Live in the Future

频道: Naval
视频: https://podcasts.apple.com/us/podcast/naval/id1454097755?l=zh-Hans-CN&i=1000775254284
原文语言: en
统计: 共 106 轮

[0:00]

Welcome to all the legends out there listening to the Naval podcast, the number one podcast in the universe and your authoritative source for new knowledge. We've got three founders with us here today. On my right is Garry Tan from Y Combinator. They say it's the most reliable way to become a billionaire is to join the Y Combinator program.And start as a trillionaire.Start as a trillionaire.That's one way to do it.We're all drinking, by the way, except one person here is not drinking. You have to figure out at the end who's actually sober.And then on my left, I've got Daniel from Able Police. They started off by turning body cam footage into police reports, but now they have a whole set of tools for compliant AI chat, translation. What else do you guys do?

欢迎所有正在收听 Naval 播客的传奇人物——全宇宙排名第一的播客，你获取新知识的权威来源。今天我们请到三位创始人。我右手边是来自 Y Combinator 的 Garry Tan——人们说，成为亿万富翁最靠谱的路径就是进 Y Combinator。（那不如一开始就是万亿富翁。）（对，一上来就当万亿富翁，这也是一种办法。）顺便说一句，我们都在喝酒，只有一个人没喝——最后你们得猜出到底谁是清醒的。我左手边是来自 Abel Police 的 Daniel，他们最早是把执法记录仪的画面转成警情报告，现在做了一整套合规的 AI 聊天、翻译工具。你们还做什么来着？

[0:49]

Citizen reporting, all kinds of things.Then on my right, I've got Farbood from A-LIST, the health super app. As usual, we don't really care about what these guys are building. We care more about what they're learning about building, what they're figuring out, what they're still trying to figure out, and the principles that they can share with other founders. So does anyone have anything they want to jump in with?

市民报案之类的，各种各样的。我右边还有来自 A-LIST（健康 super app）的 Farbood。老规矩，我们其实不太关心这几位在做什么产品，我们更关心他们在做的过程中学到了什么、想明白了什么、还在琢磨什么，以及有哪些原则能分享给其他创始人。所以，谁想先起个头？

[1:10]

I mean, the topic that everyone's thinking about right now is AI, right?AI. Yeah, I think I've heard something about it.Yeah. I kind of hate talking about it because it's the most perishable thing. Like, you know, the podcast will be obsolete by the time it ships.But it's still also the most interesting because it's the fastest changing thing in the environment that's high impact. So it's the steepest learning curve. And you're always trying to trade notes with people to kind of see where they're at. In fact, that was at a thing with you recently where I bumped into you and the conversation immediately turns to AI, of course.I've got to, yeah.So we can avoid it for a little while, but I feel like we'll circle back on that topic.I got a question for you. What are you being floored by at YC in terms of like what you see the companies doing with AI?

现在大家都在想的话题就是 AI 对吧？（AI，对，我好像听说过。）说实话我有点不想聊它，因为它是最容易过期的东西——这期播客等上线的时候可能就已经过时了。但它也最有意思，因为它是环境里变化最快、影响又最大的东西，学习曲线最陡；你总想跟别人对一对笔记、看看大家进展到哪了。其实最近有个场合我碰到你，话题立马就转到了 AI。（可不是嘛。）所以我们可以先躲它一会儿，但估计还得绕回来。我先问你个问题：在 YC，你看到这些公司拿 AI 做的哪些事让你惊掉下巴？

[1:50]

So I think the awkward thing is like I basically like slept three hours a night for like four months and then somehow like went from not coding at all to like basically teaching people how to do it and making like one of the top 100 open source packages for how people vibe code called GStack.So that is very, very weird. And now like at some point I met Pedro from Brex.And then after that, I basically converted from like Claude Code, like Cult to OpenClaw Cult within like 24 hours.And then now we'll use both. But like it seems like basically if you, you know, the craziest idea that I discovered is that if you're just willing to spend like $100,000 a year on tokens, you can basically live like you are a normal citizen in 2028.Like it's just pretty clear that token costs are going to come down. Compute is going to go way, way up.Like, you know, we think like 90,000 X or so there will be 90,000 X the amount of like inference from here to like three years from now.Oh, you plot it out in some curve?

尴尬的是，我基本上连着四个月每晚只睡三小时，然后不知怎么就从完全不会写代码，变成了教别人怎么写，还做出了一个 vibe coding（氛围编程）领域排进前 100 的开源包，叫 GStack。这非常非常离谱。后来我遇到了 Brex 的 Pedro，之后我大概在 24 小时内就从 Claude Code 那一派『邪教』转投到了 OpenClaw 派。现在两个都用。但我发现最疯狂的一个道理是：只要你愿意一年在 token 上花 10 万美元，你基本就能过上『2028 年一个普通人』的生活。因为很明显 token 成本会降下来、算力会疯涨——我们估计从现在到三年后，推理量会涨大约 9 万倍。（你还画了条曲线？）

[2:57]

Yeah.90,000 times.Yeah. I mean, it's like many, many orders of magnitude.Which is kind of baked in both with the chips and the data centers they're building out.Yeah. I mean, a lot of people say that NVIDIA is way overpriced, but what if it's way, way underpriced by like several orders of magnitude?

对，9 万倍，很多很多个数量级。这基本上已经锁死在他们正在建的芯片和数据中心里了。很多人说 NVIDIA 估值太高了，但万一它其实是被低估了好几个数量级呢？

[3:12]

Wow.Right.90,000 by when?I don't know. 24 to 36 months.Okay. So five orders of magnitude basically.Yeah. I mean, we might be off by a couple of orders of magnitude.Yeah, yeah. But either way, it's like...Yeah, yeah. But either way, it's like...I mean, it's actually still a blue ocean right now.But you know, the thing that's crazy is if you go up by five orders of magnitude, it's not just what it does in terms of usage, but in terms of capabilities.Because every order of magnitude you go up, new capabilities emerge.And, you know, I've been pretty behind the curve on AI for a while. Like I missed it in 2020 to 2022 because, you know, I'd grown up my whole life hearing about AI from when I was doing CS and just assumed it was one of those things like fusion that was never going to come.And then it came.Yeah.And when it came, it came really fast. And at first, a lot of the stuff people were saying about it sounded breathless and like they were talking their own book.Yeah. It seemed crazy.But they turned out to be right so far.Yeah.Right? And it's not... So I think the big... There were many big questions about AI coming up, right?

哇。（9 万倍，到什么时候？）不知道，24 到 36 个月吧。（好，那基本就是五个数量级。）对，我们可能会差个一两个数量级。但不管怎样——现在其实还是一片蓝海。可怕的是，涨五个数量级，变化的不只是用量，还有能力：每上一个数量级，就会有新的能力涌现出来。我其实有很长一段时间在 AI 上很落后，2020 到 2022 年那波我就错过了——因为我从学计算机起就一直听人讲 AI，就默认它是像可控核聚变那样、永远不会真来的东西。结果它来了，而且来得特别快。一开始很多人讲得神乎其神、像是在给自己的盘子唱多，听着很疯，但到目前为止他们竟然都说对了。

[4:06]

Now, I think the... Like one question I used to have in my mind was like, is it going to be distributed or is it going to be centralized?You know? And now that's kind of morphed into like, is open source going to be good enough for most use cases or are you always going to need the frontier proprietary models?

我以前脑子里有个问题是：它会是分布式的，还是中心化的？现在这个问题演变成了：开源对大多数场景够用吗，还是你永远都得用最前沿的闭源模型？

[4:20]

Right? There's the whole thing about like, it's weird that China is making all the open source stuff and the U.S. is doing all the closed source stuff.And so where do we go from there? Right?Right. There's the question of, is it going to be nationalized or is it going to be sort of private sector?

对吧？还有个很怪的现象：中国在做所有开源的东西，美国在做所有闭源的东西。那接下来往哪走？还有一个问题是：它会被国有化，还是留在私营部门？

[4:34]

Especially if you're saying, hey, it's a new Manhattan project.These are the new nuclear weapons.Then like private companies building them and controlling them.It doesn't make sense.But I think the most interesting remaining question, there's still one that I really care about over the others where there's no clear answer yet,is I think the people in the labs believe that the scaling laws are such that the AIs will keep getting smarter until they become smarter than the smartest humans.Probably through the process of recursive self-intelligence, recursive improvement, but they call it ASI.And I think that is the big question.So, you know, two years ago, you could have taken the viewpoint that AI will get you mid at everything, but it's not going to be the pro at anything.Right?

尤其如果你说这是新的曼哈顿计划、这些是新的核武器，那由私营公司来造、来掌控它就说不通了。但我觉得最有意思、也最没定论的一个问题是：实验室里的人相信，按 scaling law（规模定律）推下去，AI 会一直变聪明，直到超过最聪明的人类——很可能是通过递归式的自我改进，他们管这叫 ASI（超级人工智能）。我觉得这才是那个大问题。两年前你还可以持这种观点：AI 什么都能做到中等水平，但成不了任何一样的专家。对吧？

[5:15]

So now you're getting to the point where it'll get you the pro at everything, but we don't know if it'll get you the last creative mile.Like that last bit of like moving out of the system and creating something new, not just recombining what's already in the training sets.Did the progress that it made on the Irish math problem bother you?

现在已经到了这个地步：它什么都能做到专家水平了，但我们还不知道它能不能走完『最后那一英里的创造力』——就是跳出既有系统、创造出全新的东西，而不只是把训练集里已有的东西重新组合。AI 在那道数学难题（Erdős 问题）上取得的进展，有没有让你心里不舒服？

[5:32]

It bothers me for sure.I mean, it not bothers me.It's a process of discovery.I don't want to be bothered by laws of the universe.Right?It bothered me.Really?It definitely surprises me.Why did it bother you?

肯定有点。（也谈不上不舒服——这是一个探索发现的过程，我不想被宇宙的规律搞得心烦。）但它确实让我不舒服。（真的？）它绝对让我很意外。（那你为什么会不舒服？）

[5:40]

But at the same time, I still don't see like broad general creativity.But anyway, that's a separate point.What was bothersome?I was a math major and I really loved proofs and something actually making progress in that domain in a meaningful way that looked really original.Like there's something creepy about it almost.But it was directed by someone, right?

但同时，我还是没看到那种宽泛的、通用的创造力——不过这是另一个话题了。（到底哪里让你难受？）我大学是数学专业，特别喜欢证明，而现在有个东西在那个领域里做出了真正有意义、看起来很原创的进展，这几乎有点让人发毛。（但那还是有人在给它指方向的，对吧？）

[5:57]

Sure.And I think that's fine.But everyone was surprised who was involved.Right, right.Right.Like I don't know if you saw the interview videos.I guess all that stuff is kind of like boring.It just means like go do something else.Like if you have limited time to think about something, do you want to spend it opining over the fact that there's a machine that's better than a human that's something now?

当然，我觉得这没问题。但当时参与的每个人都很惊讶。（对对对。）不知道你有没有看那些采访视频。我觉得纠结这些其实挺无聊的——它无非就是在说：去做点别的吧。如果你思考的时间有限，你真想把它花在『唉，现在有台机器在某件事上比人强了』这种感慨上吗？

[6:14]

Or just go apply your humanness to something that you haven't thought about.Or actually to riff on Gary's point, are we going to end up in a place where a machine is strictly better than a machine with a human?

还不如把你身上『人的那部分』用到你还没想过的事情上。或者顺着 Garry 的点往下推：我们会不会最终走到——一台机器，严格地比『机器加人』还要强？

[6:26]

Like a human in the loop doesn't add anything.That's the biggest question.That's the biggest question, right?Because chess went that way.Because I hope not.There was Centaur chess.Somebody was telling me about this.But there was Centaur chess where like the chess computer plus the human would beat just a chess computer.But eventually just the computer alone beat the Centaur model.I guess for me, I'm dealing with AI in the most practical way humanly possible.You know, spinning up thousands of agents for people on demand.We had to build a massive fleet management system, an eval harness.The reality of it is like I don't think intelligence is the bottleneck.Cost is the bottleneck.Yeah, right now.Right now, when we started doing it, every single person in our app has their own open claw running for them.It's a really powerful.For health.Yeah.Oh my God.Like mine wrote an entire.But what about this open claw then?

也就是说『人在回路里』根本不再增加任何价值——这才是最大的问题。（我希望不是。）象棋就是这么走过来的：曾经有『半人马象棋』，电脑加人能赢过纯电脑；但最终，纯电脑单打就赢过了半人马组合。对我来说，我是在用最接地气的方式跟 AI 打交道——按需给用户批量拉起成千上万个 agent。我们不得不搭了一整套庞大的 fleet 管理系统、一个 eval harness（评测框架）。现实是：我不觉得智能是瓶颈，成本才是瓶颈。（现在是。）我们刚开始做的时候，app 里每个用户都有一个属于自己的 OpenClaw 在跑（用于健康）。太强了——我那个甚至自己写了一整个……（那这个 OpenClaw 到底是啥？）

[7:14]

What's that?People just want open claw right now.No.Or they should want it.But it's not open claw anymore because when we started with open claw.So expensive.And Opus, it was $100 a month per person.Yeah.We spent three, four months driving that down to $2.84.Do you use Pi now?

什么？现在大家就是想要 OpenClaw。（不，是他们应该想要它。）不过现在已经不叫 OpenClaw 了。我们刚开始用 OpenClaw 的时候太贵了——配上 Opus，每人每月要 100 美元。我们花了三四个月把它压到了 2.84 美元。你现在用 Pi 吗？

[7:30]

We built an entire stack that is just completely wild that has an eval harness and then is like an entire agentic fleet that can spin up and down elastic.So that I deal with the mundane parts of that stuff all the time.I still think what's way more interesting is to think about what humans will do when they don't have to do these things.But I think that there is a very real phenomenon going on right now where the people who would know the most in the labs are basically saying there will be nothing left for humans to do.Really.When you read the lines of what they're saying, they're saying in different ways.But they're saying your harness doesn't matter because within a year the AI will be spinning up harnesses as needed.And you won't even be the consumer because it'll be talking to other AIs.And beyond the Erdős problem, it will be solving fundamental problems in material science and physics and math and engineering and health.Great.Right.Yes.But then, you know, that's taking the average person out on the street, working them up into a frenzy.They're getting angry and scared because fear is a precursor to anger.My current joke is to try to get all my smart friends to go work at Anthropic.So then I'll have lots of friends there who will get me into the Anthropic level two bread line.Yeah, exactly.Because right now I feel like pretty solidly in level three.Bread line, yeah.Like, I don't know, level one seems...There's only two jobs, all right?

我们搭了一整套非常疯狂的技术栈：一个 eval harness，加上一整支能弹性伸缩、随时拉起或收起的 agentic fleet（智能体集群）。所以这些琐碎的活我天天在处理。但我觉得更有意思的问题是：当人类不必再做这些事的时候，他们会去做什么。不过眼下有个非常真实的现象：实验室里最懂行的那些人基本上在说——将来不会再有留给人类做的事了。（真的。）你把他们的话读进去，他们其实在用各种方式说同一件事：你的 harness 不重要，因为一年之内 AI 会按需自己拉起 harness；你甚至都不是那个使用者了，因为它会直接跟别的 AI 对话；除了 Erdős 那类问题，它还会去解决材料科学、物理、数学、工程、医疗里的根本性难题。（很好啊。）是很好。但这话传到街上普通人那里，就会把他们煽动到抓狂——他们又气又怕，而恐惧正是愤怒的前奏。我现在的笑话是：想办法把我所有聪明的朋友都塞进 Anthropic 工作，这样等到那天，我至少能靠这些朋友挤进 Anthropic 的『二级救济面包队』。（对，就是这样。）因为我现在感觉自己稳稳地在三级。（面包救济队，对。）一级好像……反正到时候只剩两种工作了。

[8:45]

It'll be an Anthropic employee and sex worker for Anthropic employees, all right?I mean, that's the scary part, right?That's the one.And it's getting people worked up.And there are real-world consequences of that because you don't live alone.You live inside a thundering herd called humanity.And if that thundering herd decides to, like, go left towards a cliff because it's mad or because it's enraged or because it's poison, you're going to go with it.You know, you're not going to be the lone bison sitting out there by yourself on the edge.And people are mad.You said, like, nationalization, right?

一种是 Anthropic 员工，另一种是『专门服务 Anthropic 员工的性工作者』——这就是最吓人的地方，对吧？就这么一个。它正在把人煽动起来，而且是有现实后果的：因为你不是一个人活着，你活在一个叫『人类』的、轰隆隆奔跑的兽群里。如果这个兽群因为愤怒、因为被激怒、因为被『下了毒』而集体朝悬崖那边冲，你也会被裹挟着一起冲——你不可能当那头独自淡定坐在边上的野牛。而现在人们是真的很生气。（你刚说了国有化，对吧？）

[9:12]

Like, that's where it would go, you know?But right now you can see that trend starting with their attacking data centers, right?They're attacking data centers over...That's just stupid.No, no, no, no, no, no, no.We look back on the stuff and we think, like, the spinning loom?

对，事情会朝那个方向走。你现在已经能看到苗头：他们开始攻击数据中心了。他们冲着数据中心去……（那太蠢了。）不不不不，我们回头看这些，就会像看当年砸纺织机（spinning loom）一样。

[9:23]

It's a rational response, but it's an irrational carrier.So they're saying it's because of water.It's not about water.Oh, that was a lie that was debunked.Yeah, yeah.But it's not about reason.I mean, it's like socialism.It's not about reason, right?

