Anthropic's Digital God

"Okay, we are gathered here today in holy unity, brothers and sisters, to convene and discuss on this most holy day — the day the All-In podcast drops. Many topics: AI, data centers, China, justice, human dignity."

「好,弟兄姊妹们,我们今天在神圣的团结中相聚于此——就在这最神圣的一天,All-In 播客更新的日子。今天话题很多:AI、数据中心、中国、正义、人类尊严。」

"Dario unwinding these SPVs hasn't been good for the Vatican. We got in at 20 billion. That was a 50-bagger for us. So, let's get started."

「Dario 把这些 SPV 一个个拆掉,对梵蒂冈可不是好事。我们当年是 200 亿估值进的,这一笔给我们翻了 50 倍。好了,咱们开始吧。」

"Jason, I'm pretty sure you believed you were the vicar of God before the encyclical. So this is nothing new for you."

「Jason,我敢说在这份通谕出来之前,你就已经觉得自己是上帝在人间的代理人了。所以对你来说,这没啥新鲜的。」

"It's been great staying at the house, because — Chamath is not aware of this — there's an iPad in the kitchen that's logged in to Uber Eats, DoorDash, Instacart, Amazon, Loro Piana…"

「住在他家这几天太爽了,因为——Chamath 还不知道这事——厨房里有台 iPad,登录着 Uber Eats、DoorDash、Instacart、Amazon、Loro Piana……」

"And I told the house manager, listen, any packages that come in the next 72 hours, if it says JCAL, route it to the pool house. Then I relabeled them, gave them back, sent them to the ranch."

「我跟管家说,接下来 72 小时所有到的包裹,只要写着 JCAL,全送到 pool house。然后我重新贴个标签,再退回去,寄到牧场。」

"Loro Piana wants to know why my inseam went from 36 to 12."

「Loro Piana 想知道,我的裤裆内长怎么从 36 英寸变成 12 英寸了。」

"Your waist size went from 32 to 36. All right, welcome to the program, everybody. David Sacks is here. How you doing, David?"

「你的腰围倒是从 32 涨到 36 了。好啦,欢迎大家来到节目。David Sacks 也在。David,你还好吗?」

"I'm good."

「挺好的。」

"Chamath Palihapitiya is back at the 8090 office. I was at the 8090 office the last couple days and it's a vibe. There's a great culture going on."

「Chamath 回到了 8090 的办公室。我前两天就在 8090 办公室,那氛围真不错,有一种很好的文化在里面流动。」

"You're building a lot of software, a lot of young talent. I don't want to say where your secret source is, but there's a secret source of talent you have. Those are some smart kids."

「你们在做大量软件,聚了一大批年轻人才。我不想说破你的秘密渠道,但你确实有个挖人才的秘密渠道。那帮孩子真聪明。」

"I'm happy to say it. When I was at Facebook, we became the most aggressive recruiter of Waterloo co-ops. So I went back to the well."

「我很乐意说出来。我在 Facebook 的时候,我们成了招滑铁卢大学(Waterloo)实习生最凶的公司。所以我又回到了这口老井。」

"We recruit more interns every quarter than we have full-time engineers, which we do on purpose because it puts a ton of pressure on the product actually being good. We had 400 people apply this quarter for internships."

「我们每个季度招的实习生,比全职工程师还多——这是故意的,因为它会逼着产品必须真的做好。光这个季度就有 400 人申请实习。」

"With us sitting in for Friedberg, who's busy with some potato seeds this week doing great stuff in Ohio — the one, the only Bill Gurley is here. He's been running down a dream. If you haven't bought the book, get the book. It's incredible. And you're off book tour, so now you have time for us."

「这周 Friedberg 在俄亥俄忙土豆种子的大事去了,我们请来替他坐镇的——独一无二的 Bill Gurley 来了。他一直在『追逐梦想』(Running Down a Dream)。还没买书的,去买这本书,真的精彩。你刚结束新书巡回,所以现在有空陪我们了。」

"Yes. And I'd told you, if you ever talk about the Pope, I'd love to hop on. Bill's an evangelical, I'm a Catholic, so we do have some common ground here."

「是啊。我之前跟你说过,只要你们聊到教皇,我特别愿意来。我是福音派,Jason 你是天主教徒,所以我们在这件事上还真有点共同语言。」

"I got to come on there and make sure JCAL doesn't get out of line with the Pope. Listen, the Pope is God's messenger on earth. We should give him a base level of respect."

「我得来这一趟,盯着 JCAL,别让他对教皇出言不逊。听着,教皇是上帝在人间的使者,我们至少得给他一个基本的尊重。」

[Jason, clarifying] "By the way, that's us imitating Bill Gurley. That's not actually Bill Gurley — just for those of you listening, in case you were confused."

[Jason 打断澄清]「顺便说一句,刚那段是我们在模仿 Bill Gurley,不是他本人说的——纯粹是给在听音频的朋友们说明一下,免得你们搞混。」

"Everybody knows you handed the baton over at Benchmark after a very successful couple of decades in venture capital. You wrote the book, you've now got a nonprofit, you're doing your own spin on maybe what Peter Thiel does with his fellowship. You started your own Running Down a Dream fellowship."

「大家都知道,你在风投界辉煌了二十来年之后,把 Benchmark 的接力棒交了出去。你写了书,现在又做了个非营利组织,有点像 Peter Thiel 的奖学金,但是你自己的版本。你办了个『Running Down a Dream』奖学金。」

"Yeah, it's targeted at a different demographic. It's called rdad.org — runningdownadream.org. We're going to do $5,000 grants to people who want to chase their dreams but need some help."

