞快，而我却完全没赶上。

And the stock is, like, ripping, and now I'm behind the ball on it.

但是呢

But, like

我觉得你得相信自己，萨姆。

I think you just have to believe in yourself, Sam.

就相信吧。

Just believe.

是的。

Yeah.

我知道。

I know.

谁能想到，我的问题其实是对自己缺乏信心呢？

Who knew that my problem is a lack of belief in my own self?

是的。

Yeah.

谁能料到呢？

Who knew?

不过我想说的是，他们其实开启了一整套新的接口，本质上就是把不同模型混在一起用。

But like, what I was gonna say about them is like, they just opened up a whole new set of endpoints, which are just like mixing models.

对吧？

Right?

就是说，让我们来负责前端，从原理上讲这完全说得通。

Where like, yeah, just like let us front them and And the physics of it makes sense.

对吧？

Right?

当然了。

Like Of course.

你只要想想这背后的原理，比如在处理超大规模数字运算时，开源项目正在快速追上来。

If you just think about the physics of all this, like you're multiplying big numbers, the open source is catching up.

传输这些模型根本不需要消耗多少能源成本。

There's no cost to shipping it like energy.

也没有什么监管垄断的问题。

There's no regulatory capture.

当然了，你肯定会选成本最低的方式来组合它们，对吧？

Like, of course you're gonna mix them for the lowest possible cost, you know?

是的。

Yeah.

我同意。

I agree.

我只是好奇，为什么到现在还没发生呢？

I'm just curious why it hasn't happened yet.

也许已经在发生了，或者

And maybe it is happening or

也可能确实发生了，但我认为目前对这些现象的衡量方式其实很不可靠。

Or maybe it is, but the measurement I I think the measurement's really suspect on what's happening on a lot of this stuff.

那正确的衡量标准应该是什么？

So what's the right measurement?

比如生成的令牌数量？

Like tokens generate, like?

不是。

No.

因为不是所有时间单位都一样。

Because the time is not all tokens are the same.

对吧？

Right?

所以很难建立一个基准。

So it's, like, very hard to, like, baseline.

我的意思是，这其实是整个问题的经济困境——我们根本找不到合适的说法来衡量它。

I mean, the this is really the economics problem of the entire thing, right, which is we don't have the right language for top measuring it.

我们连谈论生产力的合适术语都没有。

We don't have the right measure language for talking about productivity.

我们甚至没有合适的语言来界定这到底是什么类型的问题。

We don't even have the right language for actually doing like, what type of problem is this?

这到底是渐近问题，还是指数型问题？

Is this, like, an asymptotic problem, or is it an exponential problem?

对与错之间的价值到底在哪里？

What's the value being right versus wrong?

我其实写过一篇小短文谈这个，要是我现在还在上大学，第一，我会选经济学；第二，我会做大量研究，试着写论文，深入研究代币经济这个问题。

Like, I I I wrote a little mini essay about this, but I I really believe, like, if I were in in college, one, I'd be taking economics, and two, I would be doing a ton of research and trying to, like, you know, write papers and get involved in the question of, like, tokenomics.

不是那种真正的代币经济，

Not, like like, true tokenomics.

而是代币背后的经济逻辑，我觉得这会是金融领域一个巨大的突破点，同时也很有可能帮人拿到诺贝尔奖。

The economics of tokens because I think that is gonna be, like, an enormous area of financial breakthrough, but also, like, it's a great way to win a Nobel Prize.

是的。

Yeah.

说得对。

Amen.

这是个值得研究的好课题。

Good topic to study.

再谈两个跟AI相关的其他话题。

Two quick other AI topics.

我们聊聊Meta那边的情况好吗？

Should we discuss what's going on in Meta?

我觉得咱们得谈谈这个。

I think we need to discuss it.

Meta那边到底怎么了？

What's happening in Meta?

那边现在挺难的。

It's like tough times over there.

《连线》杂志有篇报道，说AI部门士气很低落。

Wired had a great story that kind of morale is low in AI.

很多人离职，高管们只好开全员会，承诺改善士气、提供更好的零食。

People are leaving a lot of all hands where executives are promising to improve morale and better snacks.

这算怎么回事？

Like, what is this?