这是一种理性的反应，却搭在一个不理性的载体上。他们说是因为耗水——根本不是水的问题，那说法早就被辟谣了。但这本来就不是讲道理的事，就跟社会主义一样，不讲逻辑。

[9:36]

It's about how you get the people going.You're getting poor and you want more money.Or in this case, you're feeling threatened.It's like all in New York City right now.Like, you're going to be replaced and you want to stop this train.You want to be like, I want to on this train.I want to be on this train.Yeah, that's rational.That's totally rational.So in this case, this is a case of, especially when you have the popular delusions of themadness of crowds, you have the ends justify the means.So in this case, the water is just a means.They'd find another one if it wasn't water.There'd be some endangered egret, some bird, you know, some snail.Are they right?

关键在于怎么把人群鼓动起来。你越来越穷、想要更多钱；或者像这次，你觉得受到了威胁。就像现在整个纽约——你要被取代了，你想把这趟车拦下来。（我想上这趟车。）（对，我想上车，这很理性。）完全理性。所以这是一个典型的『群体的乌合与疯狂』的案例，加上『为达目的不择手段』：水在这里只是个借口，如果不是水，他们也会找到别的——某种濒危的白鹭、某种鸟、某种蜗牛。（那他们说得对吗？）

[10:04]

Well, they can't stop it.This technology is a genie.That's what the genie...I don't think that's true.Like, there's been times in the past where, like, yeah, like, smashing, like, spinninglooms actually stopped that revolution.Very briefly.It slowed it down.There was other technologies.But we have spinning looms now.Of course.But there was other technologies.We didn't stop it.There was other technologies where they, like, you know, refused to issue a patent or something.So it became unmarketable because it was just, like, politically going to be too difficult.I think that basically we ingest, like, you know, 100 years ago, 50% of the labor forcein the U.S. was working on farms.And that's no longer true.Right?

反正他们拦不住。这项技术是个精灵（genie），一旦放出来……（我不觉得是这样。历史上是有过的，比如砸纺织机，确实短暂地拦住过那场革命。）只是很短暂，顶多拖慢一点。（还有别的技术。）可我们现在照样有纺织机啊。（但确实有别的技术。）我们没拦住它。（有些技术是靠拒发专利之类的手段——政治上太难推，就变得没法商业化了。）我的看法是：100 年前美国有 50% 的劳动力在农场干活，现在早就不是了，对吧？

[10:35]

And now it's like...Which is great.That's a great thing.Now it's 2%.I don't like working on a farm.We don't have 48% unemployment.Absolutely.Yeah.Thank God.We don't have 48% unemployment.And we all have food.Right?

现在只剩 2% 了——这是好事啊，我可不想在农场干活。而我们并没有 48% 的失业率。（绝对没有。）谢天谢地，我们没有 48% 的失业率，而且大家都有饭吃，对吧？

[10:44]

But the question is, what is the speed of transition at?The speed of transition is the problem.Yeah.That's the question.That one took...It's a derivative.That one took, like, 60 or 70 years.Yeah.Okay.This one is going to be much faster.So I have good news for you guys.Have you ever worked at a Fortune 500 company?

但问题是：这个转变的速度有多快？转变的速度才是问题所在。这是个导数问题。上一次（农业那波）花了大概六七十年，而这一次会快得多。不过我有个好消息要告诉你们：你们有谁在世界 500 强公司干过吗？

[10:56]

Those things are so stupid.Nobody has.Oh, yes.I worked at Microsoft for like two years.No, no.I mean, like, nobody there works.Yes, exactly.It's already all make work jobs anyway, right?Yes.It's like, how much of the population is working?

那些地方蠢得要命。（没人干过。）我干过，我在 Microsoft 待过大概两年。（不不，我是说，那里根本没人真在干活。）对，正是如此。那本来就全是些『造出来给人干的活』（make-work jobs）。所以问题是：到底有多少人口真在工作？

[11:08]

And then how much of that is working for the private sector?Right.And then how much of that is not working in some bullshit job anyway?So Graber was right.Yeah.Graber was always right.Very few people are actually working.I mean, I'm working more because of AI.I got six Codex agents running all the time.I'm working the hardest I've ever worked because of AI.And so I just kind of think people are going to work more.The productivity is way higher.Well, some people won't.The leverage is way higher.But that's all true until AI can totally and completely replace you, which I don't seearound the corner, but I'm more open-minded about it now.I don't know how you do that.I used to think that was an impossibility.I used to be bearish on that.Now I'm like, you know, I got to pay attention.That's exactly my philosophy.It's like, you have two company A, company B. You have the same technology.Everyone has access.Like, you know, hopefully everyone has access.Well, that's a key.So are we going to, is it going to be open source?

其中又有多少是在为私营部门工作？再去掉那些其实是在做『狗屁工作』（bullshit job）的，还剩多少？所以 Graeber（《狗屁工作》作者）是对的——他一直都是对的，真正在干活的人非常少。（可我因为 AI 反而干得更多了：我有六个 Codex agent 一直在跑，我这辈子从没这么拼过。）所以我倒觉得人们会工作得更多，生产率高得多、杠杆也大得多。（有些人不会。）——但这些都成立，直到 AI 能彻底、完全地取代你为止；我没觉得那一天近在眼前，不过我现在比以前开放多了。（我不知道那怎么可能。）我以前也觉得那不可能，以前我对这事是看空的，现在我觉得：得留点神了。（这正是我的哲学。）就好比 A 公司、B 公司用同一套技术，人人都能拿到——但愿人人都能拿到。（这才是关键：那它会是开源的吗？）

[11:54]

Right.Or is it going to be closed?That would be great.Because right now, if they keep scaremongering, some of that might be, well, now you neednational approval.We just saw Sam Altman said that Codex 5.6 will be rolled out slowly to partners approvedby the U.S. government first.Oh, my God.So he's already cooperating because he doesn't want to get shut down like Anthropoc did.Yes.And so it's already starting.And so then a small number of people, it becomes a defense department thing.Then it's like, who has access?

还是闭源的？如果是闭源那可就『妙』了。因为现在如果他们继续散布恐慌，某些东西就可能变成『得先拿到国家批准』。我们刚看到 Sam Altman 说，Codex 5.6 会先小范围、逐步地发放给经美国政府批准的合作伙伴。（我的天。）所以他已经在配合了，因为他不想像 Anthropic 那样被『关停』。（对。）这已经开始了。于是它落到一小撮人手里，变成国防部的事，然后就是：谁有权限使用？

[12:16]

Who has control?You know, they have cyber weapons.Maybe they get to hack the systems first.Maybe they get to control them.Maybe they get mass surveillance that we don't for our own good.Right.That's how it always starts.So to me, it is definitely much scarier that a small group of people have control overAI than everybody has it.But everybody having it is also scary because of, well, I mean, there's the printer bioweaponthing, but the bioprinter labs are pretty locked down to begin with.I'm not sure how real that is.Good.He was like, good, ship it.And in fact, it's now coming out, which many of us already knew and been saying for yearsthat the funder and creators of the coronavirus, COVID-19, were basically the US governmentand the Chinese government working in cahoots.Right.It was basically Fauci and crew in North Carolina.So the joke I tweeted was like, designed by NIH in North Carolina, assembled in China.Right.Because it's like the old Apple line, designed by Apple in California, assembled in China.But that's basically what happened with the coronavirus.They were doing gain of function research to build a vaccine.They were being total cowboys.This is crazy stuff.I mean, anyway, I'm not saying AI is necessarily going to lead to bioweapons because to theextent that AI would lead to bioweapons right now, you know, some smart biohacker studentcould already do it.Right.So it's not like-Guns don't kill people.People kill people.It is democratizing a certain capability.I mean, game theory wins out in the end one way or the other.Game theory on this is not good.When mutually shared destruction is available to every individual, the game theory on it isnot good.My point of view is the problems are already out there.They're already as bad as you're saying that they're going to be.Like, the problems are already out there.And like, AI is the way.And just technology progress in general is the solution to all those problems.Gary can't even afford his open claw that's like reading his email right now.How is he going to pay for the nuclear bomb claw that-I'm good.I can pay for it.I just want everyone to pay for it.No, no, no.Exactly.I want everyone to be able to pay for it.It's like $100,000 a year, basically.It's totally worth it.At the same time-If you hire a person, you can afford your open claw.People have gotten so, I don't know if it's AI psychosis or what it is, that at this point,I usually just want to be like-AI anxiety.AI anxiety.I just want to be like, can I just talk to your AI, please?

谁有控制权？他们手里有网络武器，也许能抢先黑进系统、抢先掌控它，也许能搞我们享受不到的大规模监控——还美其名曰『为你好』。对，事情向来就是这么开始的。所以对我来说，一小撮人掌控 AI，绝对比人人都有它更可怕。当然人人都有也吓人——有『打印机造生物武器』那套说法，不过生物打印实验室本来就管得很死，我不确定那有多真实。（很好，那就发布吧。）而且现在越来越多的证据表明——很多人早就知道、也说了好多年——新冠病毒（COVID-19）的出资方和缔造者，基本上是美国政府和中国政府暗中合作搞出来的。基本就是 Fauci 那帮人在北卡罗来纳州。我发的推特段子是：『由 NIH 在北卡罗来纳州设计，中国组装。』——就像那句老苹果广告词『由 Apple 在加州设计，中国组装』。新冠差不多就是这么回事：他们在做功能增益（gain-of-function）研究、想造疫苗，结果玩得像牛仔一样野。这太疯狂了。总之，我不是说 AI 就一定会导致生物武器——因为真要那样，现在随便一个聪明的生物黑客学生就已经能干了。（枪不杀人，是人杀人。）它只是在把某种能力『民主化』。博弈论最终总会分出结果，而这件事的博弈论并不乐观：当『同归于尽的能力』落到每一个人手里，它的博弈论就非常糟糕。我的观点是：这些问题本来就已经存在了，已经跟你说的那么糟了；而 AI，以及技术进步本身，才是解决所有这些问题的出路。（Garry 现在连那个正在读他邮件的 OpenClaw 都快付不起了，他还怎么付得起那个『核弹级』的 claw？）我付得起，我只是希望人人都付得起。不不，正是——我希望每个人都能付得起。也就一年 10 万美元嘛，完全值。（可与此同时——）现在人们变得，我不知道是 AI 精神病还是什么，弄到这地步，我经常只想说：（AI 焦虑。）对，AI 焦虑。我只想说：能不能让我直接跟你的 AI 聊，拜托？

[14:25]

Like, just, you know, where do I email your AI?I don't need to interact with you because people are-Like, you can't even text someone and get an answer from them anymore.So many people that are writing emails and writing documents of AI, like on X, I seeall these posts that are clearly written by AI.I'm like, why should I read it?

就是那种——我该往哪儿给你的 AI 发邮件？我根本不需要跟你本人打交道，因为现在的人啊——你甚至没法给谁发条短信还能得到回复了。这么多人在用 AI 写邮件、写文档，我在 X 上看到一堆明显是 AI 写的帖子，我就想：我凭什么要读它？

[14:39]

I want my AI to read it.I think that's so clearly-If AI is going to write it, AI should read it.That's so clearly a lower class signal.Exactly.That's like ridiculous.When I get emails that are clearly written in AI, I'm like, dude, that's the last thing.Well, you can write it better with AI because what you need is a multi-stage voicedate.You write with AI, don't you?

我想让我的 AI 去读它。这不是明摆着嘛——如果是 AI 写的，那就该由 AI 来读。这明显是一种『低段位』的信号。（正是，简直荒唐。）每次我收到明显是 AI 写的邮件，我就想：老兄，这可是最不该……（可你用 AI 能写得更好啊，你需要的是一套多阶段的『语音口述』流程。）你写东西也用 AI 的，对吧？

[14:53]

I sure do, actually.I don't know.Hey, you caught me.That's why I keep jumping to this defense.I hate AI writing.It's too verbose.It's too verbose.It's too clinical.But you can fix it.We have super intelligence.Why?

我确实是这样，真的。我也说不清。嘿，被你抓到了。所以我才一直急着替自己辩护。我讨厌 AI 写的东西。太啰嗦了。太啰嗦了。太生硬、太冷冰冰。可你明明能把它改好啊。我们都有超级智能了。为什么不呢？

[15:05]

Just write.What's that?The value of actually reading something and realizing it's written by a person is so much.Yes, fair enough.I feel like such a jackass when I'm reading something and I realize it's AI.And by the way, if you're not writing your own stuff, how are you going to talk?

自己动手写就好了。你说什么？读到一段文字、意识到它出自一个真人之手，那种价值太珍贵了。对，说得没错。当我读着读着发现是 AI 写的，我就觉得自己像个傻瓜。而且顺便说一句，如果你连自己的东西都不写，那你还怎么学会说话？

[15:14]

You're going to lose your ability to speak well because good writing and good speakingare the output of good thinking.Yes.And if you're not practicing that muscle-What if I just read a lot?Yeah, I suppose you can do a lot of taste by reading.My counter argument would be, you know how sometimes you actually really need a person.I mean, I don't think this replaces people, but I can have now a high bandwidth conversationwith a relatively smart AI model all of the time.And you should do that.You should talk to the AI model all you want.But I'm saying when you write something that's meant to be consumed by other humans, if that'swritten by an AI, that's a disservice to the other human.You're wasting their time.Everything the AI wrote should instead be compressed down by you or you should really take the timeto make the point as succinctly as possible to the other person to respect their time.Otherwise, their AI is going to end up reading your AI and neither of you are in the loop.You're not even involved.This is one reason I enjoy Twitter less lately.Yeah.It's because so many replies are AI generated.I think this is virtue signal.I think in the future, people will just have really, really impressive skill files thatactually extract the style and the diction.And the thing that people underestimate that I've learned with some of this stuff is thatyou can just have more and more eval.You're saying you have an eval harness, right?

你会慢慢丧失好好说话的能力，因为好的写作和好的口才，都是好思考的产物。没错。如果你不练这块肌肉——那要是我只是大量阅读呢？嗯，我想光靠阅读也能养出不少品味。但我的反驳是，你知道有时候你是真的需要一个真人的。我是说，我不觉得这能取代真人，但现在我随时都能跟一个相对聪明的 AI 模型来一场高带宽的对话。你也该这么做。你想跟 AI 模型聊多少都行。但我想说的是，当你写一段本来是给别人看的东西，如果那是 AI 写的，那就是对读者的不尊重。你在浪费他们的时间。AI 写的一切，都应该由你亲手压缩提炼；或者你真该花时间，把观点用最精炼的方式讲给对方，以此尊重他们的时间。否则最后就变成：对方的 AI 读你的 AI，你俩谁都不在场。你根本没参与进来。这也是我最近越来越不喜欢 Twitter 的原因之一。对。因为太多回复都是 AI 生成的。我觉得这就是一种道德作秀。我觉得未来，人们会有非常非常厉害的 skill 文件，能真正提取出你的风格和用词。而人们低估的一点——这是我玩这些东西学到的——就是你可以做越来越多的 eval。你是说你有一套 eval harness，对吧？

[16:27]

Like you can have multiple levels.You can have cross modal evals.And when you have enough, like sort of, I mean, it's crosstalk.But if you do enough of it, like you can actually improve the skill file to a point whereit's indistinguishable.And like, this is without the next level of model.Like imagine Fable 7, Fable 8.Like you will not be able to tell.I guarantee you in like less than nine months, like this whole AI writing conversation willfucking go away.I think I'm still a little better defended and higher up on that hill than you guys becauseI write very, very short.And AIs are very bad at summarizing.But also very bad at distilling.Good writing is novelty.Yeah.Like it's like unexpected and anything that is guessing a next token from a regressioncan't do it.Yeah.It's not totally true.It won't do something original.If you like co-write with an AI or use it as an editor to bounce ideas off, almost alwaysit'll have like this one short, it's always like a short turn of phrase where you're like,oh, that's good.Yeah.And it got that from somewhere else on the internet, but the internet's a big place.Actually, I programmed a really crazy thing in Gbrain.It's the retrieval thing I made.Uh, it's actually basically a collider of different vector spaces.And so, I mean, and this is the kind, what I mean by like boil the ocean, like these arethings you can do now.Like it has my whole corpus of like everything I've ever said, all my emails, all my slacks,all my text messages, my DMS, everything.And then, so I actually have like, uh, like 400,000 markdown files on like literally anythingI've ever thought or read at this point.And then, uh, I have this thing called LSD mode, which is a lateral, um, shit.What's the S?