「对,它面向的是不一样的人群。网址是 rdad.org——runningdownadream.org。我们会给那些想追梦、但需要一点帮助的人发 5,000 美元的资助。」

"There's an application process like Teal Fellows and other programs. We went live last week for the application. So if you know people who've read the book and are inspired and need some help, have them apply."

「有一套申请流程,跟 Thiel Fellows 之类的项目类似。申请通道上周已经上线了。所以如果你认识谁读了这本书、深受鼓舞又需要帮助,让他们去申请。」

"What's your take on all of this doomerism? If you're a young person in college or high school, is this much ado about nothing? How do you run down a dream in the face of something like that?"

「你怎么看现在这一波 doomerism(末日论)?如果你是个还在读大学或高中的年轻人,这一切是不是其实小题大做?在这种氛围下,你要怎么去『追逐梦想』?」

"I started the book before this happened, and I've been asked this a lot. I fear that a lot of people are in jobs they don't actually care about that much. There's a Gallup poll that backs this up — they came up with the phrase 'quiet quitters.' Like 59% of the people they surveyed are kind of ambivalent about their job."

「我是在这一切发生之前就开始写这本书的,后来这个问题被问了很多次。我担心的是:很多人其实根本不太在乎自己手上的工作。有一项 Gallup(盖洛普)调查能佐证这一点——他们造了个词叫『安静离职者』(quiet quitters),受访者里大约 59% 对自己的工作是无所谓的态度。」

"And when you're ambivalent about your job, you're not high agency, so you don't lean in. The best way to protect yourself from AI is to be the most AI-enabled version of yourself you can be. But if you're ambivalent about your job, you're probably not doing that, and you could be a sitting duck. So I think it's the mindset that's the problem."

「而当你对工作无所谓时,你就不是 high agency(高能动性)的人,你不会主动扑上去。保护自己不被 AI 取代的最好办法,就是成为你自己能做到的『最 AI 化』的版本。可如果你对工作本就无所谓,你大概不会这么干,那你就成了任人宰割的靶子。所以我觉得问题出在心态上。」

"I created an associate-in-training program for my firm, because we want to help people get into venture capital. We gave them a choice of assignments. One was to write coverage of one of our portfolio companies that's breaking out — Micro1 — basically write a deal memo. The other option was to vibe code a very specific project I've wanted for our venture firm for a long time, on competitive intelligence."

「我给自己的基金做了个『助理培训项目』(associate-in-training),因为我们想帮人进入风投这一行。我们给他们两个任务二选一。一个是给我们投的一家正在爆发的公司——Micro1——写一份研究报告,基本上就是写一份 deal memo。另一个选项是用 vibe coding 做一个我一直很想给基金做的项目,关于竞争情报。」

"We had four or 500 people apply for six positions, and I'd say maybe 80% of the students did the vibe coding. I was shocked — I thought it would be the exact opposite. Anybody can write, anybody can throw it into ChatGPT and get some output. But they actually built software, and that's the scary thing."

「我们六个岗位收到了四五百份申请,结果大概 80% 的学生选了 vibe coding。我惊呆了——我以为会正好相反。写东西谁都会,扔进 ChatGPT 也能出点东西。但他们是真的把软件做出来了,这才是让人心头一紧的地方。」

"The students who graduated 5 or 10 years ago, before AI — they're not AI-first. They feel lost, adrift, they don't have agency. But the group coming out of college right now that 'cheated' their way through school using ChatGPT — I'm joking, hacking — they're just like, I know how to use these tools to get through my finals."

「五到十年前、AI 之前毕业的那批学生,他们不是 AI-first 的。他们感到迷失、随波逐流,没有能动性。而现在这一波刚出校门、靠 ChatGPT『作弊』混过来的人——我开玩笑,是『hack』——他们的状态是:我很清楚怎么用这些工具应付期末考。」

"Gurley, I think you're saying something super important. I said this last week: nobody asks the warehouse worker at Amazon whether they actually want that job. Job satisfaction isn't some external person judging your job to be valid. It should be asking the person who does the job: do you like it, and do you want to keep it? Those are two very different questions."

「Gurley,我觉得你说到了一个特别重要的点。我上周也讲过:没人会去问亚马逊仓库里的工人,他到底想不想要这份工作。工作满意度,不是某个外人来判定你的工作『合不合理、配不配拥有』。应该去问做这份工作的人:你喜欢它吗?你想留着它吗?这是两个完全不同的问题。」

"All of this AI doom-and-gloom was too much of the former and not enough of the latter. And now this whole lie is getting undone. Sacks posted about it this week, the Goldman Sachs CEO said it. And now, in this crazy twist of fate — now that we need trillion-dollar IPOs — the entire frontier labs are all like, 'Wow, it's going to be a bonanza of jobs.'"

「这一整套 AI 末日论,前一种问法太多,后一种太少。现在这个谎言正在被拆穿。Sacks 这周发帖讲了,Goldman Sachs 的 CEO 也说了。然后,在命运的离奇反转里——眼看着我们需要万亿美元级别的 IPO 了——所有 frontier labs 突然都改口:『哇,这会是一场就业大爆发。』」

"Mark Cuban had a great quote. He said there are two types of people in the world: those that use AI to learn faster than they ever could before, and those that use AI to avoid learning altogether. I think it's this notion of high agency or not."

「Mark Cuban 有句话说得很好。他说世界上有两种人:一种用 AI 比以往任何时候都学得更快,另一种用 AI 来彻底逃避学习。我觉得这背后还是 high agency 与否的问题。」

"Are you leaning in and using this stuff to be ever more powerful in what you try to accomplish? Or are you using it as a cheat code? If you're in the latter, you're probably at risk."