你把这事儿给咱们理一理。

Put put this in the cycle for us.

说实话，我对这些细节也不太清楚。

Well, I don't really know any of details.

你知道的，从长远来看，我一直是Facebook的坚定支持者。

And, like, as you know, I will, like I'm a huge Facebook bull long term.

但我得认真想想，现在可能的情况是——我很好奇，作为Meta的产品工程师，现在是什么感受。

But I had had to really go say, well, here's what I think is probably happening is, like, I wonder how it feels to be like a product engineer at Meta right now.

对吧？

Right?

如果从故事的角度来看，我觉得情况确实不一样了。

Like, if you think about it, like, I'd actually argue in terms of narratives.

记得以前Meta有Facebook的时候，它的故事可以说是最好的之一。

Remember Meta had Facebook at the time, had probably one of, like, the best possible narratives.

还记得那个‘让世界更开放、更互联’吗？

Remember Make the World More Open and Connected?

记得。

I do.

那真是太棒了。

That was awesome.

对吧？

Right?

你要知道，我那时候在Meta，也就是Facebook工作。

Like, if you think about, like, like, I worked at Meta, at Facebook in that era.

我也记得那个世界越来越有同理心的年代。

I also remember the era of increasing empathy in the world.

你还记得吗？

Do you remember

是的。

that?

对。

Yeah.

那确实有过，但没那么有吸引力。

That was but that was less compelling.

我的意思是，如果你想想那些老掉牙的叙事，它们往往都会经历几个阶段。

I mean, that, like, if you think about, like, narratives, like, that are old go through several of them.

但你想想看，早期的Facebook，那时候特别鼓舞人心，大家都拼命工作，真心相信它，核心口号就是：我们到底在做什么？

But, like, if you think about it, the early Facebook, which was so inspiring, everyone worked so hard and really believed in it, right, was this, like, mantra of, like, what are we doing?

我们是在让世界变得更开放、更互联。

It's like we're making the world more open and connected.

这个口号并不是对每个人都有效，但在那个年代，对那些真心投入、认同这个理念的人来说，它极具感染力。

And that was not inspiring to everyone, but that was deeply inspiring in that era to people who worked there who worked and really really cared about that narrative.

我不禁想，当士气被用来驱动人拼命工作时，如果他们深信自己在做的事，会不会也让人做出很糟糕的行为？

And I do wonder, like, if the problem, when you think on morale like, can make people work really hard and do really shitty things if they believe in what they're doing.

就像他们心里想：我既然来了，就是自愿加入的。

Like, if they're like, I'm here and I'm I signed up for this.

这简直就是马斯克的套路。

This would be the Elon playbook.

这确实是标准操作，有趣的是，当年这其实是扎克伯格的打法。

This is the play well, this is the funny thing is this was the Mark Clear.

让世界更开放，我觉得这原本是他个人的信念。

Mean, make the world opening act, which I believe was him personally.

确实很棒。

Like, great.

对吧？

Right?

一旦有了这样的信念，做任何事都会顺风顺水。

And, like, when you have that narrative, like, you just have such tailwinds on everything.

对吧？

Right?

但问题在于，那些有选择余地的聪明人，他们会想：好吧。

And the problem with people, right, or, like, smart people who have options, right, is you're like, okay.

我现在在Meta。

Like, I'm at Meta.

我二十年前加入，是因为我相信社交软件能帮人连接，能帮人做点什么——但到了2026年，这种信念已经不像2010年那么清晰了。

I signed up in some form twenty years in because I believed in social software, helping people connect or helping people you know, whatever whatever it was that brought you, which was not as crisp in 2026 as it was in 2010.

对吧？

Right?

剑桥分析事件之后，还有那些别的丑闻。

Post Cambridge Analytica and whatever other scandals.

但你要是仔细想想，就会觉得，嗯……

But, like, if you think about that, like, you're like, okay.

现在我们又在搞人工智能。

Now we're working on AI.

那你就会想，我真想做人工智能吗？

And you're like, but do I wanna work on AI?

我是不是还该继续做社交软件？

Like, do I believe didn't I am I here to work on social software?

我到底在干嘛？

Like, what am I doing?

对吧？

Right?