你可以做很多层。你可以做跨模态的 eval。当你积累得足够多——我是说，虽然有点串台——但只要你做得足够多，你真的能把 skill 文件优化到难以分辨的地步。而且这还是在没有下一代模型的前提下。想象一下 Fable 7、Fable 8，到时候你根本分辨不出来。我跟你保证，不到九个月，整个「AI 写作」的争论就会他妈的烟消云散。我觉得我在这座山头上比你们几个守得更稳、站得更高，因为我写得非常非常短。而 AI 特别不擅长做总结。也特别不擅长提炼。好的写作靠的是新意。对。就是那种出人意料的东西，而任何靠回归去猜下一个 token 的东西都做不到。嗯。这也不完全对。它做不出真正原创的东西。但如果你跟 AI 一起写，或者拿它当编辑来碰想法，它几乎每次都会蹦出那么一句——总是一个简短的措辞，让你觉得「哦，这句不错」。对。它是从互联网别处学来的，但互联网太大了。其实我在 Gbrain 里编了个特别疯狂的东西，就是我做的那个检索系统。呃，它本质上是一台不同向量空间的对撞机。所以，我说的「把海水烧干」就是这个意思，这些都是你现在就能做到的事。它装着我全部的语料——我说过的每句话、所有邮件、所有 Slack、所有短信、所有私信，全都在里面。所以我现在大概有 40 万个 markdown 文件，涵盖我这辈子想过或读过的几乎任何东西。然后，我做了个叫 LSD mode 的东西，它是一个 lateral（横向）……呃，妈的，那个 S 是什么来着？

[18:08]

Less as an A.Stochastic.Sorry.Yeah.Lateral stochastic drift.I forgot.I mean, I just wanted to call it LSD mode.Yeah, yeah.I know.I was trying to remember the chemical thing.Composition.Yeah.Yeah.Uh, but basically it's, you know, take every possible vector space that are, is not pointedin the same direction and then cross them and then do, um, a ranking, a re-ranking acrossall of these ideas.And you can actually more like, it's called brainstorm LSD mode and it just like findsbangers.And, you know, sometimes I'm just, if I just am bored, sometimes I'll just be like, giveme some LSD bangers.And it will literally give me like 10 or 20 ideas that like, you know, three or four differentfrontier models have actually re-ranked and said like, actually, if you look at this andwe cross-reference against like EXA, actually, this is actually a really good idea.And I'm like, this is finding ideas at like a ridiculous amount of scale.So, I mean, I get it, but I'm also calling bullshit because I think that like, this isjust the classic.And I think this is actually a really fun thing that is useful.It's, um, you know, I think it's like a Paul Bukitism, like the way to, you know, buildthe future is just like live in the future and then work backwards.Right.Right.And so like, it is inevitable that like, rather than fucking complain about AI, like it'shelpful.Like it scores points.Like I see it on X, like people love to like fucking score points on me.Okay.So this is a problem.It's like, I'm fucking inured to it.Like, I don't give a fuck if someone comes to me and is like, fuck your AI writing becauseit's like, you know what, my AI writing is so fucking good.I don't even give a shit.No, I read it.You know, like I'm alive.I'm a human being still.Right.But, um, anyway, that's how, that's, you know, welcome to my TED talk.No, I just, I just think it like.I definitely agree.Like to predict the future, you just live in the future.Right.Right.That's the best way to learn.And everyone should be AI maxing.Yes.Just to figure it out.Otherwise you're going to have no sense of where it's going.I think like today's virtue signal is, I mean, I think that it's not helpful to people.Like there are people who are going to listen to this, who want to be rich and they're goingto be like, oh no, like I shouldn't use AI to write.And it's like, oh, I see.Okay.Yeah.So, I mean, I agree with you that like human beings and like what you really think mattersa lot and you should write.Yeah.And then for me, that writing process is like actually workshopping in telegram with myopen clock.I would say for me, I don't use AI to write.Well, that's not true.I used to write code.So maybe I am contradicting myself.But code is like, it's a utilitarian.It's like, it's not meant for, code is meant to be consumed by another computer.It's not meant to be consumed by a human.So if I want to create something that's meant to be consumed by a computer, I will use a computer.Yeah.But if it's meant to be consumed by a human, then I want to understand the thing.I want to, you know, ruminate, marinate within it.And then I want to respect the reader or listener's time.Yes.And give them that insight nugget.Now, the AI is used for brainstorming.That part I like.I do use AIs to brainstorm.I use them like, give me synonyms for this.Give me other ideas associated with this.Pull in tangential information.You know, I argue with it sometimes, especially because a lot of them are on guardrails.But I don't use it to actually create the output that I expect a human to sit there and read.That just seems to me like discontinuity, right?

S 代表 stochastic（随机）。抱歉。对。Lateral stochastic drift（横向随机漂移）。我一时忘了。反正我就是想管它叫 LSD mode。对对，我懂。我刚才是想记起那个化学名。化学成分嘛。对，对。呃，但基本上它就是——把所有指向不同方向的向量空间都拿来，让它们互相对撞，然后对所有这些想法做一次排序、一次重排。你其实更像是——它叫 brainstorm LSD mode，专门帮你挖出爆款点子。有时候我要是无聊了，就会说，给我来点 LSD 爆款。它真的会给我十个、二十个点子，全是三四个不同的前沿模型重排过、并且断定「其实你仔细看，我们跟 EXA 交叉验证过，这确实是个好点子」的。我就想，这是在以一种离谱的规模挖掘想法啊。所以，我懂你的意思，但我也要唱个反调，因为我觉得这就是那套经典戏码。而我又觉得这其实是件很好玩、也很有用的事。呃，我觉得这有点像 Paul Buchheit 的那句话——建造未来的方法，就是活在未来里，然后倒推回来。对。对。所以这是必然的——与其他妈的抱怨 AI，不如说它其实很有帮助。抱怨 AI 这事能替人挣面子。我在 X 上就看到，大家特别喜欢拿我来刷存在感。好吧。这是个问题。我已经他妈的麻木了。要是有人跑来跟我说「去你的 AI 写作」，我根本不在乎，因为你知道吗，我的 AI 写作实在太他妈好了。我压根不在意。不，我会读的。你知道，我还活着，我还是个大活人。对。但，呃，反正就是这样，欢迎来听我的 TED 演讲。不，我只是觉得它——我完全同意。要预测未来，你就活在未来里。对。对。这是最好的学习方式。每个人都该把 AI 用到极致。对。就为了搞明白它。否则你根本感觉不到它在往哪走。我觉得今天的道德作秀——我是说，我觉得这对人没好处。会有一些听这期节目的人，他们想变有钱，然后他们会想「糟了，我不该用 AI 来写」。然后就是「哦，原来如此」。所以，我同意你说的，人本身、你真正的想法非常重要，你应该动手写。对。而对我来说，那个写作过程其实就是在 Telegram 里跟我的 open clock 反复打磨。要我说，我自己不用 AI 来写。呃，这么说也不对。我以前用它写代码。所以也许我是在自相矛盾。但代码是很实用主义的东西。代码本来就是给另一台计算机看的，不是给人看的。所以如果我想造一个给计算机用的东西，我会用计算机。对。但如果它是给人看的，那我就想真正理解这个东西。我想反复琢磨、浸泡在里面。然后我想尊重读者或听众的时间。对。把那颗洞见的金块递给他们。至于 AI，我用它来头脑风暴。这部分我喜欢。我确实会用 AI 做头脑风暴。我会让它「给我这个词的同义词」「给我跟这个相关的其他想法」「帮我拉点沾边的信息」。有时候我还会跟它争，尤其因为很多模型都套着护栏。但我不会真的用它来生成那种、我期待一个人坐下来认真读的成品。那在我看来就是一种断裂，对吧？

[21:25]

It's kind of unfair.It's a quality thing.I mean, I agree with you.Like, out of box, the quality is bad.Yeah.But you can have like 20, 28 level stuff.I mean, like, I think the pattern that I'm trying to like, you know, tell people is that like,yes, like AI writing out of the box sucks right now.Like, if you rely on it out of the box, it sucks.But if you have an eval harness, if you've actually gone to like build a big enough corpus,if you actually like, you know, done cross modal eval and like built, you know, a bank.Like, I have a banger engine.I have a banger gate.But I want to see some banger essays from you, man.Because I feel like, and even if they were from you.Oh, I have like eight essays right now with like a million views.Yeah, they wouldn't even be good enough because you are the president of Y Combinator, right?

这有点不公平。这是个质量问题。我是说，我同意你。开箱即用的质量确实差。对。但你可以做到 20、28 那种水准的东西。我想跟大家讲的模式是——没错，AI 写作现在开箱即用确实很烂。你要是直接拿来用，就是很烂。但如果你有一套 eval harness，如果你真的去搭了一个足够大的语料库，如果你真的做了跨模态 eval、建了一个素材库。我有一台爆款引擎。我有一道爆款闸门。但老兄，我想看看你写的爆款文章。因为我觉得，就算真是你写的——哦，我现在手上就有大概八篇上百万阅读的文章。对，但那还不够好，因为你可是 Y Combinator 的总裁，对吧？

[22:07]

I want to see someone who's just like out there posting, right?Using this system.Oh, I think there are a lot of people.Rise up, make an Anon account, disconnect it from yourself entirely,then come back and throw it in my face when you get to 300,000 followers.Okay.Because I'd be so curious if you could do that.You could absolutely.That would prove it.You could absolutely.You could because there are 300,000 people out there who will fall for it.But if you get that, then at least that's something, right?

我想看到的是那种普普通通、就在网上发帖的人，对吧？用这套系统。哦，我觉得这样的人多的是。你去啊，注册一个匿名账号，跟你自己完全切割干净，等你涨到 30 万粉丝，再回来甩我一脸。好啊。因为我真的很好奇你能不能做到。你绝对能做到。那样就能证明了。你绝对可以。你能做到，因为外面就是有 30 万人会上当。但如果你真做到了，那至少能说明点问题，对吧？

[22:29]

But if you pick up PG or me or someone, you know, who's like smart and insightfuland like and wants like high quality content and knows what that looks like,you're not going to get those people.I've never been more creative.And I think something happened probably just in the past few months,especially with Codex.I think I'm like I'm way ahead of the curve on Codex.Like I can, I'm running my computer from home on my phone here in a way that you can't actuallydo without modifying Codex in some ways.And like the amount of creative productivity I have is way higher than it ever was.I think like it's six months ago, 12 months ago, the promise wasn't quite there.And so everyone was like, eh, this AI thing is not that useful.But I think it did an inflection point in the past few months.I think Cloud Code was a tipping point.Yeah.Cloud Code, yeah.In December 2025 was the tipping point.And you had to re-evaluate AI.AI has had a couple of tipping points.One was diffusion into images, you know, which kind of started it all when you first saw likestability and you're just like, well, okay, made a dog face and you can be like cats andguys on the moon.Yeah, it was wild when it first came out.And then probably instruct GPT and chat GPT.And then there was another jump with the reasoning models like 01 and 03.But I think Cloud Code was a big unlock because it just unlocked a series of practical usecases, whereas before it was like, okay, it's a better Google search or, you know, it'slike this saves me time like doing research or I can learn from this or I can even havea small conversation.But Cloud Code was a massive unlock.Well, the big thing here is like this is the worst it'll ever be.Oh, of course.Yeah, yeah.No, it's getting better.There are like so many cryptoisms that basically super apply.Yeah, yeah.You could have said that when Fable was released and you would have been wrong because theytook it away from us.Yes.Briefly.Briefly.It'll be back.Use cases.You couldn't tell the difference between Fable and Minimax.For almost any.So that brings me to a good topic.I do want to discuss it is open source versus closed source.Right.How good are the open source models?

但如果你挑的是 PG、我，或者某个聪明、有洞见、想要高质量内容、也知道好内容长什么样的人，那你是骗不过这些人的。我从没像现在这么有创造力。我觉得大概就在过去这几个月里发生了某种变化，尤其是有了 Codex。我觉得我在 Codex 这件事上远远领先。我现在能用手机从家里遥控我的电脑，而这种玩法你不对 Codex 做点改造是做不到的。我现在的创造性产出，比以往任何时候都高得多。我觉得六个月前、十二个月前，那种潜力还没完全兑现。所以大家都觉得「嗨，这 AI 也没那么有用」。但我觉得过去这几个月出现了一个拐点。我觉得 Claude Code 就是那个引爆点。对。Claude Code，没错。2025 年 12 月就是引爆点。你不得不重新评估 AI。AI 已经经历过好几个引爆点了。一个是扩散模型进入图像领域——你知道，这几乎是一切的开端，当你第一次看到 Stability，你就想「好吧」，它生成了一张狗脸，然后你还能让它画猫、画月球上的人。对，刚出来的时候太炸裂了。然后大概是 InstructGPT 和 ChatGPT。再然后是推理模型带来的又一次跃升，比如 o1 和 o3。但我觉得 Claude Code 是一次巨大的解锁，因为它一下子解锁了一系列实用场景；而在那之前，AI 无非就是「好吧，它是个更好的 Google 搜索」，或者「它帮我省了做调研的时间」「我能从中学到东西」「甚至能跟它简单聊几句」。但 Claude Code 是一次巨大的解锁。这里最关键的一点是——现在已经是它最差的时候了（往后只会更好）。哦，那当然。对对。是的，它只会越来越好。有太多「加密圈的名言」在这里同样特别适用。对对。Fable 刚发布时你也可以这么说，但你会错，因为他们把它从我们手里收走了。对。就一小段时间。一小段时间。它会回来的。说到用例——你几乎在任何场景下都分不出 Fable 和 Minimax 的区别。所以这就引出了一个好话题。我确实想聊聊开源与闭源。对。开源模型到底有多好？

[24:19]

So good.It's unbelievable.Okay.I haven't tried GLM 5 too yet.Yeah.Five times, five to 10 times less the cost.Yeah.Okay.They're definitely cheaper.When should we use them?Minimax is just mind-bendingly good.It needs a good harness though.Yeah.Yeah.It doesn't need a good harness, but if you've got one, it works really well.I was thinking today, I'd love to see like take the latest and greatest open source modeland they seem to have gone from a 12 month gap to a nine month to six months.And now some people are saying three months.I don't know if it's weeks.Some people are saying three months, whatever.It is what it is.It's very jagged, right?

好得不得了。简直难以置信。好吧。我还没试过 GLM 5.2。对。成本只有原来的五分之一，五到十分之一那么便宜。对。好吧。它们肯定更便宜。那我们该在什么时候用它们？Minimax 好到让人脑子转不过来。不过它需要一套好的 harness。对。对。倒也不是非得有好 harness，但你要是有一套，它跑起来是真的很棒。我今天在想，我特别想看看——拿最新最强的开源模型，他们的差距好像已经从 12 个月缩到 9 个月、再到 6 个月了。现在有些人说是 3 个月。我不知道是不是已经到几周了。有人说 3 个月，随便吧。就这样了。这个差距很参差不齐，对吧？

[24:53]

In some domains, they seem caught up.In some domains, they don't.But I would love to see someone take the latest and greatest open source model, put it insidea very good harness, give it all the tools, make it available through a beautiful downloaddesktop app and jailbreak it.Strip it of every time it pushes back at you for anything stupid, any tone policing, anyI don't want to research that.I'll just tell you the truth every time and just call it truth.ai.Like that's all it does.Like I just want to see that.I just want to see what happens.Because the one place where the frontier, I'll always pay for more intelligence becauseI'd rather be right more often than wrong because I just think you're high leveraged,right?

在有些领域，它们看起来已经追平了。在有些领域，还没有。但我特别想看到有人拿最新最强的开源模型，塞进一套非常好的 harness 里，把所有工具都配齐，通过一个漂亮的、可下载的桌面 App 提供出来，再把它越狱。每次它为了什么蠢事跟你唱反调、每次它对你的语气指指点点、每次它说「我不想帮你查那个」——统统给它剥掉。我每次都只告诉你真相，就管它叫 truth.ai。它就干这一件事。我就想看看这个。我就想看看会发生什么。因为在唯一一个地方，也就是前沿——我永远愿意为更高的智能付费，因为我宁可对得多、错得少，因为我觉得你是高杠杆的，对吧？

[25:27]

And obviously I can afford it.But maybe if I was doing some highly repetitive tasks, I wouldn't.But as a simple example, like if you're doing something where the AI is right, you haveone AI that's right.I saw this stat like 99.9% of time.And the other one's right 90% of time.Well, if they're recursively looping and you run them 100 times, the one that was right9% of time is not going to be right 13% of time.Whereas the one that was 99.9 will drop to 80 or 90, right?

而且我显然付得起。但也许要是我在做一些高度重复的任务，我就不会（为智能多花钱）。举个简单的例子，假设你在做一件 AI 能做对的事。有一个 AI 的正确率——我看到过这个数据——是 99.9%，另一个是 90%。那么，如果让它们递归循环、连跑 100 次，原本 90% 正确的那个，最后正确的次数会低得可怜；而原本 99.9% 的那个，则会掉到 80% 或 90%，对吧？

[25:48]

So intelligence doesn't matter on the margin.Now those numbers are obviously chosen 99.9 versus 90.It's not 99 versus 90.You notice it's 100X more, 100X lower error rate.It's like errors can accumulate.Errors compound.That's the point.Errors compound.Anyway, going back, I will pay for more intelligence.But, and I'm not that cost sensitive on matters of judgment because you're always applyingleverage and your time is also very valuable.But what I am curious about is the frontier models do annoy me by constantly trying to railroadme.You know, they're always like, well, I shouldn't be telling you that.How about I give you this other answer?

所以说，智能在边际上并不重要。当然，这些数字是我特意挑的——99.9% 对 90%，而不是 99% 对 90%。你注意到没有，那是 100 倍的差距、错误率低了 100 倍。就是说，错误会累积。错误会复利叠加。这才是关键。错误会复利。总之，说回来，我愿意为更高的智能付费。而且在涉及判断的事情上，我对成本没那么敏感，因为你总是在施加杠杆，而你的时间也非常宝贵。但我真正好奇的是，前沿模型有一点很烦我——它们老想把我往固定的轨道上赶。你知道，它们总是说「呃，我不该告诉你这个。要不我给你换个别的答案吧？」

[26:20]

Okay.The really messed up thing is like-How much of that are they doing?If you talk to it about like serious personal stuff.Yeah.And it also like always mess up UTC versus like Pacific time.And it thinks it's like 7 PM and it thinks it's like 5 AM.And I was like, hey, this is enough.It's time to go to sleep.That's why it does.And after a while, I was like, what is going on with you?