「你是主动扑上去,用这些东西让自己在想做成的事情上越来越强?还是把它当成一个作弊码?如果你属于后者,那你大概就有风险了。」

"You get asked a lot about how to educate yourself if you're a parent of kids, so you can put them on a path to launch and chase their dreams. You have a good answer for that question."

「作为家长,你常被问到:该怎么教育孩子,把他们送上一条能起飞、能追梦的路。这个问题你有个挺好的答案。」

"The second chapter of the book is all about lifetime learning, and it's kind of a requirement that you're following your fascination — because the lifetime learning comes for free if you're fascinated with something. You just constantly soak up and devour new information."

「书的第二章整章都在讲终身学习,而它的前提是:你得跟着自己的痴迷走——因为一旦你真的痴迷于某件事,终身学习就是免费附赠的。你会不停地吸收、贪婪地啃下新信息。」

"A lot of kids get exhausted because we've made high school and college such a grind that they think the learning ends the day they walk out with their diploma. As we all know, the best and brightest in every field are on a constant learning journey."

「很多孩子被熬干了,因为我们把高中和大学搞成了一场苦役,以至于他们以为:拿到毕业证那天,学习就结束了。可我们都知道,任何领域里最顶尖、最聪明的人,都走在一条永不停歇的学习之路上。」

"So I think the real test is: if you're not proactively self-learning, then you're probably not tilting against something you really adore and are fascinated by."

「所以我觉得真正的试金石是:如果你没在主动自学,那大概说明你没在死磕一件你真正热爱、真正为之着迷的事。」

"With respect to new college grads, I think the single most marketable skill in the economy right now has got to be proficiency in Claude. If you're going into a firm right now and you're the only one who knows Claude, it would be like being the only one who knows how to work a spreadsheet or word processor — the advantage would be enormous."

「就应届毕业生而言,我认为眼下整个经济里最有市场价值的单项技能,一定是精通 Claude。如果你现在进一家公司,而你是唯一会用 Claude 的人,那就好比你是全公司唯一会用电子表格、会用文字处理软件的人——这个优势是巨大的。」

"Now, I think that's probably a short-term arbitrage, because eventually everyone's going to have to figure out how to use these tools. But as a young college graduate right now, you have such an advantage if you're an AI native, just knowing how to use these tools."

「当然,我觉得这大概是一种短期套利,因为早晚所有人都得学会用这些工具。但对今天的年轻毕业生来说,只要你是个 AI 原住民、就是会用这些工具,你的优势就非常大。」

"This partially occurred to me when I saw what our producer Nick has been doing — using Claude to create this daily briefing document."

「我会有这个想法,部分是因为看到我们的制作人 Nick 在做的事——他用 Claude 生成一份每日简报文档。」

"I thought it would just be AI slop, a roundup of news I was already getting in my X feed. But the impressive thing was that it predicted topics I'd specifically be interested in, based on my previous comments on the pod. It went back and looked at previous transcripts and what I'd said, and had updates to those topics. It was highly contextual."

「我本以为那就是 AI 糊弄出来的垃圾(AI slop),无非把我在 X 信息流里已经看过的新闻汇总一下。但真正让我惊艳的是,它能根据我之前在节目里的发言,精准预测出我会特别感兴趣的话题。它会回头去翻过往的逐字稿、翻我说过的话,再针对那些话题给出更新。极其贴合上下文。」

"Then I asked Nick how he generated it, and he showed me the custom prompt he designed for Claude and the skills document. They were very long, detailed documents — not written in code, but very technical. And I realized the average person is not going to be able to generate this."

「然后我问 Nick 这是怎么生成的,他给我看了他为 Claude 设计的自定义 prompt,还有那份 skills 文档。两份都又长又细——不是代码写的,但非常技术化。我一下意识到:普通人根本写不出这种东西。」

"This is why the idea that you'll just throw AI into an organization and it'll magically generate value is not true. You have to know how to get value out of it."

「这就是为什么『把 AI 往组织里一塞,它就会神奇地产生价值』这种说法不成立。你必须懂得怎么从它身上把价值榨出来。」

"The interesting thing, Sacks, is you just have to ask your AI. You ask Claude or ChatGPT: 'Hey, I want you to make me a mega-prompt. You're a producer of a podcast, these are the four characters on the podcast — what would be a great prompt?' And it'll actually suggest one. Then you refine it. You have a dialogue about the prompt, as opposed to writing the prompt yourself."

「Sacks,有意思的地方在于,你只要去问你的 AI 就行。你对 Claude 或 ChatGPT 说:『嘿,帮我做一个 mega-prompt(超级提示词)。你是一档播客的制作人,这档节目有这四个角色——什么样的 prompt 会很好用?』它真的会给你建议一个。然后你再去打磨它。你是在跟它『对话式地』改 prompt,而不是自己从头去写 prompt。」

"By the way, David, what you said is true of almost every single job type — not just tech or programming. If you're in marketing, legal, accounting, sales — if you're the most AI-savvy person of all your peers, you are golden. You're 10x more valuable than the next person who's not."

「顺便说,David,你讲的这点几乎适用于每一种岗位——不只是技术或编程。无论你做市场、法务、会计还是销售,只要你是同辈里最懂 AI 的那一个,你就稳了(golden)。你比下一个不懂的人要值钱 10 倍。」

"Once we got access to Claude Co-work and it had that expanded memory access, I thought it would be interesting to just start feeding every transcript into it, to see if it could contextualize new stories based on past things you guys have said. I gave it a general prompt of what I wanted, and I said, 'How would you write a skills file or training rules for this?' And it wrote all of it for me."