我觉得这正是我想说的，之前也提过，就是AI的意义缺口问题。但归根结底，人得做自己真正在乎的事。

And I think that's, like, kind of the thing I mean, I've talked about this before as, like, the meaning gap with AI, but, like, I just think people need to do things they really, really care about.

我跟初创公司聊的时候也总说，有人来路演，讲得跟商学院案例似的。

This is the thing I always talk about with startups too is I'm like, you know, when people come and pitch baby companies, sometimes they sound like business school cases.

听着是门好生意，可你到底为啥非得做这个？

We're like, that's a good business, but why on earth do you care about this?

你懂我的意思吧？

You know?

到底谁会在意啊？

Like, what, like, what like, who cares?

对吧？

Right?

想让团队有干劲、让公司成功，非得有这份热忱不可。

You really need that level of passion to keep an organization happy and successful and make it win.

说到信息这块，你确实是有这种热情的。

And, like, I mean, you have that with the information.

很多人看法不一，但真正做信息报道的那些人，他们是真心相信你做的事，所以才愿意干下去。

It's like you can lots of people have different opinions, but ultimately, like, the reporters who work on the information, like, deeply believe in what you do, which is why they do it Yep.

也正因如此，他们才愿意为你工作。

And why they work for you.

对吧？

Right?

我觉得关键区别在于，你要么得付给员工足够多的钱，让他们不在乎别的，只在乎你赚钱。

That's like the I think the big differentiator is you you just have to have you either have to pay people so much money they don't care.

对吧？

Right?

他们信的是你能发财。

They believe in you getting rich.

这其实就是华尔街或者对冲基金的思路。

Like, that's the Wall Street or the hedge fund answer.

对吧？

Right?

举个我特别喜欢的例子，比如你找到那些做在线游戏的创业者。

Or, you know, one of my favorite examples is if you find founders who are working on, like, online gaming.

对吧？

Right?

只要他们自己真心投入，目标明确——我就想赚最多的钱。

It's fine as long as they're in it and they're like, I'm here to make maximum money.

而且他们还坦诚承认这一点。

And if they're honest about that.

对吧？

Right?

他们就是会说：没错，我就是冲着钱来的。

Like they're like, oh yeah, that's what I care about.

我就是为了赚钱才干这个的。

I'm here for money.

这样就会吸引一批特定的人，他们基于这种信念去做相应的事。

And then you get a certain type of people that do a certain set of things based on that belief.

对吧？

Right?

但如果你不这么做，你的任务就是筛选出那些真正关心你使命的人，要知道大多数人并不在意。

But but if you're not doing that, your your job is to filter to the people who like actually give a shit about your mission, understanding that most people don't.

没错。

Right.

是的。

Yeah.

好的。

Okay.

那个小小的Meta手表。

That little watch Meta.

我感觉那里还会出更多事，尽管我们目前一无所知。

I feel like there will be more shoes to drop there without knowing anything.

看起来真是个艰难的时期。

It just seems like a tough time.

他们只需要找到那个清晰明确的使命。

They just need to find what that crisp mission is.

然后重新调配人力，让真正认同的人加入进来。

And then like resort their human capital to People are like, I'm in.

这至少是我的感觉。

You know, is my sense.

一年前，马克用来成功吸引大量顶尖人才的AI使命，是‘个人超级智能’这个理念。

Well, the mission around AI that Mark used to successfully recruit a lot of top talent a year ago was this idea of like a personal super intelligence.

他经常公开谈论这一点，说现在很多前沿AI模型都聚焦在提升效率上，但他认为人类会发现更多样的应用场景，也会更倾向于这些用途。

Like he really and he talks publicly about this all the time that, you know, so many of the frontier AI models are focused on like productivity maxing, but he thinks that humans are gonna find far more use cases and gravitate to more use cases of AI.

我说这话是出于对他的极大推崇和信任。

Look, and I say this as a huge, enormous fan of his and believer in him and like the whole thing.

但这个说法不如‘让世界更开放、更互联’来得清晰有力。

I just, it's not as crisp as make the world more open and connected.

是的。

Yeah.

我同意。

I agree.

我同意。

I agree.

Anthropic现在有麻烦了。

Anthropic's in trouble.

你有关注这件事吗？

Have you been following this?