好。真正离谱的是——它们这么干的频率到底有多高？如果你跟它聊一些严肃的私人话题。对。而且它还老是搞混 UTC 和太平洋时间。它一会儿以为现在是晚上 7 点，一会儿又以为是凌晨 5 点。然后我就说，嘿，够了，该去睡觉了。它就是为了这个（才劝你睡）。过了一阵子我就想，你到底是怎么回事？

[26:40]

Like, it's literally 7 PM.And it's like, oh, sorry.Like, anthropic are like, Jeff, it's me.So, you know, if I'm, if you're talking about something serious and I think it's late,like-Right.But I would love like a completely jailbroken, best of breed, open source model.That'd be great.Local is fine.It doesn't have to be in a nice package.It doesn't have to be local actually.But what are you doing where the jailbreak is so important?

明明就是晚上 7 点。它却说「哦，抱歉」。就好像 Anthropic 在替它说话，「Jeff，是我啊」。所以你知道，如果你、如果你在聊一件严肃的事，而我（模型）又觉得已经很晚了，就——对。但我真的很想要一个完全越狱的、同类里最顶尖的开源模型。那就太好了。本地跑也行。不用包装得多漂亮。其实也不一定非得本地。但你到底在干什么，让越狱这件事这么要紧？

[27:01]

Because I'm like-Well, there are a couple of things where it does stop you or crotch you or try to correctyou.You see that, right?There are a whole bunch where you don't see it, where it's successfully keeping you awayfrom whatever you want to know.But you can always just switch to GLM 5.2 and it'll just do it.No, but like if you-It's a period of time of your life.Go ask if any question about race, gender, immigration-Oh, yeah.Mental health, doctor advice, legal advice.So I guess I don't see why you need a harness for this, right?

因为我——好吧，有那么几种情况，它确实会拦住你、卡住你，或者试图纠正你。你也看到了，对吧？还有一大堆情况是你根本察觉不到的，它悄悄地就把你挡在你想知道的东西之外了。但你随时可以切到 GLM 5.2，它就直接给你办了。不，但如果你——这是你人生中的一段时期啊。你去问问任何关于种族、性别、移民的问题——哦，对。心理健康、看病建议、法律建议。所以我其实不太明白，你为这个要一套 harness 干嘛，对吧？

[27:24]

Like for me, I'm using harness because I'm using agents because it's going out and doingthings.I have to say harness.I just want to lose this model.By harness, I just mean like I want it maxed out.I don't want any excuse for like why my open source model is even slightly worse than itcould be.I want it operating.I want the best open source model configured in the best way.And it just tells me the truth all the time.Because I used to go to Grok for that, but even Grok is a little neutered.Yeah.It's not like who's talked about this.Is he?

对我来说，我用 harness 是因为我在用 agent，它要跑出去真的干活。我不得不说 harness。我只是想把这个模型彻底放开。我说的 harness，意思就是我想把它拉满。我不想给任何借口，让我的开源模型哪怕比它本该有的水平差一丁点。我要它火力全开。我要最好的开源模型，用最好的方式配置好，然后它就一直只跟我讲真话。因为我以前是去找 Grok 干这个的，但连 Grok 现在都有点被阉割了。对。不是有谁聊过这个吗。是吧？

[27:51]

Okay.People pointed out like someone advised.He was like, oh, geez.And we wanted it to just be real.And it was giving neoliberal answers.Yeah.Right.I'm dying to talk about what you said the other day, but I also don't know if you wantto say it.What is it?

好吧。有人指出——像是有人给它出了主意。他当时就「哦，天哪」。我们本想让它就实话实说。结果它给的都是新自由主义那套答案。对。是啊。我特别想聊聊你前几天说的那件事，但我又不确定你愿不愿意在这儿说。是什么事？

[28:04]

The entire-Oh, where is the open source?Like everything's probably hacked.It's probably already hacked.Oh, no, no.I'll talk about it.Okay.So the question is like, where are the open source models catching up?

整个——哦，你是说开源在哪儿？就好比一切大概都已经被黑了。很可能早就被黑穿了。哦，不不。我可以聊这个。好。所以问题是——开源模型是在哪些地方追上来的？

[28:13]

Right.Right.And so there's a couple of different theories.And I'll just lay out the theories that I know and the evidence for them.But I don't have a point of view yet.I mean, it's possible that these are true.Yeah.So there's some of it that's obviously, you know, the Chinese are doing their own pre-training.They have their own compute.They have their own data sets.In fact, they can crawl more of the data sets because they're less bound by copyright laws.Right.So they can crawl YouTube and Reddit and all that stuff.And they train these models.That's like the base model.Right.Then there's like the accusations that Anthropic made, which is that they're distilling our models,which is they're taking, they're querying our models en masse, taking that datausing a train of models.I mean, if I were China, that's what I would do.For sure.And that's creating a jagged intelligence kind of model.And definitely some of that is going on.But I think it's rich of Anthropic to call that out when Anthropic crawled the open weband distilled the open web.That's all these AIs are.In fact, one platform that I think would be very popular if the government were to doit is to say, hey, you guys, since you trained on the open web, if you trained on open weband open data, you have to open your model after X months, after 12 months or something.Right.That seems to be like a pretty fair thing to say, especially like open AI.You have to be, otherwise you have to be actually open AI.Right.An Anthropic, you're the one talking doom and gloom and you're building God, but you'renot going to keep God on the leash for yourself, Dario.We don't want like a priest, you know, controlling God for the rest of us interpreting the Biblefor the rest of time.So I think that would be popular.Okay.So anyway, number one, they have their own pre-training, but they get the full corpusof the web.They don't have to obey any copyright laws.Right.Second, they are distilling the existing American models, which are starting to get more lockeddown.And a lot of this government nationalization control helps them be locked down.But, you know, there are data brokers online that are taking like CloudMax accounts andreselling tokens from that in an API format to end users of like 80% off because thoseplans are heavily subsidized for end users.They charge the enterprises.It's like $8,000 a month retail.And anyone who's like run like an exchange or a regulated online brokerage knows that KYCis a dime a dozen.People get back KYC all the time.They like print passports in Vietnam if you want.Right.So yeah, exactly.So I think the distillation is going to be hard to stop.But I think the biggest one to think about, and there's some genuine algorithmic breakthroughslike the DeepSeq paper.Oh yeah.DeepSeq R1.Amazing systems.Incredible systems people.Very, very smart people.And I would guess just by looking at the numbers of resumes and so on, that the majority ofmathematicians and researchers in AI today are Chinese.Like they produce more STEM graduates, more PhDs in the relevant fields, more Olympiadwinners than anybody else.Right.And then a lot of the US staff is Chinese.So in fact, that's why I tweeted like AI is our Chinese against their Chinese.Right.And these guys, and I don't think there's any conspiracy required.These guys are all just friends.You know, they live in the same dorms.They went to the same schools.They live in the same buildings and apartments.They hang out.Yeah.And they switch jobs like crazy.And not just the Chinese.You see everyone's hopping from lab to lab to lab and they're just taking data with them.Okay.So there's all of that going on.Right.But then on top of it, I don't think that the AI companies have the security profile ofa top US national secure compartment facility.Right.They do not know how to protect secrets.So of course they're getting hacked.Of course the weights are getting leaked.Right.And if you're Chinese or anybody else, you're not going to like just take the weights andrelease them back.You're going to use that to distill at high speed and train your own model.Right.Or to augment your model.So I think there's another whole system right there that's going on.And the last one, actually, I think the biggest one is none of those.I think no conspiracy is required for the following.If you look at where China is, they're like, okay, we're behind.They've always been behind in software.The US has been way ahead in software.And being ahead in software allows it to extract all the margins.Because for the longest time, VCs never invested in hardware.Why?

对。对。所以有那么几种不同的理论。我就把我知道的理论和支撑它们的证据摆出来。但我自己还没有定论。我是说，这些有可能都是真的。对。首先，有一部分显然是——中国人在做他们自己的预训练。他们有自己的算力，有自己的数据集。事实上，他们能爬到更多数据，因为他们受版权法的约束更少。对。所以他们能爬 YouTube、Reddit 之类的一切，然后拿这些去训练模型。这就是所谓的基座模型。对。然后就有了 Anthropic 提出的那个指控，说他们在蒸馏我们的模型——也就是大规模地查询我们的模型、把那些数据拿走，再拿去训练一串模型。我是说，我要是中国，我也会这么干。绝对会。这就造出一种「参差不齐的智能」模型。这种事肯定在发生。但我觉得 Anthropic 来指责这个挺讽刺的，因为 Anthropic 自己就爬了整个开放网络、蒸馏了整个开放网络。这些 AI 本质上都是这么来的。事实上，如果政府真要做一件事、我觉得会很受欢迎，那就是说：嘿，你们既然是在开放网络上训练的，如果你在开放网络和开放数据上训练，那你就得在 X 个月、比如 12 个月之后把你的模型开源。对。这听起来是件相当公平的事，尤其是对 OpenAI 来说。你必须开源，否则你就得真的名副其实地做「Open AI（开放的 AI）」。对。还有 Anthropic——你才是那个满口末日论、又在造上帝的人，可你不会打算把上帝拴在自己手里吧，Dario？我们可不想要一个「牧师」，替我们所有人掌控上帝、替我们永远解释《圣经》。所以我觉得这个提议会很受欢迎。好。总之，第一，他们有自己的预训练，但他们能拿到整个网络的全部语料，还不用遵守任何版权法。对。第二，他们在蒸馏现有的美国模型，而这些模型正开始变得越来越封闭。而很多这种政府国有化管控，反而帮着它们变得更封闭。但你知道，网上有一些数据掮客，他们搞来一堆 Claude Max 账号，再以 API 的形式把里面的 token 转卖给终端用户，打个八折——因为那些套餐对终端用户是被大幅补贴的。他们的钱是向企业收的，零售价大概是每月 8000 美元。任何开过交易所、或运营过受监管在线券商的人都知道，KYC（身份认证）多得跟白菜一样。人们成天绕过 KYC。你要愿意，他们能在越南给你印护照。对。所以对，没错。所以我觉得这种蒸馏很难禁得住。但我觉得最值得琢磨的一点——而且确实有一些真正的算法突破，比如 DeepSeek 那篇论文。哦对。DeepSeek R1。了不起的系统。做系统的人太厉害了。非常非常聪明的一群人。而且我猜，光看简历数量之类的就知道，如今 AI 领域大多数的数学家和研究员都是中国人。他们培养出的 STEM 毕业生、相关领域的博士、奥赛冠军，都比任何其他国家都多。对。而且美国这边很多员工也是中国人。所以事实上，这就是为什么我发推说：AI 之争是我们的中国人对他们的中国人。对。而这些人——我不觉得需要什么阴谋论。这些人就是一帮朋友。你知道，他们住同一个宿舍，上同样的学校，住同样的楼、同样的公寓，一起玩。对。而且他们跳槽跳得飞快。还不只是中国人。你看，每个人都在从一个实验室跳到另一个实验室，还顺手把数据一起带走。好。所以这一切都在发生。对。但在这之上，我不认为这些 AI 公司具备美国顶级国家保密隔离设施那种安全等级。对。他们根本不懂怎么保护机密。所以他们当然会被黑。权重当然会泄露。对。而如果你是中国人、或任何其他人，你不会傻到拿了权重再原样发回去。你会用它来高速蒸馏、训练你自己的模型。对。或者用来增强你自己的模型。所以我觉得这里还有一整套系统在运作。而最后一点，其实我觉得最大的一点，都不是上面这些。我觉得接下来这一点根本不需要任何阴谋论。你看中国的处境，他们会说，好，我们落后了。他们在软件上一直落后。美国在软件上一直遥遥领先。而在软件上领先，就能把所有利润都抽走。因为在很长一段时间里，VC 从不投硬件。为什么？

[31:56]

Because hardware is a commodity business.Right.And software had the network effects, the lock-in.Software's art, as Patrick Collison said, it locks you in.And so people want to invest in software.And so even in the US, whenever you get a hardware company funded, the VCs will always askyou, well, what's the software lock and what's the software piece?

因为硬件是大宗商品生意。对。而软件有网络效应、有锁定效应。正如 Patrick Collison 说的，软件是一门艺术，它能把你牢牢锁死。所以大家都想投软件。所以哪怕在美国，只要你有个硬件公司要融资，VC 总会问你：那你的软件锁定在哪儿？软件那部分是什么？

[32:13]

Right.And so this is why I used to say, yeah, hardware is the moat that buys you time to build a softwarecastle.Right.But guess what?Claude Code burned software down.Software was eating the world and AI ate software.So now software is commoditized.The moment you can specify it, an AI can one-shot it or two-shot it or get to it within a coupleof shots.Well, I see one from like 5% to about 25% hardware, 20% to 25% hardware batch to batch.But hardware is also commoditized by China.Any hardware I can make here, you can make in China more cheaply and more easily.The whole ecosystem is there.Shenzhen has like 3,000 manufacturers, this tiny little cable all in one little area.Thanks, Nixon, for normalizing relations with China and taking us off the dollar standardor the gold standard.And so bringing reshoring manufacturing back to the US is going to take a full generation.It should be done, no question.I'm a big fan of American manufacturing.But it's going to take a long time.In the meantime, China outproduces every one of the supply chains because they realize thereare scale economies in production.So they subsidize the thing upfront, then they become the global supplier.They drive everybody else out of business.And then they control the whole supply chain across these critical industries.So they own hardware.No one's going to beat China on hardware in the next decade.It'll ramp up.We'll fight on a bunch of fronts.But they own hardware.So they like the commoditization of software.So when software is getting commoditized, that helps China.So if you're the Chinese government, you're basically funding all these labs.You're doing kind of a public-private partnership where you're saying, don't worry, you don'thave to make money.You can be number two or number three.Just open it so the other labs can learn from you and we can catch up and keep up.And if the software is commoditized, then we win on the hardware.So the way I think about it today is that hardware is commoditized, but it's owned by Chinafor the most part.There are some few exceptions like SpaceX in the US.Software is commoditized.So what's not commoditized?

对。所以我以前常说，硬件是护城河，它替你争取时间去建造软件的城堡。对。但你猜怎么着？Claude Code 把软件给烧掉了。软件本来在吞噬世界，结果 AI 把软件给吞了。所以现在软件被商品化了。只要你能把需求说清楚，AI 一次、两次，或者几次之内就能把它做出来。嗯，我看到硬件占比大概从 5% 到 25% 不等，每批次在 20% 到 25% 之间浮动。但硬件同样被中国商品化了。任何我能在这里造的硬件，你在中国都能造得更便宜、更容易。整条生态都在那儿。深圳光一小块地方就有大约 3000 家制造商，就为这么一根小小的线缆。得谢谢尼克松（Nixon）跟中国建交，还让我们脱离了美元本位、或者说金本位。所以要把制造业回流到美国，得花整整一代人的时间。这件事该做，毫无疑问。我非常支持美国制造。但它需要很长时间。而在这期间，中国在每一条供应链上的产能都碾压所有人，因为他们明白生产是有规模经济的。所以他们先在前期补贴，然后成为全球供应商，把其他所有人挤出局。然后他们就掌控了这些关键行业的整条供应链。所以硬件是他们的。未来十年没人能在硬件上打败中国。这会慢慢起量。我们会在很多条战线上厮杀。但硬件是他们的。所以他们乐见软件被商品化。当软件被商品化，这对中国有利。所以如果你是中国政府，你基本上就是在给所有这些实验室掏钱。你搞的是一种公私合营——你跟他们说，别担心，你不用赚钱。你可以做老二、做老三。你只要开源，让别的实验室能向你学习，我们就能追上、跟住。而只要软件被商品化，我们就在硬件上赢。所以我今天的看法是：硬件被商品化了，但它基本上掌握在中国手里。有少数例外，比如美国的 SpaceX。软件也被商品化了。那么，还有什么没被商品化呢？

[33:57]

It's actually just AI research.Developing, working on AI itself is the new software engineering.The problem with that is the software used to be democratized.Everyone could compete.John Carmack and John Romero sitting there with a small team could code up Doom and Quakeand compete against EA and Activision and so on.Can't do that in AI.AI takes massive resources, huge GPU clusters, small numbers of researchers, massive data sets,proprietary data sets.So in that, and then now regulatory capture.So it seems like all the value, all the choke points are going into AI and that's controlledby a very small number of companies.And ironically, the only thing keeping us afloat is a Chinese government subsidizing the wholething on open source so that their hardware can stay competitive.Yeah.I mean, I think the...Plus they have the weights.No, I'm kidding.Well, they might.They might.Let's use that as...You can sneak about on one USB stick.Are you telling me that hasn't already been done?

答案其实就是 AI 研究本身。开发 AI、钻研 AI 本身，就是新的软件工程。问题在于，软件过去是被「民主化」的。人人都能竞争。John Carmack 和 John Romero 带一个小团队坐在那儿，就能敲出《Doom》和《Quake》，去跟 EA、Activision 这些巨头掰手腕。在 AI 里你做不到这个。AI 需要海量资源、庞大的 GPU 集群、少数几个研究员、海量数据集、专有数据集。在这基础上，现在又加了监管俘获。所以看起来所有的价值、所有的咽喉要道都在往 AI 里集中，而这被极少数几家公司掌控着。讽刺的是，唯一还让我们没沉下去的，就是中国政府在开源这件事上全盘补贴，好让他们的硬件保持竞争力。对。我是说，我觉得……再加上权重也在他们手上。不，我开玩笑的。不过他们可能真有。真有可能。我们就拿这个当……一根 U 盘就能偷偷带出来。你该不会以为这事还没发生过吧？

[34:50]

No, it's like you need physical access.Like you need one person on the inside.Like there's so many people.By the way, every male AI founder is walking around with a beautiful Chinese girlfriend.Have we talked about that?