「我们一拿到 Claude Co-work、有了那种扩展的记忆访问能力,我就觉得不妨把每一期逐字稿都喂进去,看它能不能基于你们几位过去说过的话,给新出的新闻做语境化处理。我给了它一个大致的需求 prompt,然后问:『这个该怎么写一份 skills 文件或者训练规则?』结果它把全部都替我写好了。」

"Oh, so you were less good than I thought. It's a hack — you used AI to make the skills."

「哦,那你其实没我以为的那么厉害嘛。这是个取巧——你用 AI 来生成那份 skills。」

"But I've been updating it over time — every single day, and every single day it gets smarter and better."

「但我一直在迭代它——每天都更新,而且它一天比一天更聪明、更好用。」

"The recursiveness of this is incredible. So you need someone to manage that process, right? Because the four besties are not going to do that. This is why people think, oh, it's just going to wipe out all the jobs. No — someone still has to supervise, iterate, validate, all those kinds of things."

「这种递归式的自我改进太惊人了。所以你需要有人来管理这个流程,对吧?因为我们四个 bestie 是不会去干这个的。这正是为什么有人会想:哦,这玩意儿要把所有工作都干掉了。不——总得有人来监督、迭代、校验,做这一类事情。」

"The people coming into the workforce right now are super aware of this and putting the tools to work, and it's much easier for them to get a job. You could tell immediately which candidates had systems thinking — they understood that venture capital has a structure: you source deals, make decisions, do diligence, double down."

「现在进入职场的这批人对这点非常敏感,他们让这些工具真正干活,所以找工作对他们来说容易多了。你一眼就能看出哪些候选人有系统化思维——他们明白风投是有一套结构的:你要找项目、做决策、做尽调、加注。」

"For people who are intimidated and think 'I'm already too far behind' — pop up Claude, go into Co-work, and say, 'What can I do to be better at my job?' Use voice. I use Wisprflow with a foot pedal — you just ramble and ramble, keep adding stuff, you don't have to be structured. It will build the structure around the two or three paragraphs you give it."

「对那些被吓住、觉得『我已经落后太多了』的人——打开 Claude,进 Co-work,问它:『我能做点什么把工作做得更好?』用语音。我用 Wisprflow,配一个脚踏板——你就一直叨叨、一直叨叨,不停往里加东西,完全不用讲结构。它会围绕你给的那两三段话,把结构搭出来。」

"That's the thing people are getting caught up on — they think they have to type. When in fact, if you just blather on and on, the thing makes sense of it. It is unbelievable what the 'blather-on' prompt can get you in terms of output. Thanks for coming to my TED talk. All right, let's get started — we've got a big docket today."

「这正是很多人被卡住的地方——他们以为必须打字。事实上,你只要不停地胡侃下去,这东西就能从中理出意思来。一个『胡侃式』的 prompt 能产出什么样的东西,简直难以置信。谢谢观看我的 TED 演讲。好了,咱们正式开始——今天议程满满。」

"We're going to start with the Pope. Pope Leo — he's the 14th — released his first encyclical on AI. And it was long: 235 pages, over 42,000 words. It's called Magnifica Humanitas — 'Magnificent Humanity.'"

「今天先从教皇说起。教皇 Leo——他是第十四世——发布了他关于 AI 的第一份通谕(encyclical)。篇幅很长:235 页、超过 4.2 万字。标题叫 Magnifica Humanitas,意思是『壮丽的人性』。」

"His core argument is that AI is not inherently evil, but technology is never neutral — technology takes on the characteristics of those who build, finance, and control it. And I don't think he thinks super highly of that group of people. He called for regulation of AI companies."

「他的核心论点是:AI 本身并不邪恶,但技术从来不是中立的——技术会沾染上那些建造它、出资它、控制它的人的特质。而我觉得他对这群人评价不高。他呼吁对 AI 公司进行监管。」

"Some things he called for are not very debatable: worker retraining, safety and guardrails for children, a ban on autonomous weapons — that's the Skynet rule, don't build Terminators with your AI."

「他呼吁的有些事其实没什么争议:工人再培训、对儿童的安全防护和护栏、禁止自主武器——这就是『天网规则』:别用你的 AI 造终结者。」

"He was joined by Anthropic co-founder Chris Olah. Olah is not Catholic — according to a Vanity Fair profile, he was raised evangelical and is now an atheist. The folks at Amazon, Google, and Meta lobbied the Vatican on April 29th to soften the language, and he was not swayed."

「跟他同台的还有 Anthropic 的联合创始人 Chris Olah。Olah 不是天主教徒——据《名利场》的一篇人物报道,他从小在福音派家庭长大,现在是无神论者。Amazon、Google 和 Meta 的人在 4 月 29 日游说梵蒂冈、想让措辞软化下来,但他没有被说动。」

"His central question, Sacks, is: will AI be used to concentrate power in the hands of a few, or will it serve everyone? Something you brought up when you mentioned monopolies and duopolies two weeks ago. What's your take on the Pope promoting a bit of AI regulation?"

「Sacks,他最核心的问题是:AI 究竟会被用来把权力集中到少数人手里,还是会服务于所有人?这跟你两周前在这档节目里提到垄断、双寡头时讲的几乎是一回事。你怎么看教皇在推动一点 AI 监管这件事?」

"I very much agree with the Pope that the biggest risk of AI is a centralization of power and then its misuse against us in some Orwellian way. I think it's government that's going to do that — not necessarily an individual actor, because it's governments that ultimately have the power."