是吗？

Are they?

还是他们营销做得太厉害了？

Or are they doing incredible at marketing?

对。

Yeah.

我们一遍又一遍地在讨论他们。

We're talking about them again and again and again and again.

我觉得

I think

Anthropic 这回真火了。

Anthropic's killing it.

他们搞了个模组，公关效果好到不能再好了。

They've successfully released a mod like, couldn't ask for better PR.

可他们还没发布模型呢。

Well, haven't released the model.

那就更妙了。

Even better.

他们根本不用花一分钱。

They don't have to pay for it.

没错。

No.

我们还用不了。

We can't use it.

我们不能用Fable。

We can't use Fable.

但这不正是更好吗？

But isn't that even better?

简直完美，他们根本不用做任何事。

Like, the best There's not

他们的服务器现在根本没在超负荷运行。

their their servers aren't burning right now.

这就相当于埃隆说：我当然能去火星，但我就不去，因为有人告诉我不准去。

That would be the equivalent of, like, Elon getting to say, oh, I definitely can go to Mars, but I'm not going to because someone would tell my told me I'm not allowed to.

是的，没错。

Oh yeah.

这太棒了。

That's pretty.

这简直是完美的营销。

It's like the best possible marketing.

就像是

Like, it's like

这还挺搞笑的。

That's pretty funny.

我真希望推出一些大家觉得特别厉害、结果我却不能发布的功能，这样我就不用付钱了，而你们还是会注册并用我的其他产品。

I would love to release things that everyone believes are so powerful that I'm not allowed to release them and therefore don't have to pay for them, you're gonna sign up and use my other stuff.

你说得对。

And you're right.

那东西一上线，我们所有人肯定都会惊呼：天哪。

The second that thing comes out, we're all gonna be like, oh my god.

它到底好在哪里？

What was so great about it?

他们网站上真的有小提示，我最近刚好登录了Anthropic的官网，看到了。

They literally have little things on their I saw this because I happened to log in to the actual website, Anthropic recently.

上面还写着：Fable 5 已被禁用。

And, like, they have a little thing being like, Fable five is disabled.

我觉得这简直是史上最牛的营销了。

I'm like, that's the greatest marketing I've ever seen.

他们真该直接拍广告说：Fable Five 已停用，因为太出色了。

Like, they should literally be making ads being like, Fable five is disabled because it's too good.

对吧？

Right?

没错。

Yeah.

你可能会说，我也不确定。

And you're like, I don't know.

你还想怎么样呢？

What more do you want?

我觉得人们其实比公众以为的要聪明得多。

Like, I think people that people are much smarter than, like, the public gives them credit for.

对吧？

Right?

很多时候，人们总爱过度解读，觉得背后有什么高明的策略，其实根本不是那么回事。

Like, a lot of times, I think people, like, overread into things and think that, like, there's some great strategery when there isn't.

这次的情况不一样，这明显就是完美的设计，我敢打赌他们完全是故意的。

In this case, it's like I mean, this is perfect, and I'm sure they're doing it a 100% on purpose.

你懂的吧？

You know?

对。

Yeah.

这确实挺有意思的。

Well, it's fascinating.

上周末我们听说，亚马逊曾主动联系政府，报告了一些安全漏洞。

We had a story over the weekend that Amazon had reached out to the government finding some vulnerabilities.

但又有人说，他们其实是等达里奥没接电话之后才去联系政府的，这说法根本说不通。

And then someone else reported they only reached out to the government after Dario didn't pick up the phone call, which makes no sense.

还有人说，这本来就是他们和政府定期会议的一部分。

And then someone else reported it was part of their regularly scheduled meeting with the government.

但现在，Anthropic 自己的合作伙伴也开始谈论这个模型了，虽然他们并没有特别强调它的能力。

But now you have Anthropic's own partners that are involved in kind of talking about, you know, this model, although they weren't necessarily talking about the power of it.

我觉得这次 Anthropic 在华盛顿的最新动态中，最值得关注的是，外界对人工智能公司里外籍员工能接触哪些信息越来越关注了。

And I think one of the most interesting things to come of this latest round with Anthropic in DC is the growing scrutiny on foreign nationals inside AI companies and what they can have access to.