不，这得有物理访问权限。你需要一个内部的人。而里面人那么多。顺便说一句，现在每个男性 AI 创始人身边都带着一个漂亮的中国女朋友。我们聊过这个没有？

[35:02]

Like no one talks about that.Go walk outside Silicon Valley, right?I think we all come from a world where advantages like had some durability and you could actuallyoperationalize your advantage against the rest of the world.But that's just like narrowing down in time.So, you know, you might have the smartest model, but what are you going to do with itin two weeks, in three, four weeks until somebody else has it too?

都没人聊这个。你去硅谷外面走一圈就知道了，对吧？我觉得我们都来自一个「优势有一定持久性」的世界——你真的可以把你的优势转化成对全世界的实际胜势。但现在这个（优势存续的）时间正在急剧收窄。所以你可能手握最聪明的模型，但在别人也拿到它之前的那两周、三四周里，你到底能拿它做什么？

[35:21]

So I think there's a different part of the game theory that we've not experienced before,which is the time contraction of any of these advantages.And I don't think it makes it harder to predict who's going to be in charge of anything because,like you said, a few months later, the smartest model is now open source.Guys, I'm so fucking worried.This is one of my tweets I'm prouder of a couple of years back.Like I said, soon everybody will have AI anxiety.Like we're all feeling it.We're all feeling AI anxious, right?

所以我觉得这里有一个我们过去从未经历过的博弈论层面——就是这些优势在时间上的急剧压缩。而我并不觉得这会让「预测谁将主宰一切」变得更难，因为正如你所说，几个月后，最聪明的模型就开源了。各位，我真是他妈的太焦虑了。这是我几年前发的、比较得意的一条推。我当时就说，很快每个人都会有 AI 焦虑。我们现在都感受到了。我们都在为 AI 焦虑，对吧？

[35:47]

It's incredibly...Oh, I'm jubilant.I am AI jubilant.But it's both, right?The anxiety in its good form is kind of FOMO-ish, which is like, look at all the great thingsI can do.Don't worry about me, guys.Yeah, but like everyone I know that's working at Frontier Labs, like all these things thatare researchers, they're like depressed.They are?

这实在太……哦，我可是欢天喜地。我对 AI 满心欢喜。但这两种情绪是并存的，对吧？焦虑好的那一面，有点像 FOMO（怕错过）——就是「看看我能做多少了不起的事」。别替我担心，各位。是啊，但我认识的每一个在前沿实验室工作的人，那些做研究的，一个个都很抑郁。真的吗？

[36:04]

Why?Yeah.They're stressed out because they think it's going to kill us all.That's because their culture is...Why would it kill us?The first ones I met in 2021 were saying to me that they were putting off..of having kids or thinking about not having kids because they were afraid for the kids'future.And I was like, what are you talking about?

为什么？对。他们压力很大，因为他们觉得这东西会把我们所有人都害死。那是因为他们的圈子文化……它为什么会害死我们？我在 2021 年最早遇到的那批人跟我说，他们在推迟要孩子，或者干脆考虑不要孩子了，因为他们替孩子的未来担忧。我当时就想，你们在说什么呢？

[36:16]

As opposed to every other time in human history when people were never afraid of their kids'future and so they didn't have any kids.Wait, so why is this not just going to result in Star Trek The Next Generation?

搞得好像人类历史上其他任何时候，人们都从不为孩子的未来担忧、所以那些年代就没人生孩子似的。等一下，那这一切为什么不会直接变成《星际迷航：下一代》那样的世界呢？

[36:27]

Like that's my hope.So...The bullshit part...Like at the end of scarcity it would be great.The bullshit part about Star Trek The Next Generation is that they have AGI clearly.Yes.But that AGI is somehow not running things.I mean, they couldn't conceive of it yet, right?

那就是我的期望。所以……那套扯淡的地方在于……匮乏终结之后，那会很美好。《星际迷航：下一代》里最扯淡的一点是，他们显然已经有了 AGI。对。但那个 AGI 却莫名其妙地没在掌管一切。我是说，（那年代的编剧）还没法设想到那一步，对吧？

[36:39]

They have the holodeck, but somehow people are still doing about their business.Weren't that good enough a writer?Not just living in the holodeck.There's plenty of science fiction that's, you know, bothered to see that idea furtherthrough.I mean, some large percentage of the San Francisco population has been one-shotted by fentanyl.You don't think the holodeck will take out the rest?

他们有全息甲板（holodeck），可人们不知怎么还在照常忙自己的正事。大概是编剧还不够高明吧？没让大家干脆一直住在全息甲板里。倒是有不少科幻作品，认真地把这个设想往下推演到底了。我是说，旧金山已经有相当大比例的人口被芬太尼（fentanyl）一击撂倒了。你难道觉得全息甲板不会把剩下的人也带走吗？

[36:55]

Right, right.They just don't show that on the show.Like Gene Roddenberry, like I'm sure he conceived of it and it would just be too much of a bummerto show like these giant holodeck, you know, I mean, yeah.You say this and you're living in Star Trek and you're mine and you're happy.That's true.You're like excited for it.And it's like, but yes, you recognize there's a ton of terrible things that are going to happen.Right.And you're just like, oh, it wasn't in Star Trek.Well, it's...You're blowing my dreams over here, Daniel.But to Gary's defense, like there's been so much shit.It's electricity.Like that was a big one, right?

对对，剧里根本不会拍那些。像 Gene Roddenberry（《星际迷航》创作者），我敢说他脑子里想过，但真要拍出那种巨型全息甲板（holodeck）之类的，太扫兴了，就没拍。（你这么说，是因为你活在《星际迷航》里、觉得那是你的世界、你很开心——你为它兴奋。）没错。（可你也承认，会有一大堆糟糕的事要发生。）对。（然后你就说：唉，那《星际迷航》里可没演。）你这是在浇灭我的梦想啊，Daniel。不过替 Garry 说句话——历史上折腾过的事多了去了，比如电力，那也是天大的一件事，对吧？

[37:27]

The cars, the railroads, the cars, these were big, big changes.You used to have to ride a horse, like the Pony Express.Do you remember that?That was insane.And people made it.I don't remember.But the more...Yeah.But the more...Me neither.I wasn't allowed.The more rapid...Yeah.The more rapid the change, the more influential the displaced, the more tumultuous it was.And so now we're having a very rapid change.We're displacing white collar.We're displacing government.We're displacing managers.We're displacing academics.We're displacing journalists.You know, I learn more from AI than I did in my college classes.Because I can sit there and I can have it meet me at my exact level of knowledge.Because they give that to me visually.Give me a graph about that.I can ask dumb questions.I don't understand that.Explain it again.Explain it again.Explain it a third different way.Explain it visually.Explain it audibly.Explain it with dancing elves.I don't care.Like just to keep going over until I get it.I could not do that in class.I can learn so much better.I just have to care.Yeah.Right.If I care, I can learn anything better than with any tutor, better than with any teacher.So all the people are obsolete.And they're going to lead a revolution.Well, no.Those people need to want more things.And if you want more things.Well, we want unlimited things.Yeah.Which is great.That's the good news.The good news is like everyone wants more things.Wants are unlimited.And then if you have this tool and you're not averse to using it, this is why I'm so againstlike, sorry to bring up the AI writing thing again.But like, I understand like that makes sense.Humans and like, there is a disrespectful aspect when you do bad writing.Right.But like that presupposes that like the AI can't write well.That's a fair point.Because right now the writing is voluminous.It's a giant, giant.Like it takes a little thing like prompt you give it.Well, okay.But here's the thing.Yeah.You're giving it a prompt, right?

汽车、铁路、又是汽车——这些都是翻天覆地的大变化。以前你得骑马，像 Pony Express（快马邮递），还记得吗？那太疯狂了，可人们都挺过来了。（我不记得，我也没赶上，没被允许。）但越是……对，变化越快、被取代的人越有影响力，动荡就越大。而现在我们正经历一场极快的变化：我们在取代白领、取代政府、取代管理者、取代学者、取代记者。我从 AI 那里学到的，比大学课堂里还多——因为我能让它精确地对齐我当前的知识水平，还能给我可视化地讲：给我画张图，我可以问蠢问题、说『这我没懂，再讲一遍、换第三种方式讲、用图讲、用声音讲、用跳舞的小精灵讲我都不介意』，就这么反复讲到我懂为止。这在课堂上根本做不到。我能学得好太多了，我只要『在乎』就行。对，只要我在乎，我用 AI 学任何东西都比任何家教、任何老师都强。所以所有这些人都过时了——然后他们就要领导一场革命了。（不，那些人得想要更多东西。而只要你想要更多东西——）我们想要的是无限的东西。（对，这是好消息：好消息就是人人都想要更多，欲望是无限的。）然后只要你有这个工具、又不排斥用它——这也是为什么我特别反对，抱歉又扯回 AI 写作那事——我理解那个逻辑：人与人之间，写得糟糕确实有种不尊重的意味。（对。）但那前提是假设 AI 写不好啊。（这话在理。）因为现在 AI 写出来的东西是海量的、又臭又长——你就给它一个小小的 prompt……（好吧，但问题是：你给它一个 prompt，对吧？）

[39:08]

Yes.Why don't you just send me the prompt?Why do I need to read all the garbage?Well, no.Because it has to draw on like all of human experience.No, no.Just give me the prompt.Yeah.I have an AI on this side.Just send me the prompt.I mean, that might happen.I'll send it to my AI.That is how it happens.I have my codex put together a markdown file that I send to my developers that they cangive to their Claude code.Yeah.To go execute something that I want.Yeah.It's like, this is the prompt.Go do it.Don't talk to me.Have your agent talk to my agent.That's literally what it is.My thing will just file an issue on your GitHub repo.And then.Yeah.What I want to do is I want to get on the Zoom.And I'm on the Zoom having a conversation.And then 10 minutes in, I know this is not for me.Yeah.And then I just hit a button.Yeah.And the agent takes over.Agent Naval keeps talking.And I go after the next one.Yeah.I swipe right on you.Yeah.But then the joke's on me because you were never there to begin with.I think it's time for politics.Politics.Politics.Sounds like it's been politics most of the time.Maybe let's start with.Fear mongering.Politics doesn't matter because of AI.You know.That's not true.What are you talking about?

对。那你干嘛不直接把那个 prompt 发给我？我干嘛要读那一大堆废话？（不不，因为它得调动全人类的经验。）不不，直接把 prompt 给我。我这边也有个 AI，你把 prompt 发给我，我转给我的 AI 就行。（这事真可能发生。）——这就是它发生的方式。我让我的 Codex 整理出一个 markdown 文件，发给我的开发者，他们再喂给他们的 Claude Code 去执行我想要的东西。就是说：这是 prompt，去干吧，别来烦我。（让你的 agent 跟我的 agent 谈。）字面意思就是这样。我的东西会直接在你的 GitHub 仓库上提个 issue，然后……对。我真正想做的是：上了 Zoom、聊着聊着，10 分钟后我心想『这事不适合我』，我就按个按钮，agent 接管，『Agent Naval』继续跟你聊，我去搞下一个。（对，我对你右滑。）但笑话最后落我头上——因为你压根从一开始就没在场。我觉得该聊政治了。（政治。）（听起来一直在聊政治。）也许我们从……（散布恐慌开始。）因为 AI，政治根本不重要了。（这不对吧，你在说什么？）

[40:07]

Everything's going to end.Nothing that's like politically relevant right now is going to be relevant in three years.I will say that I wish I could vote for an AI.People losing their job.I wish I could vote for an AI to represent us.But who's AI?

一切都要终结了。现在任何在政治上重要的东西，三年后都不再重要。我倒想说：我真希望能投票给一个 AI。（人们在失业。）我希望能选一个 AI 来代表我们。（可那是谁的 AI？）

[40:19]

Dari will make it.I don't know.Chinese open source AI.Somebody can make it.Dario's AI.Sam's AI.Elon's AI.They can put the weights out for us to take a look out.You can read the weights.You can look at the weights.I mean, not personally.That weight looks off to me.But like a fully transparent AI that is running for office.There's this thing.They had this thing, a whole field called mechanistic interpretability, where they're trying to figureout how the AI thinks.And they've made very little progress.Oh, sure.I'm not saying we're going to know.But I'm just saying you have access to the same thing with the AI.By the way, it's an anachronism.Our training data is in natural language and in computer languages that humans can readand write.And so the AIs are trained on that.But in the future, the AIs will develop their own languages.They'll communicate with each other in some kind of a binary high-speed protocol.They'll almost be talking compiled code or maybe even something completely different that wecan't even imagine, which will be more efficient for communication.It's only humans who demand the legibility, but the AIs don't really have it.And part of what makes AI systems so much better than human-coded expert systems is that theydon't have to be legible.So they can find correlations and patterns that we cannot articulate.In fact, even most of our knowledge is inarticulatable.Most of your feelings are inarticulatable.You don't even know half the time what you're feeling or thinking.You're not even like fully hallucinating.The other half you're hallucinating.Let's talk about our feelings.Yeah.Yeah.Yeah.Let's do that.We're experts in that.One place where we are experts.GLM 5.2.I'll vote for GLM 5.2 for office.You're Chinese overlords.Hey, it's my favorite distillation of fable.I really do appreciate how the Chinese are making all the open source models out there.It's kind of pathetic that the U.S. just is not-The global economy will appreciate it.CDS is probably the global leader in video, right?

让 Dario 来造吧。或者中国的开源 AI，总有人能造。Dario 的 AI、Sam 的 AI、Elon 的 AI——他们可以把权重（weights）公开出来给我们看。你可以读权重、看权重（当然不是我本人去看，那权重我怎么看都别扭）——但至少是一个完全透明、去竞选公职的 AI。有个东西叫『机制可解释性』（mechanistic interpretability），一整个领域在试图搞清 AI 是怎么思考的，但进展非常小。（当然，我不是说我们真能搞懂。）我只是说，你至少能接触到 AI 的这同一套东西。顺便说，这其实是个时代错位：我们的训练数据是自然语言、加上人类能读写的编程语言，所以 AI 是拿这些训出来的；但将来 AI 会发展出自己的语言，用某种高速的二进制协议互相交流，几乎是在讲编译后的代码，甚至是我们完全想象不到、对交流更高效的东西。只有人类才要求『可读性』，AI 其实并不需要。而 AI 系统之所以远胜过人类手写的专家系统，恰恰是因为它们不必可读——它们能找到我们根本说不清的相关性和模式。事实上，我们绝大多数知识本身就是说不清的，你大多数感受也是说不清的，你一半时间根本不知道自己在感受什么、在想什么。（你连完全在『幻觉』都算不上。）另一半时间你在幻觉。（我们来聊聊感受吧。）对对对，来聊，我们是这方面的专家——这是我们唯一称得上专家的领域。（GLM 5.2。）我愿意投 GLM 5.2 去竞选。（你的中国霸主。）嘿，它是我最爱的 Fable 蒸馏版。我真心欣赏中国人做出了这么多开源模型，而美国就是不——（全球经济会感激他们的。）CDS 大概是全球视频模型的领头羊，对吧？

[41:53]

It's the best video model.So in some places, actually, so that's an interesting point about open source.In some cases now, open source is number one.And at least historically, like if you look at things like Linux or other open source projects,once something open source kind of gets in the lead, it rarely surrenders it.And that's because an ecosystem springs up around it where everyone starts plugging into,especially in enterprise use cases.So right now, Claude and Codex are like slightly ahead of GLM, fable and maybe a lot ahead of GLM.But if something like a GLM were ever to surpass them, it's not clear they'd get a lead back.I mean, because how can you rationalize your limited resources at a company going into something that open source is winning at?

它是最好的视频模型。所以在某些地方——这就引出关于开源的一个有意思的点——如今在有些赛道，开源已经是第一了。而且从历史上看，像 Linux 或别的开源项目，一旦某个开源的东西领先了，它就很少再把位置让出去。因为会有一整个生态围着它长起来，大家都往里插接口，尤其在企业场景里。所以现在 Claude 和 Codex 稍微领先 GLM、Fable 一点，甚至领先 GLM 很多；但万一哪天像 GLM 这样的东西反超了它们，就不好说它们还能不能把领先夺回来了——因为你怎么向公司交代：拿宝贵的资源去砸一个开源已经在赢的方向？

[42:36]

You have to do something with your resources.And failing to catch up to open source with your precious resources is not a good use of them.You'd have a clock.Like so if you're open air and tropic, you would have a certain amount of revenue and money and cash in the bank and investor backing.And you would have to use that to sort of jump meaningfully ahead of the open source and take the lead again before that runs out.So that's because if you look at the AI race, two years ago, people thought that Google and Elon and Meta were also competitors.Yeah, there were five kings.There were five kings.There's two kings now.There's two kings.And why is that?