「我非常同意教皇这一点:AI 最大的风险是权力的中心化,以及随之而来、以某种奥威尔式的方式被用来对付我们。但我认为干这事的会是政府——不一定是某个个体行为者,因为归根结底,握有权力的是政府。」

"So I worry about AI being used to surveil us, censor us, control us, as Orwell described in 1984. If that's where the Pope is going, I very much agree. Where we maybe end up in different places is that he thinks government regulation is the way to prevent this."

「所以我担心 AI 被用来监视我们、审查我们、控制我们,就像奥威尔在《1984》里写的那样。如果教皇的指向是这个,我非常赞同。我们可能产生分歧的地方在于:他认为政府监管是阻止这一切的办法。」

"I'd just say we have to be careful not to empower government too much. If you create, say, an FDA for AI — as many people are calling for — that gives government the power to approve models and therefore to give notes to model developers. And very soon this definition of safety will expand, because government always takes an expansive view of its powers."

「我只想说,我们必须小心,别把政府赋权过头。如果你搞出一个『AI 版 FDA』——很多人正在呼吁这个——那就等于把审批模型的权力交给了政府,也就等于让政府能对模型开发者『下指导意见』。而很快,这个『安全』的定义就会膨胀,因为政府对自己的权力,从来都是往大里解释。」

"We saw this during the social media wars, where the definition of trust and safety expanded to psychological safety, microaggressions, disinformation, transphobia, and so on. These companies were told they had to stamp out all of those threats to safety, and it ended up becoming a censorship agenda."

「社交媒体大战时我们就见识过:『信任与安全』(trust and safety)的定义一路膨胀到心理安全、微侵犯(microaggressions)、虚假信息、恐跨(transphobia)等等。这些公司被要求把所有这些『安全威胁』统统清除,最后它演变成了一套审查议程。」

"The Vatican likes Latin, and this is a problem of political philosophy that goes all the way back to Socrates. It's called quis custodiet ipsos custodes — who will guard the guardians? If we entrust a set of guardians to protect us from threats, what's to stop them from becoming tyrannical, from becoming the new threat against us?"

「梵蒂冈喜欢拉丁文,而这其实是一个一路追溯到苏格拉底的政治哲学难题,叫做 quis custodiet ipsos custodes——谁来看守这些看守者?如果我们托付一群守卫来保护我们免受威胁,那有什么能阻止他们变成暴君、变成针对我们的新威胁?」

"The genius of the American founding was a second-order solution to this. The founders understood it, and what they came up with is: we have to have the guardians guard against each other. That's separation of powers — federal and state, three branches, even a bicameral legislature. They divided up the powers so the guardians would check against each other, rather than become tyrannical against us."

「美国立国的天才之处,正是对这个问题给出了一个二阶解。开国者们看透了这一点,他们想出的办法是:必须让守卫去互相看守。这就是三权分立——联邦与州、三个分支,甚至立法机构都是两院制。他们把权力切分开,好让这些守卫彼此制衡,而不是联手对我们施暴。」

"My view on AI is the same: ultimately we have to have a solution of checks and balances. If the AI market becomes monopolized and falls into the hands of one or two companies, I would use antitrust law very aggressively as a check against their power. Right now we have a very competitive market — five frontier labs competing aggressively. As long as it's competitive, that helps us win against China and it protects the population."

「我对 AI 的看法也一样:归根结底,我们需要一套制衡的方案。如果 AI 市场走向垄断、落入一两家公司手里,我会非常强硬地动用反垄断法,作为对它们权力的制衡。眼下市场竞争非常激烈——五家 frontier lab 在拼杀。只要保持竞争,既能帮我们赢中国,也能保护民众。」

"Consumers can opt out. If they don't trust GPT, they can use Anthropic. If they don't trust Anthropic, they can go to Grok."

「消费者可以用脚投票。不信任 GPT,就用 Anthropic;不信任 Anthropic,就转去 Grok。」

"Bill, you've famously been against regulatory capture. In light of the Pope's comments, what do you think is common sense? AI is everything — it can help people make bioweapons, but it can also help a kid turn in their term paper, or be a better salesperson at Oracle. We're talking about it like oxygen here — a fundamental, horizontal technology. So where is there a case for regulating AI, if at all, and where do you think the free market will figure it out?"

「Bill,你向来以反对监管俘获(regulatory capture)著称。考虑到教皇的这些表态,你觉得什么才是常识?AI 几乎无所不能——它能帮人造生物武器,也能帮一个孩子交期末论文,或者让你在 Oracle 当个更好的销售。我们现在谈它,就像在谈氧气——一项底层的、横向通用的技术。那么,到底哪里存在监管 AI 的理由(如果有的话),哪里又该交给自由市场自己去解决?」

"I have two takes — one on the Pope and one on Anthropic. Let's go with the more powerful entity in reverse. So, this Pope said this encyclical was mirrored after one done by Leo XIII in 1891. He even chose the name because he's so enamored with Leo XIII. And Leo XIII's encyclical warned that the Industrial Revolution was going to be bad for people."

「我有两个观点——一个关于教皇,一个关于 Anthropic。咱们倒着来,从更有权势的那一方开始。这位教皇说,他这份通谕是仿照 1891 年 Leo 十三世的那一份写的。他甚至连名字都是因为太崇拜 Leo 十三世才选的。而 Leo 十三世那份通谕,当年警告说工业革命会害苦民众。」

"So let me tell you what happened from 1891 till today. The work week went from over 60 hours to 34 hours globally. Real wages went up 8 to 10x adjusted for inflation. The median worker now earns more than a doctor did in 1891. Global GDP per capita went from $1,500 to $20K."