政府现在已经明确表示，Anthropic 的外籍员工不得接触这个模型。

The government's now said that foreign nationals inside Anthropic can't have access to it.

我有点担心这个做法。

I kind of worry about that.

如果一个在美企工作的员工能接触什么软件，竟然要取决于他的国籍，那局面就太糟糕了。

I, I think we're be in a very bad place if all of a sudden the software you have access to working on is dependent on your nationality when you're employed by a Us company.

这确实不太对劲。

Seems bad.

这对美国公司来说恐怕也不利。

It's probably bad for The Us companies.

对吧？

Right?

大概是吧。

Probably.

对吧？

Right?

我觉得这可能是唯一有问题的。

Like, think that's only one that's bad.

没错。

Yeah.

我们一直受益于全球的人才。

We've benefited from talent around the world.

确实如此。

Yeah.

关键还是看你到底觉得这些技术是不是真的那么独特和强大。

It really just depends, like, if you actually think these things are, like, uniquely powerful or not.

对吧？

Right?

如果你觉得，我们发明了暗物质，别人根本不可能独立发明出来，除非去偷。

If you're like, oh, we've invented dark matter and like, it's impossible for someone else to invent it unless they steal it.

如果问题是，天哪，那个叫什么来着？我怎么突然想不起来了？

If it's if it's, oh god, what's the, god, why am I blanking on this?

谁从众神那里偷走了火？

Who stole fire from the gods?

不，我不知道。

No, I don't know.

得了吧。

Oh, come on.

这可是希腊神话里特别重要的一个故事。

This is like one of those important Greek myths.

抱歉。

Sorry.

我不

I don't

我对这个不太了解。

know about that.

普罗米修斯。

Prometheus.

普罗米修斯。

Prometheus.

谢谢。

Thank you.

如果这就像普罗米修斯的故事一样，众神握着火种，你的任务就是确保普罗米修斯别把火偷走，那倒也说得通。

It's like, if this is like a Promethean thing, where like the gods have fire and your only job is to make sure that Prometheus doesn't steal it, then sure.

把外国人挡在外面。

Keep the foreign nationals out.

有道理。

Reasonable.

但我觉得更可能的情况，其实是像四分钟跑一英里那样。

But, like, I don't I I think it's much more likely that it's kinda like the four minute mile.

对吧？

Right?

一旦你知道这事能成，别人也就能做到，而且他们真的会去做。

Which is, once you know it's doable, other people can do it and they will do it.

这样的话，这策略就太蠢了。

In which case, like, that's a stupid strategy.

对吧？

Right?

因为他们根本会自己去造火。

Because, like, they'll just make their own fire.

是的。

Yeah.

这正是现在芯片和中国正在发生的事。

Well, that's what's happening in chips and China.

再说，你要是回过头看，就在英伟达崛起之前，好几年前，很多人都看不起英伟达，觉得他们不过如此。

And this is, like, look, if you go back, like, remember the before Nvidia, like several years ago, people really poo pooed Nvidia because they're like, dah.

是的。

Yeah.

其实很简单。

They're it's easy.

我们谁都做得出这种芯片。

Like, we can all make these chips.

对吧？

Right?

结果呢，大家都觉得，哎呀，无所谓。

And like, it turns out like and and like everyone's like, dah, whatever.

反正他们也不会有事。

It doesn't you know, they're they'll be fine.

这不过是个十二个月的短期现象，撑不了多久。

Like, it's a twelve month trade, and, like, it's not gonna sustain.

从长远来看，说不定还真有可能。

And, like, that actually abstractly might eventually be true.

对吧？

Right?

但事实证明，从好几年的时间来看，这并不成立。

But it's turned out to be not true on a several year basis.

实际上，事情比想象中难多了。

Like, it's turned out to be much harder.

对吧？

Right?

而且，越来越多的人开始意识到这个问题。

And, like, there's come more compounding who will realize.

所以我觉得，这正是很多这类问题的关键所在。

So think this is, like, this is the big balance in, like, a lot of these.

我的意思是，这跟火箭项目也有关系。

I mean, there's into the rockets.

我的意思是，你知道的，就是埃隆那边给停了。

I mean, like, you know, dropped by Elon.