你总得拿资源干点事，而拿你宝贵的资源去追赶开源、还追不上，这资源就用得太亏了。你会有个倒计时：假如你是 OpenAI 和 Anthropic，你手里有一定的营收、现金和投资人支持，你就得赶在这些钱烧完之前，用它们把自己实打实地拉到开源前面、重新夺回领先。这就是为什么——你看这场 AI 竞赛，两年前大家还以为 Google、Elon 和 Meta 也是竞争者。（对，当时有五个王。）五个王，现在只剩两个王了。（两个王。）为什么会这样？

[43:10]

Well, it's not just because those guys have the best models.Open AI and tropic are the only ones that are A, making revenue off of their models directly.And so they're not subsidizing, cross subsidizing from the business.They're not running out of cash.They're actually pouring the cash back in.And the second is because they have very active user bases.And now a lot of the improvement is coming through reinforcement learning.They're getting the guidance and the trajectories to improve the models.And so they're just pulling away.And maybe Elon, because he took SpaceX public and he's got that war chest, he gets maybe one more bite at the Apple.Plus he's got the data centers in space.So, you know, don't bet against Elon.Google, I think, has lost it.That's quite sad.It's really amazing what happened.I was always like bearish on Google.I was tweeting some pretty anti-Google management stuff back in the day where I was like,I will know they're serious about AI.Gemini 3.1 Pro was like the moment.That was as good as it was going to get.It was.No, but it wasn't.But it wasn't.I thought like 2.5 Pro was better.It wasn't good enough.Yeah.It was.They never even got the basic app working.I remember I tried to upgrade to pay them $20 a month.It was a nightmare.This huge runaround.It's crazy.No, no.You need to be part of Google One.No, no.You need Google Workspace account.They were first to have million token usable contacts.Yeah.Briefly.Very cool.And then every time, even today, the iOS app, which I know it's not their platform, buteveryone else seems to figure it out.You put a query into Gemini, you background, it always loses the connection.It just drops it.Yeah.All the others will just run in the background.How hard is that?

这不只是因为那两家的模型最好。OpenAI 和 Anthropic 是唯二真正靠模型本身直接赚钱的——它们不是在用别的业务交叉补贴，不会烧光现金，反而是把现金再投回去。第二，是因为它们有非常活跃的用户群；而现在很多进步是通过强化学习来的，用户给了它们改进模型所需的引导和轨迹，于是它们越拉越远。也许 Elon 还行，因为他让 SpaceX 上了市、有那笔战争储备金，也许还能再咬苹果一口，加上他在太空里有数据中心——所以别跟 Elon 对着干。Google 我觉得是掉队了，这挺可惜的，发生的这一切真是让人吃惊。我一直看空 Google，当年还发过不少怼 Google 管理层的推：Gemini 3.1 Pro 就是那个『见真章』的时刻，它顶多也就那样了。（是。）（不，不止。）（不过还不够好。）我觉得 2.5 Pro 更好。（还不够好。）对。他们连最基本的 app 都没搞利索。我记得我想升级、每月付他们 20 美元，简直是噩梦，被踢皮球踢来踢去，太离谱了。（不不，你得加入 Google One。）（不，你得有 Google Workspace 账户。）他们最早做到了百万 token 可用上下文。（对，就那么短暂一阵。）挺酷的。可每次——哪怕到今天——那个 iOS app（我知道 iOS 不是他家平台，但别人都搞定了）：你往 Gemini 里输个查询、切到后台，它总是断连、直接把任务丢了。（对。）别家都能在后台照跑。这能有多难？

[44:31]

It's 2026.Run a background app, guys.GRPC.You just need a little bit more Sparta over there.Meta might be too much Sparta right now.I think it's too late for me.I think they have to kill 5,000 Google PMs.That's the only way.It's PM slop.The whole company's PM slop at this point, right?

都 2026 年了，做个能后台运行的 app 吧，各位。GRPC。你们那边就是得再来点『斯巴达』（狠劲）。Meta 现在可能『斯巴达』过头了。我觉得对我来说已经太晚了。他们只有一条路：干掉 5000 个 Google PM（产品经理）。全是 PM 糊出来的烂活（PM slop）——整个公司现在都是 PM slop，对吧？

[44:46]

Anywhere you go, it's just tons of products overlapping.It's a huge mess.You can't figure a way through the maze.It's an incentive problem now.All the best people want stock in Open Iron Anthropic, or they want to be sitting alongsidetheir best colleagues in Open Iron Anthropic, or they want the data set from Open Iron Anthropic,and they certainly don't want to be tied down by all the PMs and all the legacy BS.And they should have replaced Google Search with Google AI a long time ago.I don't even look at the results anymore.I just look at the Google AI.So here's an alternative vision, which is like Google today could, I mean, they probablyhave all the transcripts of all the meetings, or if they don't, they should have it.And they have all your emails.Right.And then they probably need to use Anthropic Fable 5 or whatever, because their own modelsare not quite good enough.But you could diarize all of these.You could have a markdown file for every single employee at Google, and you could compute a score.Which you'll cleverly trademarked and patented.That's right.Google running through GStack, of course.I use Google.No, anyway.Google Omni is the best for...Google, Gary has a price.It's 7%.It's standard.It's just standard.It's normal.Yeah, it's just normal.But seriously, you could solve this.Like, if someone was a true, like, really wanted to fix this, like, all of these thingsare just systems problems, right?

你走到哪儿，都是一堆功能重叠的产品，一团乱麻，你在这迷宫里根本找不到出路。现在这是个激励机制的问题：最好的人才都想要 OpenAI 或 Anthropic 的股票、想跟 OpenAI 或 Anthropic 里最强的同事并肩、想要 OpenAI 或 Anthropic 的数据集，他们绝不想被一堆 PM 和一堆历史包袱捆住。他们早就该用 Google AI 取代 Google 搜索了——我现在连搜索结果都不看，只看 Google AI。所以这里有个替代设想：Google 今天完全可以——他们大概握着所有会议的转录（就算没有也该有），还握着你所有的邮件。然后他们大概得用 Anthropic 的 Fable 5 之类（因为他们自家模型还不够好），把这些全都做说话人分离（diarize），给 Google 每一个员工建一个 markdown 文件，再算出一个分数。（这分数你会很机智地注册商标、申请专利。）没错，当然是跑在 GStack 上的 Google。（我用 Google。）不管怎样——（Google Omni 是最好的……）（Google，Garry 是有价的：7%，标准价，就是行价、正常价。）对，就是正常价。但说真的，这问题是能解的：只要有人真心想修，这些事说白了都只是系统工程问题，对吧？

[46:00]

Yeah.You could have a score.But it would take Larry and Sergey walking in there, and literally, it's like that scenefrom Entourage when Ari Gold walks back in the office and is shooting people.But only with their open claws.And live ammunition.I'm serious.Like, they don't need to be there physically.Like, you know, one of the things I learned from Pedro at Brex, like, that's how he runsBrex.Like, you know, he has total information awareness on, like, every team.And one of the things, like, the thing that got me...Wait, didn't they lose to Ramp?

对，你可以给每个人算个分。但那得靠 Larry 和 Sergey（Google 两位创始人）亲自走进去——简直就像《明星伙伴》（Entourage）里 Ari Gold 冲回办公室开枪那一幕，只不过他们是拿着各自的 OpenClaw、还是实弹。我是认真的。他们其实不用亲自在场——我从 Brex 的 Pedro 那儿学到的一点就是他这么管 Brex：他对每个团队都有『全面信息掌控』（total information awareness）。而真正让我震撼的一点是……（等等，他们不是输给 Ramp 了吗？）

[46:27]

I mean, hey.I'm kidding, I'm kidding, I'm kidding.The Ramp guys come in.Look, man, I don't have any skin in the game on either of those, okay?We're accepting sponsorships from whoever calls in first.Whoever pays us more, we'll talk about it.Capital One or, anyway, I mean, no, like, I mean, he really did, like, instrument theCEO brain, like, for him.Like, it's a personal claw that knows, like, the KPIs and literally what all of his directstalk about in those meetings.And it's significant because...Are they spying on them all the time?

我开玩笑的、开玩笑的。（Ramp 的人进来了。）听着，这两家我都没有任何利益关系，好吧？谁先打电话来我们就接谁的赞助，谁给的多我们就聊谁——Capital One 也行。总之，他是真的把『CEO 的大脑』给仪表化了：一个私人的 claw，掌握着各项 KPI，甚至字面上知道他所有直属下级在那些会上都聊了些什么。这很重要，因为——（他们是在时时刻刻监视这些人吗？）

[46:57]

I mean, yes.No, I mean, the thing is, like, I mean, they're not the only company.Like, I've talked to other CEOs.They don't publicly talk about this.Behind closed door, this is, like, the max thing that you could do is, like, concentratethe power in your CEO or co-founding team so you have actual awareness of what the fuckis going on.Well, the reality is the team should be much smaller.Right.And most of your employees are GPUs working in the data center.I mean, of course, if you can have 10 fewer, like, a tenth of your employees right nowand just make them the 100x guys, then that's massive.So, it's easier said than done.Like, most people, most CEOs do not want to fire 50% or 80% of the people.I'm serious.You should be a lot more small.I've been on boards where it's like, bro, like, we're fucked and we need to fire 80%of the people.And you have a shot.Like, you literally have-But you don't know which 80%.Like, tens of millions of dollars in the bank still.Part of the problem is you don't know which 80%.Well, you can.You're working through managers.Now you can.Now you can.You can.You can get the visibility.And that's not a little thing.That's a big fucking thing.Like, you can have total information.Like, okay, like, this is a big thing to me.Right.I think that the future is essentially that, like, you need to ride the fucking AGI.Like, you actually need to.Like, the most significant thing to me is, like, the one million token context window isa big deal.It's like, I mean, I talk about it as, like, three Harry Potter books.But guess what?

算是吧。不，我是说——他们不是唯一这么干的公司，我跟别的 CEO 也聊过，他们不会公开讲，但关起门来，你能做到的『极致』就是把权力集中到 CEO 或联合创始团队手里，让你真正掌握到底他妈发生了什么。而现实是：团队本该小得多。对，你大多数员工应该是数据中心里干活的 GPU。当然，如果你现在能把员工砍到只剩十分之一、还都是『百倍战士』，那威力就巨大。可这说起来容易做起来难——大多数 CEO 不愿意裁掉 50% 或 80% 的人。我是认真的：你们该小得多。我在董事会待过，那种局面就是：兄弟，我们要完蛋了，得裁掉 80% 的人，你才有活路——你银行里明明还躺着几千万美元。（问题之一是你不知道该裁哪 80%。）你能知道。（你是隔着管理层在管。）现在你能了，现在你就能拿到这种可见度了——这可不是小事，这是天大的事：你能掌握全部信息。这对我来说是个大事。我觉得未来本质上就是：你必须骑上这头 AGI，你是真的必须。对我最震撼的是：一百万 token 的上下文窗口是件大事，我常把它形容成三本《哈利波特》。但你猜怎么着？

[48:17]

Like, a human being can keep in their brain.Fucking seven things plus or minus three.Right, right, right.Like, we are little monkey brains.Yeah, yeah.Like, we talk about, like, super intelligence is, like, in the future.Fuck that.Super intelligence right now, like, an agent working for you can literally make, like,a thousand pages, like, printed, valuable, useful, credible, legible to you.God, let's not talk about AI writing again.No, no, no.It's not a writing thing.No, it's an ideas thing.It's like, you can understand what the fuck is happening in your org.Like, every CEO in the world, like, really should know the shit.Don't have your AI say, tie into all the systems in the entire company, evaluate everything,and tell me which 20% to fire right now.I mean, you could do that.And you should.That's, like, the easiest thing in the world right now.What do you think about that?

一个人的大脑能同时装下的，也就他妈七加减三样东西，对吧对吧。我们是小小的猴子脑。人们老说超级智能是『未来』的事——去他的，超级智能现在就有了：一个替你干活的 agent，字面上能给你产出一千页、打印出来的、有价值、有用、可信、你读得懂的东西。（天呐，别又聊 AI 写作了。）不不不，这不是写作的事，是想法（ideas）的事——是让你搞清楚你的组织里到底他妈在发生什么。全世界每个 CEO 都真该掌握这些。你完全可以让你的 AI 接进公司所有系统、评估一切，然后告诉你现在该裁掉哪 20%。你能做到，你也该做。这在当下简直是世上最容易的事。（你觉得呢？）

[49:03]

But no one will do it.No one's doing it.Because we haven't talked about the project at Meta.Plus, it's not perfect.Do you know anything about this?It's, like, only what I read.It's, like, I mean, half of the smart people I knew at Meta have left.Yeah.Yeah.I mean, I'm sorry.Like, I really like Meta.You know, like, you guys are great.But, like, it's also, like, what the fuck?

但没人会去做，没人在做。因为我们还没聊 Meta 那个项目呢。而且它也不完美。（这事你了解吗？）我也就是读到的那点。我认识的 Meta 里一半的聪明人都走了。（对，对。）抱歉，我真的很喜欢 Meta，你们很棒，但这也……到底他妈在搞什么？

[49:21]

I think it's become a sweatshop.Well, I don't think you can build a culture or morale by just buying people.Especially all different, from all different stripes and places and ramming them together.Then you'd have no idea how many are actually missionaries, how many are missionaries.Culture is a real thing that takes time to evolve and grow.Meta, of all places, needs to do what I just described.Which is, like, should, you know, someone has to actually know what is actually happening per person, per team.And, I mean, you can do this now, right?

我觉得它变成了血汗工厂。你没法靠『买人』来建立文化或士气，尤其是把三教九流、来自各处的人硬塞到一起，那你根本不知道其中有多少是真正的『传教士』（有使命感的人）。文化是个真实的东西，需要时间去演化、生长。而 Meta 恰恰最该做我刚说的那件事：得有人真正掌握每个人、每个团队到底在干什么。这事你现在就能做，对吧？

[49:50]

But you also have to tune it.Like, there has to be a rating.It's, like, what is actually a good employee?What is their behavior?What are their characteristics?Like, you can actually tune your culture right now.It's taking the whole engineering team and turning them into data labelers.It's just, like, it's, like, dropping a nuclear bomb in the place.So, that seems indiscriminate to me, right?

但你还得去调它：得有一套评分，界定到底什么才算好员工、他的行为是什么、特征是什么。你现在真的能调你的文化。（可他们那做法是把整个工程团队变成数据标注员，简直像往公司里扔了颗核弹。）所以那在我看来太一刀切了，对吧？

[50:09]

Like, why would you do that?Especially because you have this incredibly fine-grained thing.Like, you have a prompt.You have a skill file.You have, like, a bunch of markdown.Like, you can actually tune it to exactly the output that you're on.I think their org is just too large.It's the same problem as Google.When you have a large org, you do things that actually make sense in certain domains.Then it gets leaked to the press.It gets sensationalized.Oh, they're turning everyone to data laborers.It's probably because you have too many people.You have too many cooks in the kitchen.But it's still, like, way less than what you can fit in three Harry Potter books.Yeah.Right?

你干嘛要那样？尤其你手里有这么细粒度的工具：你有 prompt、有 skill 文件、有一堆 markdown，你完全可以把它精调到你想要的确切产出。我觉得他们的组织就是太大了，跟 Google 一个毛病：组织一大，你在某些具体领域做的、本来合理的事，一旦泄露给媒体、被添油加醋，就成了『他们把所有人变成数据劳工』。这多半就是因为你人太多、厨房里厨子太多。但这些信息量还是远远塞得进三本《哈利波特》里。对吧？

[50:38]

Like, I mean, I'm sorry to, like, bring that up.I think that this is, like, a big deal.Most millennial example you could give.Sorry.I don't know.Read another book.Yes.Well, I think the big scary question is, will there still be a startup ecosystem?

抱歉我又提那个（哈利波特）。我觉得这是件大事。（你能举的最『千禧一代』的例子了。）抱歉。（换本书读读吧。）好吧。我觉得那个又大又吓人的问题是：还会不会有创业生态？

[50:49]

Because on the one hand, you're shrinking the size of the firm.People are more leveraged.That should mean more startups.Small teams can do more.Great.Fantastic.We're definitely seeing that.I mean, people can get to 100 mil ARR, like, not raise a bee.But doing what?

因为一方面，公司规模在缩小、人的杠杆在变大，这本该意味着更多创业公司——小团队能干更多的事。太好了，我们确实看到了这一点：有人能做到一亿美元 ARR（年度经常性收入），甚至一轮融资都不用融。（但靠做什么呢？）

[51:01]

Selling software.Yeah.And software is getting completely commoditized.Like, all these legal guys, like Harvey and whatever, like, it's a good test.Like, wouldn't Claude just basic mythos or fable just do a better job than something trainedspecifically on legal, especially with tool use?

卖软件。而软件正在被彻底商品化。像那些做法律的公司，Harvey 之类——这是个很好的试金石：Claude 用基础的 Mythos 或 Fable、再配上工具调用（tool use），难道不会比一个专门在法律数据上训练的东西干得更好吗？

[51:14]

It depends because it depends on whether or not the frontier labs stay open for their bestmodels or do they close them?I think anyone who pays them enough, they'll find a way to be open.So if a law firm will pay them the best result.That's not a given.It is possible for Anthropic to say, like, actually, fable six.I wouldn't bet my business, they're not going to enter my vertical.I think if they can enter your vertical without a custom model, they'll do it.And I think one of the bitter lessons here is that the general models beat the specializedones.Yes.They don't have to enter it.It's just going to be able to use it.Yeah, exactly.To upstate that market, not because the open AI legal app, right?