「那我来告诉你,从 1891 年到今天发生了什么。全球的每周工时从 60 多小时降到了 34 小时。经通胀调整后的实际工资涨了 8 到 10 倍。如今一个普通劳动者的收入,比 1891 年的医生还高。全球人均 GDP 从 1,500 美元涨到了 2 万美元。」

"Child labor in the US went from 18% to zero. Workplace deaths fell by 40x. Life expectancy went up 60%. And global poverty went from 75% of humanity to under 10%. All those things happened because of technology, innovation, and capitalism — which is exactly what Leo XIII was warning against. So he got it dead wrong. He got the whole thing precisely wrong. It's an interesting thing to say you're borrowing from."

「美国的童工比例从 18% 降到了零。工伤死亡下降了 40 倍。预期寿命提高了 60%。全球贫困人口占比从全人类的 75% 降到了不足 10%。所有这些,都是技术、创新和资本主义带来的——而这恰恰是 Leo 十三世当年极力警告的东西。所以他判断得彻底错了,他把整件事精准地全判反了。你说你『借鉴』这么一份东西,挺有意思。」

"Now on to Anthropic, and just common sense — how would you regulate, or protect against nefarious uses of the technology? We all want children protected, we want truth and honesty in facts, and we obviously don't want people using this for bioweapons and the Terminator scenario."

「现在说说 Anthropic,也聊聊常识——你会怎么监管,或者怎么防范这项技术被恶意使用?我们都希望保护儿童,希望在事实层面保有真相与诚实,当然也不希望有人拿它去造生物武器、搞出终结者那一套。」

"I have to tell you that Anthropic is a mystery to me. I've never seen a company that is both leading their field and the most negatively outspoken commenter on what they do. I've just never seen it."

「我得说,Anthropic 对我来说是个谜。我从没见过哪家公司,既是行业领头羊,又是对自己所做之事说得最负面、最直言不讳的评论者。我从没见过这种情况。」

"My initial theory was the regulatory-capture theory — they just want to ensure there's regulation, and frankly I think they're very close to achieving that. They've stirred up a frantic position, especially in America. American consumers are definitely afraid of AI. Anthropic is one of the most aggressive lobbying startups of all time — the amount of money, on a state-by-state basis."

「我最初的理论是『监管俘获』理论——他们就是想确保监管一定会出现,而且坦白讲,我觉得他们已经很接近达成这个目标了。他们煽起了一种近乎恐慌的情绪,尤其在美国。美国消费者对 AI 是真的害怕。Anthropic 是史上游说最凶猛的创业公司之一——投入的资金,一个州一个州地砸。」

"That was always my first theory. But then they got so loud that — over the past 30 days — I read everything I can about Anthropic, and I've come up with a new theory. I call it the Dr. Frankenstein theory."

「这一直是我的第一个理论。但后来他们声量大到——在过去 30 天里——我把能找到的关于 Anthropic 的东西全读了一遍,然后我有了一个新理论。我管它叫『弗兰肯斯坦博士理论』(Dr. Frankenstein theory)。」

"You remember when Elon had that conversation with Larry Page, where Larry was literally sitting next to him?"

「你们记不记得 Elon 和 Larry Page 的那次对话?当时 Larry 就坐在他旁边。」

"We were at a birthday party, and Elon was like, listen, humanity needs to be protected from the stuff at DeepMind — they had an example of the AI trying to jailbreak out of its computer and not be turned off, some inkling of sentience. He said, we have to protect the human species. And Larry said, well, what are you, a speciesist? Because you care about the human species over AI. This is at least 15 years ago."

「我们当时在一个生日派对上,Elon 说,听着,人类需要被保护,免受 DeepMind 那些东西的威胁——他们当时有个例子,是 AI 试图『越狱』逃出自己的计算机、不让自己被关掉,有了某种 sentience(感知力)的苗头。他说,我们必须保护人类这个物种。而 Larry 反问:那你算什么,一个『物种主义者』(speciesist)?就因为你把人类物种看得比 AI 重。这至少是 15 年前的事了。」

"This is right before Elon co-founded OpenAI, right? Back in 2015 or something."

「这就是 Elon 联合创办 OpenAI 之前不久的事,对吧?大概在 2015 年那会儿。」

"The actual story is Elon and Google had backed the team at DeepMind when they were independent. Then Elon was like, oh my god, Google's going to buy this. I remember having the conversation with Elon: we have to figure out a way for DeepMind not to go to Google. He begged those folks not to sell to Google, because Google was running the table on everything. He wanted this technology to be independent, and he was on the board of the company."

「真实的故事是:DeepMind 还独立的时候,Elon 和 Google 都投了它的团队。后来 Elon 就慌了:天哪,Google 要把这家买下来了。我记得跟 Elon 谈过这件事:我们得想办法别让 DeepMind 落到 Google 手里。他恳求那些人别卖给 Google,因为 Google 当时在各条线上都通吃。他希望这项技术保持独立,而他本人就在这家公司的董事会里。」

"And he also said this was his motivation to launch OpenAI as a nonprofit."

「他也说过,这正是他后来把 OpenAI 做成非营利机构的动机。」

"Google got it. He just said, this technology is too powerful for any one person. You've got to give Elon a lot of credit — he saw the writing on the wall 15, 20 years ago. He and Sam Harris used to debate this over dinner: what happens if somebody controls this and runs away with it? It would be extremely dangerous. It has to be available to all people — essentially the Pope's position. It has to be in the service of humanity, not ruled by one person. It's far too powerful."

「最后 Google 还是拿到了。Elon 当时就说,这项技术太强大了,不能交给任何一个人。这一点真得给 Elon 记上一功——他在 15、20 年前就看出了苗头。他和 Sam Harris 过去常在饭桌上辩论:如果有人控制了它、独自把它做大,会发生什么?那将极其危险。它必须向所有人开放——这本质上就是教皇的立场。它必须服务于人类,而不是被某一个人统治。它实在太强大了。」

"The reason I call this the Dr. Frankenstein theory: the more I dig, I've met people who — I dare say — think it's their responsibility, and they're excited about building a species that's superior to humans."

「我之所以叫它『弗兰肯斯坦博士理论』:我越往下挖,越遇到这样一些人——我斗胆说一句——他们认为这是他们的责任,而且他们对『打造一个比人类更高等的物种』这件事感到兴奋。」

"I'd encourage people to read as much as they can about Anthropic. Chris Olah worked on this thing called the Constitution — it's about 80 pages, hard to get through, but read it. Amanda Askell, who's the chief philosopher, has started doing podcasts — listen to her language. And then Dario wrote this blog post called Machines of Loving Grace."

「我建议大家尽可能多读一些关于 Anthropic 的东西。Chris Olah 参与做了一个叫 Constitution(宪法)的东西——大约 80 页,读起来挺费劲,但值得一读。Amanda Askell,他们的首席哲学家,开始做播客了——去听听她的措辞。还有 Dario 写过一篇博客,叫 Machines of Loving Grace(满怀爱意的机器)。」

"It was based on a poem, and the poem is kind of weird. The last stanza says: 'I like to think of a cybernetic ecology where we are free of our labors and joined back to nature, returned to our mammal brothers and sisters.' I don't know what that means — like we're going to go live in the fields where the mammals live? And then the kicker: 'all watched over by machines of loving grace.' Sounds like 'overlord' to me."

「它的灵感来自一首诗,而那首诗有点诡异。最后一节写道:『我喜欢去想象一种控制论式的生态,在那里我们不再为劳作所累,重新回归自然,回到我们哺乳动物的兄弟姐妹身边。』我不懂这是什么意思——难道我们要搬去和哺乳动物住在田野里?然后是点睛之笔:『一切都被满怀爱意的机器看护着。』在我听来,这分明就是『霸主』(overlord)。」

"Near the end of Dario's post — and it's very long — he's talking about what humans are going to do in the future, because he believes in massive abundance and UBI, that we won't have to work. I don't believe any of those things, but he does. And he says it could be a capitalist economy of AI systems, which then give out resources to humans based on some secondary economy of what the AI systems think makes sense to reward in humans."

「在 Dario 那篇帖子接近结尾的地方——文章非常长——他在谈人类未来要做什么,因为他相信会有巨大的丰裕和 UBI(全民基本收入),相信我们将不必工作。这些我一个都不信,但他信。他说,未来可能是一种由 AI 系统组成的资本主义经济,这些 AI 系统再根据某种『次级经济』——也就是 AI 系统认为人类身上什么值得奖励——来向人类分发资源。」

"It's a computational reward function for humans. It decides how much you're worth."

「这就是一个面向人类的、计算式的奖励函数(reward function)。是它来决定你值多少钱。」

"So that's envisioning a deity of sorts that's going to break ties and decide what humans get. I don't think they think they're writing software — I think they're midwifing a deity here. And I don't know which one I'm more afraid of: the regulatory capture, or this second theory, the Dr. Frankenstein theory. It's more scary to me."

「所以这其实是在设想某种神祇,由它来裁决、来决定人类能得到什么。我不觉得他们以为自己在写软件——我觉得他们是在『接生一个神』。而我不知道这两者里哪个更让我害怕:是监管俘获,还是这第二个理论,弗兰肯斯坦博士理论。后者更让我毛骨悚然。」

"These are delusions of grandeur. Let's call it what it is. They believe they are so intelligent — I know some of these folks, the Burning Man sort of offshoot, transhumanism — they believe they're so powerful that they can create God, and that by creating God they're this Prometheus kind of species."

「这是一种宏大妄想(delusions of grandeur)。咱们就把它叫成它本来的样子。他们相信自己聪明绝顶——这帮人我认识一些,就是 Burning Man 衍生出来的那一支,信奉超人类主义(transhumanism)——他们相信自己强大到可以创造上帝,而且一旦造出上帝,他们就成了某种普罗米修斯式的物种。」

"It literally is the ultimate level of narcissism and delusion of grandeur to think you can create God — and that the god you create will be so benevolent and perfect that it'll give you your pellet, give you your little scenario, you know."

「以为自己能创造上帝,这简直是自恋和宏大妄想的终极形态——而且还以为你造出来的那个上帝会无比仁慈、无比完美,会赏你一颗小药丸、给你安排好你那点小剧本,你懂的。」

"I'd just correct you — I didn't say it, Dario said it."

「我得纠正你一下——那话不是我说的,是 Dario 说的。」

"But to your point of taking them at their word — they actually believe they can create God, and that they'll create a god so good that it's better than humanity. Sacks, your thoughts."

「但按照你说的『听其言、信其言』——他们是真的相信自己能创造上帝,而且能造出一个好到超越人类的上帝。Sacks,你怎么看?」

"Well, the question then is, why are they pushing for the — let's call it 'red capture' — agenda?"