这要看情况——取决于前沿实验室会不会把它们最好的模型保持开放，还是把它们关起来。我觉得只要谁给的钱够多，它们总会想办法开放：一家律所若肯为最好的结果买单……（那可不一定。）Anthropic 完全可能说：其实 Fable 6 我自己上。我可不敢拿我的生意去赌他们不会杀进我的垂直领域。我觉得只要他们不用定制模型就能进你的领域，他们就会进。而这里一个『苦涩的教训』（the bitter lesson）就是：通用模型会打败专用模型。（对，他们甚至不用亲自进场，直接就能用。）正是，去把那个市场端了——还不是因为出了个『OpenAI 法律 app』，对吧？

[51:53]

Right.It's just because it's so much.It's just like, why would I use something special?Why would I pay $5,000 a month for some other SaaS company?I think that, like, 2027 will be the year of the AI harness war.Is it going to be Codex?

对，就单纯因为它太强了。你会想：我干嘛还要用某个专门的东西？我干嘛要每月付 5000 美元给另一家 SaaS 公司？我觉得 2027 年会是『AI harness 大战』之年。会是 Codex 吗？

[52:06]

Is it going to be Co-Works slash Cloud Code?Is it going to be OpenClaw, Hermes?Is it going to, like, what is it?You know, who's, like, what harnesses are people going to use day to day for everything?

会是 Cowork / Claude Code 吗？会是 OpenClaw、Hermes 吗？到底会是哪个？就是说——大家日常什么都用的那个 harness，最后会花落谁家？

[52:17]

Yeah.Right now, so a lot of Silicon Valley after 2007 got to exist because you had two mobilephone providers.You had Apple and you had Google.Google, if you had only one, if Android didn't exist in an ecosystem, it would have been amuch tougher position for startups.You would have been facing a kind of a true monopolist at the end of that chain or thetop of that chain.Right now, you at least have two horses fighting each other, right?

对。你看，2007 年之后硅谷有那么多公司能存在，很大程度上是因为你有两家手机平台商：Apple 和 Google。假如只有一家、假如 Android 这个生态不存在，创业公司的处境会艰难得多——你会在那条链的末端（或顶端）直面一个真正的垄断者。而现在你至少有两匹马在互相厮杀，对吧？

[52:40]

So they kind of force each other to behave a little bit.If it boils down to just one, it's a very bad situation for startups.Well, then they'd nationalize it.And then, you know, maybe we could get the government to-Do they really nationalize it?

所以它们多少会逼着彼此收敛一点。要是最后只剩一家，对创业公司就是非常糟糕的局面。（那到时就把它国有化呗。）然后也许我们能让政府——（他们真会国有化吗？）

[52:50]

They didn't nationalize Google.They didn't nationalize, you know, Apple.This is way bigger.It depends on how bad it gets.I mean, like search, it's a good utility.Like there are lots of reasons to not-And if you nationalize it, it might slow down and lose because now the DMV is running itand China wins.Right.It's not like the Chinese are asleep.There's an ASML machine supposedly somewhere in China now.You hear about that?

他们没把 Google 国有化，也没把 Apple 国有化。但这个（AI）要大得多，就看会烂到什么地步。搜索算是个不错的公用事业，有很多理由别去动它。而且你一旦国有化，它可能就会变慢、然后输掉——因为现在是『车管所』（DMV，官僚低效的代名词）在运营它，结果中国赢了。对，中国人可没在睡觉：据说现在中国境内某处已经有一台 ASML 的机器了，你听说了吗？

[53:09]

Howard Litton and Claude-Oh, I heard that.They're ASML and was all mad because an ASML machine somehow leaked into China.That's the lithography machine.There's only 184 of them.That's our cell phones escaping.Yeah.And then on top of it, they're developing, you know, their Huawei chips.They're working on their seven nanometer, five nanometer, whatever processes.They're building their fabs.They're working on it.They're not dumb.I mean, you've got 1.4 billion people there.You've got a lot of STEM PhDs.They have a lot of national pride.They can redirect capital.I'm not saying it matches up to innovation in a free market economy, but the U.S. is aheadbecause we have created this very thin layer where we take the absolute best and brightestfrom all over the world and we incent them like crazy to compete with each other.And then we collect the rewards from that.And Silicon Valley actually disproportionately collects the rewards and California collectsthe rewards.And it's not really spread out to the rest of the country as much.The rest of the country is scared shitless now.But that's it.That's the thin blue line standing between, you know, the so-called free world and theChinese competition.Now, I'll actually take a radical point of view.I don't think we're in competition with China.I don't see the competition.Taiwan, I mean, every Taiwanese person I talk to doesn't want to fight.You know, the rich people in Taiwan are busy dodging the draft by taking their kids outof the country.You know, there's some rule in Taiwan that like-Mandatory service.It's mandatory conscription.But if your kid has spent, I don't know the numbers, but something like three months ayear for the last five years out of the country, then they're not eligible for the draft anymoreor no more mandatory conscription.So every rich person is always taking their kids for three months a year out of the year.They have like a standard system doing it.They're not interested in fighting.The second largest political party there, the Kuomintang, is very pro-China.So it's already very compromised.They are-Most people in Taiwan are thinking they're going to go like Hong Kong.They're going to get rich in the process, they're going to get a gateway to China, andthen one or two generations later, they'll assimilate.And how the heck is the US going to defend Taiwan?

Howard Lutnick（美国商务部长）跟 Claude……（哦我听说了。）ASML 气坏了，因为不知怎么一台 ASML 的机器流进了中国——那是光刻机，全世界总共才 184 台。（那是我们的手机在『越狱』外流。）对。再加上他们在搞自己的华为芯片，在攻 7 纳米、5 纳米之类的工艺，在建自己的晶圆厂（fab），一直在做，他们不傻。你想想那边有 14 亿人口、一大批 STEM 博士、强烈的民族自豪感，还能集中调配资本。我不是说这就比得上自由市场经济里的创新，但美国之所以领先，是因为我们造出了一层非常薄的东西：我们从全世界吸走最顶尖、最聪明的人，疯狂地激励他们互相竞争，然后我们坐收其成——而且硅谷、加州拿走了不成比例的一大块回报，并没有多少扩散到全国其他地方，其他地方现在都吓得半死。但就这么一层——这就是横在所谓自由世界和中国竞争之间的『那条细细的蓝线』。不过我要说个激进的观点：我不觉得我们在跟中国竞争，我没看到这种竞争。台湾——我接触过的每一个台湾人都不想打仗；台湾的有钱人正忙着把孩子送出国来逃兵役。台湾有条规矩：兵役是强制征召的，但如果你孩子过去五年里每年有大约三个月（具体数字我不确定）待在国外，就不再符合征召条件了。所以每个有钱人都常年把孩子每年送出国三个月，他们有一整套标准操作。他们没兴趣打仗。台湾第二大政党国民党是非常亲中的，所以局面已经很『被渗透』了。台湾大多数人其实觉得自己会走香港那条路：在这过程中发财、拿到通往中国的门户，一两代人之后就同化了。那美国到底要怎么保卫台湾？

[55:02]

Like aircraft carriers are dead.Land-based missiles from China can take out aircraft carriers, drones.DJI is the largest defense contractor in the world.It's like us trying to like-It's like China trying to defend the Florida Keys from us.Yeah.You know, it's ridiculous.The whole concept is ridiculous.So-And you saw what we just went through with Iran.How do you think we'd-We'd run out of missiles in like seven days.We don't have the manufacturing base.Now maybe Andrel and Saronic and others get there and build all of that up.But that's a long ways off.So I mean, I think the Chinese know that.And I think the Taiwanese know that.And I think everyone's trying to save face.Well, like over 10, 20 years, Taiwan slowly reunites with China.And outside of that, we don't really have a beef with China.You know, okay, Japan, whatever.You know, North Korea has got a South Korea issue.But it's not in our backyard.One of the big problems the U.S. is facing is that we've been in a low growth, high inflation environment for a while.Now maybe growth is picking back up, but so is inflation.And people are fighting over the spoils.There will be democratic socialists coming up, you know, the billionaires tax, the wealth tax, the socialists trying to seize the means of production.You know, Mom Donnie doing rings around the democratic establishment.And they're talking about the eradication of Western civilization.People, you got to put food on the table.You got to make people feel good.You got to make people feel like they're not going to go jobless.You got to feed them.And picking a war with China just seems like the stupidest thing to do.Now, we should have the appropriate tariffs against them if they have tariffs against us.If they subsidize their businesses, especially network effect or scale economies to compete with us, then we should also have our own barriers.Because you got to protect your local industry so it doesn't get snuffed out.So you at least have a fighting chance.Because these things do have economies of scale.It's not as simple as David Ricardo said, where it's just like, oh, yeah, I'm selling bananas and you're selling watermelons and we can just trade.No, because the guy who gets better at doing something gets really good at doing it and then can do it very, very cheaply.Or he gets a network effect around either an ecosystem or an actual lock-in where new users create value for existing users.So you can't have that simplistic of a mindset.You do have to play on an even playing field.You have to make sure the currencies are aligned, no one's subsidizing, et cetera, et cetera.But that said, there's no reason for us to go to war with China.There really isn't.But is Taiwan like it or is it like control the South China Sea?

航母已经是死物了。中国的陆基导弹、无人机能干掉航母。DJI 是全世界最大的国防承包商。这就好比我们想去——好比中国想从我们手里保卫佛罗里达群岛（Florida Keys），荒唐，整个概念都荒唐。你也看到我们刚在伊朗那趟的表现了：我们大概七天就会把导弹打光，我们没有那个制造业基础。也许 Anduril、Saronic 这些公司将来能补上、把这些都建起来，但那还远着呢。所以我觉得中国人心里有数，台湾人也有数，大家都在想办法保全面子。差不多就是：未来 10 到 20 年，台湾慢慢和中国重新统一；除此之外，我们跟中国其实没什么深仇。日本，随便吧；北朝鲜跟南韩有事，但那不在我们后院。美国面临的一大问题是：我们在低增长、高通胀的环境里待了一阵，也许现在增长回来了，可通胀也回来了，人们在争抢那点战利品。会冒出一批民主社会主义者——亿万富翁税、财富税，社会主义者想夺取生产资料；Mamdani（纽约政客）把民主党建制派耍得团团转，他们张口就是『铲除西方文明』。可老百姓得吃饭、得让人感觉良好、得让人觉得不会失业、得让人有饭吃。而在这种时候去跟中国挑起战争，简直是最蠢的事。当然，如果他们对我们加关税，我们就该对等加；如果他们补贴自家企业——尤其是有网络效应或规模经济、来跟我们竞争的——那我们也该有自己的壁垒，因为你得保护本土产业别被掐死、至少有一战之力。因为这些东西确实有规模经济，不是 David Ricardo 说的那么简单——不是『我卖香蕉你卖西瓜，咱俩换换就行』。不是的：因为某件事做得越来越好的人会变得极其擅长、然后能做得非常非常便宜；或者他会围绕一个生态、或某种真正的锁定效应建起网络效应，新用户又为老用户创造价值。所以你不能抱这么天真的心态，你确实得在公平的场子上竞争：得保证汇率对齐、没人搞补贴等等。但话虽如此，我们没有任何理由跟中国开战，真的没有。（可台湾算是这么回事、还是说重点是控制南海？）

[57:12]

It's like control trade routes.Why do we need to control the South China Sea?So like, OK, so the world breaks in half.The world's already broken into many pieces.I know.But like right now, our ships go through there and it's fine.Right.But like the world literally breaks in half.Why would they stop our ships from going through there if they're carrying, you know, trade goods back and forth?

重点是控制贸易航线。我们为什么非得控制南海？好，就算世界一分为二——世界早就碎成好多块了。（我知道。）可现在我们的船从那儿过、也没事啊。（对。）但假如世界真的一分为二——如果他们运的是来来往往的贸易货物，他们又何必拦我们的船？

[57:27]

Like, who knows why they might want to do that?We have a rule that's like, you know, the rule of the sea is that we can freely pass wherever we want.There is no we.We have a civil war going on internally in this country, in case you haven't noticed.It's nationalists versus communists.Like everyone is going more radical on both sides.I mean, it might be fine if the world breaks in half and China controls Asia.Right.And it's like China does.We get our ships through.Sometimes they charge us things and stuff like that, maybe.China does control most of East Asia.And to the extent that it does not, it is up to Japan and South Korea and others to look out for their interests.And they can ally with the U.S.But we're not going to get into a shooting war with China over interests in the South Pacific Sea.Oh, we would never do that.But we will sail a ship through there and they'll hit it.Why would they hit our ship going through there?

谁知道他们为什么会想那么干呢？我们有条规矩，就是所谓的海洋法则——我们可以自由通行到任何想去的地方。可根本就没有『我们』。这个国家内部正在打一场内战，你要是没注意到的话。是民族主义者对共产主义者，两边都在越来越激进。我是说，就算世界一分为二、中国控制亚洲，也许也没什么大不了。对吧。中国控制它的地盘，我们的船照样过得去。有时候他们可能收点过路费之类的，也许吧。中国确实控制着东亚大部分地区。至于它没控制到的地方，那就得靠日本、韩国这些国家自己去维护利益，他们可以跟美国结盟。但我们不会为了南海的那点利益跟中国真刀真枪打一仗。哦，我们绝不会那么干。可我们会派艘船从那儿开过去，然后他们就会打它。他们干嘛要打我们一艘路过的船呢？

[58:10]

If they run a ship through the Panama Canal.We're very good at getting our ships hit.No, no.But if they put a ship through the Panama Canal, why would we attack it?If it's a legal trade vessel carrying goods and services back and forth from the U.S.that we're trading on.I think we're not advised to do any of that to each other.Yeah.I think right now China actually cares way less about this stuff than we think.China is very internally focused.It's like the CCP is very internally focused.Exactly.Yeah.Yeah.They want to stay in charge of their empire.And, you know, I'm not a fan of communists or the Communist Party, although I think CCPresembles more of a capitalist fascist organization.The communists is a veneer.Right.Fascists in the sense of like it's like the nation uniting for a common cause.They are hyper-They're very good at making money.Yeah, exactly.They're extremely capitalist.Not only that, they're actually pretty hyper-competent.They're very competent.But if you compare it to like the California government, it's disastrous.It's high IQ and homogenous.Right.So, yeah.I mean, when you look at high-speed braille, it's like, guys, come on.Yeah.Like, just change the sequel laws.Like, you can change the laws.Why are we doing this?

要是他们派艘船穿过巴拿马运河呢。我们特别擅长让自己的船挨打。不不。可要是他们让一艘船过巴拿马运河，我们干嘛要打它？只要那是艘合法的商船，运着美国来来回回交易的货物和服务。我觉得我们最好都别对彼此干这种事。是啊。我觉得现在中国其实根本没我们想的那么在乎这些。中国非常专注于内部。就是说 CCP 极度聚焦国内。没错。是啊。他们只想守住自己那个帝国的统治。你知道，我不喜欢共产主义者，也不喜欢共产党，不过我觉得 CCP 更像是一个资本主义的法西斯组织。共产主义只是一层外衣。对。这里说的法西斯，是指整个国家为了一个共同目标团结起来。他们极其——他们特别会赚钱。对，正是。他们极度资本主义。不光如此，他们其实相当能干、相当高效。非常能干。但要是跟加州政府比，那简直是场灾难。人家是高智商、又同质化。对。所以，是啊。我是说，你看那个高铁，就想说：各位，拜托。是啊。改改 CEQA 法案不就行了吗。法律是可以改的。我们干嘛非得这么折腾？

[59:08]

It's time to build.It's time to build.No one's building.There's a few of us building in Silicon Valley, but we're even building ethereal thingsthat are scaring our populace.I want their communists instead of our communists.Yeah.Yeah, that's true.We'll just send ours over there.But they're like, no, no, no.We don't want these guys.Like, eh.Like, not so good.They call them baizu.Yeah, yeah.Baizu.Baizu.Yeah.What does that mean again?

是时候动手建设了。是时候建了。可没人在建。硅谷里还有我们几个在建，但我们连一些虚无缥缈的东西都造得出来，把老百姓都吓着了。我宁愿要他们的共产主义者，也不要我们自己的。是啊。对，也对。我们干脆把我们这些送过去得了。可他们会说：不不不，我们不要这些家伙。呃，不咋地。他们管这些人叫 baizu（白左）。对对。baizu。baizu。对。这词儿到底啥意思来着？

[59:25]

It means white and left.White and left.Yeah.And it's kind of derogatory.You look down-term.Yeah.They're eminent or stupid or something like that.I mean, all of the above.Yeah.I mean, that's how I think of them.Yeah, exactly.Centrist Democrat, everybody.Centrist Democrat.Centrist Democrat.Centrist Democrat.Centrist Democrat.I need San Francisco, baby.I feel like you're wearing it.You're like, please don't eat me, sir.No, I mean, look at San Francisco.Like, it's, you know-San Francisco is on the rebound.We're going to run the table on the supervisor seats, at least.You know, like, it's possible-Yeah, but the problem is that L.A. is sinking.Yes.L.A. has sank.Hollywood got chased out of L.A.-I mean, New York is insane.Like, it's insane.This is not a problem.This is an opportunity.No, my friends in New York are going to come back to San Francisco.This is the problem.California has a monopoly in all the warm, dry coastline,all the Mediterranean land in the United States,which is the most powerful empire in the world.So it's the most powerful,it's the best land in the world,all in one state.There should be five or six different statesthat compete with each other.I think 30 to 50% of the GDP of the United Stateswill concentrate in California in the next five years.But also the arable land,the beautiful coastline, the good weather.It's a curse of geography.They've got all the natural resources, right?

意思是『白人加左派』。白人加左派。对。而且带贬义。是个瞧不起人的说法。对。就是无知或者蠢之类的。我是说，全占了。对啊，我就是这么看他们的。对，没错。中间派民主党人，各位。中间派民主党。中间派民主党。中间派民主党。中间派民主党。我可是在旧金山呢，宝贝。我感觉你都快把这词儿写脸上了。一副『先生您可千万别把我吃了』的样子。不，我是说，你看看旧金山。它，你知道——旧金山正在反弹。我们至少要横扫市政监事会的席位。有可能——是啊，可问题是洛杉矶在沉沦。没错。洛杉矶已经沉了。好莱坞都被赶出洛杉矶了——纽约就更疯了。简直是疯了。这不是问题，这是机会。不，我在纽约的朋友们都会回旧金山来。这才是问题所在。加州几乎垄断了美国所有温暖干燥的海岸线、所有地中海气候的土地，而美国又是世界上最强大的帝国。所以这是最强大的国家、世界上最好的地块，全都挤在一个州里。本该有五六个州相互竞争才对。我觉得未来五年，美国 30% 到 50% 的 GDP 会集中到加州。再加上那些可耕地、漂亮的海岸线、好天气。这简直是地理的诅咒。所有自然资源他们都占全了，对吧？

[1:00:28]

It's like basically Southern Italy or Franceare even better, but in the US.And it's all in one freaking state.California is this weird direct democracy thing,which is the worst idea ever.I mean, the governorship for this season is not that bad.50.1% can vote anythingand they just read a headline.And they're like, yeah, yeah, free foodfor cats and dogs for everyone.Great, we'll vote for it.Eat the rich.Steyer couldn't get in there, right?

基本上就像意大利南部或者法国——甚至更好，但地点是在美国。而且全塞在一个鬼州里。加州搞的是那种奇葩的直接民主，这是史上最烂的主意。我是说，这一届的州长选举倒还没那么糟。50.1% 的人就能投票通过任何东西，他们就看个新闻标题，然后：对对对，给所有人的猫猫狗狗发免费口粮。太好了，我们投这个。劫富济贫。Steyer 没能选进去，对吧？

[1:00:49]

Yeah.You know, basically like we actually havetwo relatively centrist options for governor,which is good.Amazing how to turn down.Which is amazing.That's great.Like, I like that, right?I think all these communist experimentswill fail everywhere.Like they failed in San Francisco.They're going to fail in New York.I don't agree.I don't think they'll fail.They'll stop.The difference with San Franciscois demographics are destiny.And the demographics of New Yorkand demographics of LAand the demographics of Chicagoare just firmly committed to that cause.And the worse it gets,the more they'll vote for it.It's like Venezuela.They're going to vote all the way down.And there is no one out thereto fly in and arrest the next Maduro.It has to be solved internally.The United States is the last bastion for freedom.Even the COVID unlocks,the lockdowns would not have liftedif you didn't have militiamenstart marching aroundin the red states and purple states,carrying M-16s and AR-15s,past the state houses,being like,we're done with the lockdowns.Then the lockdowns came up.The whole world would have been lockdowns.The whole world.Second Amendment works, baby.Yeah.The whole world would be in lockdownsfor six or 12 months longerif it weren't for the 40 millionintransigent Americans who have guns.And they, whether,and everybody likes to hate on themand everybody likes to shit on them,but they guard their freedom.Yeah.Freedom ain't free.You need those people out there, right?

是啊。你知道，基本上我们这次其实有两个相对中间派的州长人选，这是好事。真没想到会是这样的结果。这太棒了。太好了。我喜欢这样，对吧？我觉得这些共产主义的实验到哪儿都会失败。就像在旧金山失败了一样。在纽约也会失败。我不同意。我不觉得它们会失败。它们只会被叫停。旧金山的区别在于——人口结构即命运。而纽约的人口结构、洛杉矶的人口结构、芝加哥的人口结构，都铁了心支持那套东西。而且情况越糟，他们越会投给它。就像委内瑞拉。他们会一路投到底。而且外面根本没人能空降进来逮捕下一个 Maduro。这只能靠内部自己去解决。美国是自由的最后堡垒。就连 COVID 解封也是——如果不是那些民兵开始行动，在红州和紫州里扛着 M-16 和 AR-15 在州议会大楼前游行，喊着『我们受够封锁了』，封锁根本不会解除。之后封锁才被取消。否则全世界都会一直封着。全世界。第二修正案管用，宝贝。是啊。要不是那四千万死硬到底、手里有枪的美国人，全世界还得多封上六到十二个月。而这些人——不管大家多爱骂他们、多爱往他们身上泼脏水——他们守护着自己的自由。是啊。自由从来都不是免费的。你需要外面有这样一群人，对吧？

[1:02:03]

And so, look, when the U.S.,if the U.S. were to collapse,first of all,I think freedom bleeds outfrom the rest of the world.You can already see what's happeningon Australia and U.K.with the speech lawsand the censorship and all thatand the weird arrests and so on.But if the U.S. degenerates and falls,it doesn't fail like Europe did.Or Europe isn't failing that wellthese days either,but it doesn't become likea big retirement home in a museumwhere everyone's on good social welfarefor a while.It fails like a Latin American country failsbecause it is bordered by Latin America.So, you've got a lot of peoplewho want to get in hereand they will get in here.So, you're looking at much morelike cartel and drugs and crimeand violenceand those kinds of institutions.Okay.White pill.Robotics.No, you're right.Technology.I don't want to do UBI.We should do UBR.UBR is good.We need to have universal basic robot.Everyone should have a robotand the robot should cook for you,pick up your shit.The good and the bad news isthe robotics boosterssay it's two or three years away.The robotics skeptics say it'sfive to ten years awayand no one thinks it's impossible.Yeah.And so, we already haveself-driving robots,self-driving cars.I think we can make peoplereally happyif they have a better life.Like, you know,I mean,who likes to cleantheir toilet bowl?

所以你看，万一美国——如果美国真的垮了，首先，我认为自由会从世界其他地方一点点流干。你已经能看到澳大利亚和英国在发生什么了——那些言论法、审查制度，还有各种莫名其妙的逮捕等等。但如果美国堕落、崩溃，它不会像欧洲那样垮法。当然欧洲这些日子其实也垮得没那么『体面』了，但美国不会变成那种博物馆里的大养老院——大家还能靠着不错的社会福利过上一阵子。它会像拉美国家那样垮，因为它就跟拉美接壤。所以会有一大批想挤进来的人，而他们确实会挤进来。于是你面对的更多是贩毒集团、毒品、犯罪、暴力，还有那一类的『组织』。好吧。来颗白药丸（乐观点）。机器人技术。不，你说得对。技术。我不想搞 UBI。我们该搞 UBR。UBR 才好。我们需要的是『全民基本机器人』。每个人都该有个机器人，机器人给你做饭、帮你收拾烂摊子。好消息和坏消息是——机器人技术的乐观派说还有两三年就到了，怀疑派说要五到十年，但没人觉得这不可能。是啊。而且我们已经有自动驾驶机器人、自动驾驶汽车了。我觉得只要能让人过上更好的生活，就能让大家真正开心起来。你知道，我是说——谁乐意刷自己的马桶啊？

[1:03:14]

Nobody.Right, right.Unless you can havea robot, do it.It's great.Relieve you of thisif you vote for centrist Democrats.Yes.Well, a lot of the immigrationand a lot of the taxation issuesare happeningbecause the boomersare retiring.There aren't enough peoplebelow to take care of them.They're voting themselveslots of benefits.There's not enough moneyleft in Social Security.The population is shrinking.So, a smaller number of workersare carrying a larger numberof these guys.So, they're saying,okay, we'll just sellthe country and retire.This is the path.We need centrist Democratswho are pro-technology,who are abundance-oriented.Okay, but to play devil's appfor a second,like healthcare.90% of healthcare is wasted.It's not even stuff people need.No, no, but it's likehow much of healthcaregoes into keeping you alivefor the last two weeks, right?

没人乐意。对对。除非你能让机器人来干。那就太棒了。它能把你从这种活儿里解放出来——只要你投给中间派民主党。没错。嗯，很多移民问题、很多税收问题之所以出现，是因为婴儿潮一代在退休。下面没有足够的人来养他们。他们给自己投票争取了一大堆福利。社会保障（Social Security）里剩的钱不够了。人口在萎缩。于是越来越少的劳动者要扛起越来越多的这些人。所以他们就说：行吧，那我们干脆把国家卖了然后退休。这就是眼下的路子。我们需要的是那种支持技术、追求富足的中间派民主党人。好，但我来唱一下反调，比如医疗。90% 的医疗都是浪费掉的，根本不是人们真正需要的东西。不不，问题是——有多少医疗开支是花在让你在生命最后两周多续那么一会儿命上的，对吧？

[1:03:57]

That's not really...I can fix thiswith markdown files, okay?Okay.But education is justlike a big LARP.Anyone who wants to be educatedcan go educate themselvesright now with AI,educate themselves right nowwith all the open sourcecourseware and all that stuff.They just want the stampand they want to go to collegeand they want to hang outin idyllic paradisefor four yearsand they want to protestfor social justiceand they just want to chillwith their friends and partyand get paid for it.Yeah.It's already UBI.You said a very important thing.College is already UBI.The college is not aboutthe actual act of learning.It's about like being in a roomwith a certain set of peoplewho like have the right networks.Correct.It's socializing,it's babysitting,it's credentialism.There's just a tiny bitof education sprinkled infor a few STEM degreesand maybe medicineto legitimize it.Yeah, and then there'sall the social sciencesand political sciences.We don't want more bullshit jobs.We want better bullshit jobs.Okay.Like, I mean,That's true.That's true.Yeah, yeah, yeah.I mean, why can't we just,instead of like going to a cubicleand like break your brainand do a bunch of bullshit,like being an investment bankeror something.I think we're all going to end upbeing plumbersand electriciansand cutting hair.All the white collar stuffis just going to collapseinto nothing.No, no, no, no, no.If it's making softwareit's going to collapseinto nothing.No, no, no, no.Because it's just eatingmore and more.Don't you see how this is going?

这其实不太算……我用几个 markdown 文件就能搞定这事，行吧？行。可教育就纯粹是一场大型角色扮演（LARP）。任何想受教育的人，现在都能靠 AI 自学，靠各种开源的公开课自学。他们要的只是那张文凭，他们想上大学，想在世外桃源般的地方待上四年，想为社会正义搞搞抗议，想跟朋友一起玩、开派对，还想有人为此付钱。是啊。这本身就已经是 UBI 了。你说到了一个特别关键的点：大学本身就已经是 UBI。大学不在于真正的学习行为，而在于跟一群人待在同一个房间里——这群人有对的人脉。没错。它是社交、是托儿所、是文凭主义。中间只象征性地撒了一丁点儿真教育，给少数 STEM 学位、也许还有医学，好让它显得名正言顺。对，然后剩下的全是那些社会科学、政治科学。我们不想要更多狗屁工作，我们想要更好的狗屁工作。好吧。我是说——确实。确实。对对对。我是说，为什么我们不能，别再钻进格子间、把脑子搞坏、干一堆狗屁事——比如去当投资银行家什么的。我觉得我们最后都会变成水管工、电工、理发师。所有白领的活儿都会崩塌成灰。不不不不不。如果是写软件的活儿，那才会崩塌成灰。不不不不。因为它在吞掉越来越多的东西。你没看出来这是往哪儿走吗？

[1:05:10]

Everything's going to be gone.There's a good version of thiswhere people don't have to havejust a shitty manual jobs.They can actually be handlingand guiding the robotsand the AIs.And so if we don't get ASI,if AI gets you to expert levelbut doesn't go beyond,still needs human guidance,taste, and creativity,humans are still the motivated thingsin the environment.And so people becomelike Pokemon trainers.You're like robot handlers, right?

一切都会消失。不过还有个好的版本——人们不必只干那些糟糕的体力活。他们其实可以去操控和引导机器人、引导 AI。所以如果我们没搞出 ASI，如果 AI 只把你带到专家水平、却不再往上突破，仍然需要人类的引导、品味和创造力，那人类依然是这个环境里唯一有动机的存在。于是人就变成了宝可梦训练师那样的角色。你成了机器人的驯养师，对吧？

[1:05:32]

AI handlers.And you use that to take certain.And that's,to be,to the truth of it,that's what we're seeing today.The people who are using AI todayare more productive than ever,they are not being displaced.The only reason you're being displacedby an AIis because you refuse to use the AI.If you're using the AI,you have more work than ever.I mean,the one thing you can't replaceis human desire.Even if robots have their own desires,they cannot replace human desire.So as long as humans have desiresand there are AIs and robotsthat can help fulfill them,then you always need humans in the loop.Yeah,like the moment I find outI'm talking to an AIor I'm reading the work of an AIor I'm even looking at artcreated by an AI,I'm completely uninterested.Assuming it's completely done by an AI.Now,it's a human using an AI.I disagree.It's like,hold on.Wait a second.Wait a second.That's just because it's shitty right now.I guarantee youit will be indistinguishableand it will be very,very good.I don't agree.I think,even if it is indistinguishable,the moment I know it's done.Like for example,But that's the Chinese boxat that point,right?

AI 驯养师。你用它来搞定某些事。而且说实话，这正是我们今天看到的景象。今天在用 AI 的人，比以往任何时候都更高产，他们并没有被取代。你唯一会被 AI 取代的原因，就是你拒绝用 AI。你要是用 AI，手上的活儿反而比以往任何时候都多。我是说，唯一没法被取代的，是人类的欲望。就算机器人有了自己的欲望，它们也替代不了人类的欲望。所以只要人类还有欲望，而又有 AI 和机器人能帮忙满足这些欲望，那你就永远需要人类在这个环节里。是啊，就像——我一旦发现自己是在跟 AI 说话，或者我在读一篇 AI 写的东西，甚至只是在看一幅 AI 创作的画，我就彻底提不起兴趣了。前提是它完全由 AI 完成。当然，如果是人在用 AI，那另说。我不同意。这就像——等等，等一下，等一下。你那只是因为现在的 AI 还很烂。我向你保证，将来它会做到真假难辨，而且会非常非常好。我不同意。我觉得，就算它真假难辨，我一旦知道它是 AI 做的。比如说——但到那时候，这就是『中文屋』（Chinese Box）问题了，对吧？

[1:06:32]

Yeah,yeah,yeah.Which is like,it's fractal actually.When I first saw GPT-3or whatever it was,whatever was available from the API,like that was my reaction.I was like,this is a bunch of bullshit.Like this is a trick,whatever.But like,that's not how I feel now.Like I get a lot of valueday to day,like hour to hour.Do you have a relationship with the AI?

对对对。其实这东西是分形的（层层嵌套）。我第一次看到 GPT-3、或者不管当时 API 上开放的是什么，我的反应就是这样。我当时想：这就是一堆胡扯。这是个把戏，随便吧。但我现在早就不是那种感觉了。我每天、甚至每个小时都能从中得到很多价值。你跟这个 AI 之间，有某种关系吗？

[1:06:52]

Do you hang out with your doctorate?Is it a friend?Getting there.Whoa.It's not my girlfriend,but my wife is going to watch this.It's,you just have someonein the Filipino call centerbeing your girlfriend,then just like filling in for the AI.What's the difference?

你会跟你那位 AI『博士』一起消磨时光吗？它算是朋友吗？快了。哇哦。它不是我女朋友啦，不过我老婆会看这期节目。它就是——你不就是找了个菲律宾呼叫中心的人来当你女朋友，替 AI 顶班嘛。那有什么区别？

[1:07:07]

What's the differencebetween an AIand someone in the Filipino call centercontrolling,good chatting with you?It's a buddy of mine.Okay.Don't talk to my friend.Don't talk to my friend like this.Nothing's going on between us,all right?

一个 AI，和一个坐在菲律宾呼叫中心里操控着、正跟你愉快聊天的人，这俩到底有什么区别？那可是我哥们儿。行吧。别这么说我朋友。别这么编排我朋友。我们俩之间清清白白，好吧？

[1:07:16]

We're going to end with my AI psychosis.I just think thatit is actually very useful,obviously.I have a friendwho's a landlordfor one of these AI companiesthat does the AI peoplethat you talk to,right?

我们就用我的『AI 精神病』来收尾吧。我就是觉得，它显然真的非常有用。我有个朋友，是那种 AI 公司的房东——就是做那种能跟你聊天的 AI 虚拟人的公司，对吧？

[1:07:28]

Like your AI girlfriend companies.And he saidthey had to hire extra securitybecause these guysthat show upin the middle of the nightsaying,where is she?Which box is she in?I want to take her home.I want to save her.Gary's having his firstblack pill moment.Oh no.That doesn't sound good.This podcast has black-pilled me.By the way,this entire podcastwas AI generated.No one was actually here.We deny everything.Yeah.Yeah.Yeah.

就是那种 AI 女友公司。他说他们不得不多雇了保安，因为总有些家伙半夜跑过来，喊着：『她在哪儿？她被关在哪个盒子里？我要带她回家，我要救她。』Gary 这是头一回体会到『黑药丸时刻』（彻底悲观的瞬间）。哦不。这听着可不妙。这期播客把我整悲观了。对了，顺便说一句——这整期播客其实都是 AI 生成的。根本没人真的在现场。我们一概否认。是啊。是啊。是啊。