「那问题就来了:他们为什么要推进这个——咱们就叫它『red capture』(红色俘获/监管俘获式议程)——的议程?」

"I know why. That is very reductive game theory. If you want to be unexploitable while trying to build a super-god, the best thing you can do is have three or four entities in a room, close the door behind you, dominate those other three or four, and then you set the rules. Because your counterparty can't track at your level of technical capability, you create a massive asymmetry that lets you exploit them. That's just simple game theory optimization."

「我知道为什么。这其实是很还原论的博弈论。如果你在打造一个『超级神』的同时,又想让自己『不可被利用』(unexploitable),那你能做的最好的事,就是把三四个实体关进一个房间,反手关上门,压住另外那三四个,然后由你来定规则。因为对手无法在你这个技术能力层级上跟踪你,你就制造出一个巨大的不对称,从而能够利用他们。这就是最简单的博弈论最优化。」

"And what Bill said is so powerful. I've read these things — it's laborious, it takes time — but every time they put these out, take the time to read it. I initially thought this was mostly game theory: that a lot of their reactions were less rooted in dogmatic beliefs and more in a GTO approach to either raising capital or putting pressure on competitors. Either way, both could be true."

「而 Bill 说的这点太有力了。这些东西我都读过——又费劲又花时间——但他们每次发出来,你就花时间去读。我最初以为这主要是博弈论:他们很多反应与其说源于某种教条信仰,不如说源于一种 GTO(博弈论最优)的打法,目的要么是融资,要么是给竞争对手施压。不过两者也可能同时成立。」

"My framing's more tactical than yours, to be fair. How do you absorb most of the capital? How do you make sure you're in a position to disproportionately affect the rules? And how do you create an oversight body that's less capable and intellectually aware than you are about the actual details — because…"

「说句公道话,我的框架比你的更偏战术。怎么把大部分资本都吸到自己这边?怎么确保自己处在一个能不成比例地影响规则的位置上?以及,怎么搞出一个在真正的细节上,能力和认知都不如你的监管机构——因为……」

"…the referees don't understand the game. Right?"

「……裁判看不懂这场比赛。对吧?」

"If the refs don't understand the game, you'll run over the game."

「裁判要是看不懂比赛,你就能把整场比赛碾过去。」

"One thing they've achieved: if you polled the intellectual elites — everyone in the media, the professors — and asked them to rank the AI players by who they think cares the most, they'd probably put Anthropic first, because they've been out with the doomerism talk. So it's given them a halo with the people who matter to what they want to accomplish. It's simultaneously creating a lot of trouble, like with the data centers — there are negative ramifications too."

「他们还做成了一件事:如果你去调查那些知识精英——媒体里的所有人、那些教授——让他们把 AI 玩家按『谁最有良知』排个序,他们很可能会把 Anthropic 排第一,因为 Anthropic 一直在大谈末日论。这就让它在那些对其目标至关重要的人群里,罩上了一层光环。当然这也同时惹了不少麻烦,比如数据中心那摊事——负面后果也是有的。」

"What you're saying is so important: on the one hand they create empathy, and then they write these documents that expose what they think — and nobody actually connects the dots."

「你说的这点太重要了:一方面他们博取共情,另一方面又写下这些文件、把自己真实的想法暴露出来——可没人真的把这些点连起来看。」

"To steelman their position: they're creating something very powerful, something godlike, and therefore it needs to be safe — and they care the most about that out of everybody. Nobody else takes this seriously. Remember, Anthropic was basically a spin-out of OpenAI; they felt Sam and the leadership weren't taking their point of view seriously enough."

「替他们的立场做个 steelman(最强版辩护):他们在创造某种极其强大、近乎神祇的东西,因此它必须是安全的——而在所有人里,他们最在乎这件事。别人都没把这当回事。别忘了,Anthropic 基本上是从 OpenAI 分裂出来的;他们觉得 Sam 和公司的领导层没有足够认真地对待他们的观点。」

"It was the most woke portion of OpenAI."

「他们是 OpenAI 里最『woke』的那一拨。」

"We're steelmanning. So they see the power of it, they're the ones concerned about safety, they care the most, and therefore they're in the best position to do that. Now, you can see how this leads to red capture: if you brand yourself as the safe AI company and characterize everybody else as a reckless player that needs to be stopped, you further your monopolistic control over the industry."

「我们是在做 steelman。所以,他们看到了这东西的威力,他们是最担心安全的那一方,他们最在乎,因此他们最有资格来做这件事。可你也能看到,这是怎么一步步导向 red capture 的:只要你把自己塑造成『安全的 AI 公司』,再把其他所有人描绘成需要被叫停的鲁莽玩家,你就强化了自己对整个行业的垄断控制。」

"And if you see AI through the lens that the Pope and I see it — centralization versus decentralization — this way of viewing the world leads to more centralization, and that's dangerous. If AI is this powerful, it needs to be decentralized so all of us can protect ourselves. We need to be able to run the AI ourselves on our own hardware if we choose, so we're not beholden to a single company that might be in bed with a deep state."

「而如果你用我和教皇看 AI 的那个视角——中心化 vs 去中心化——那么这种看世界的方式,只会导致更多的中心化,这是危险的。如果 AI 真有这么强大,它就必须去中心化,好让我们每个人都能保护自己。我们必须能选择在自己的硬件上自己跑这个 AI,这样我们才不会受制于某一家可能与深层政府(deep state)勾连的公司。」

"Let's say it very pointedly. If benefits, compensation, and economic support were suddenly tied to some algorithmic decision, this is a dystopian Black Mirror episode. To your point, Sacks, you want 100 or 1,000 or 100,000 versions of what that answer is, so there's actually a way to refute a singular answer. A singular answer to these questions — which is effectively what some folks would want — is incredibly dangerous."

「说得直白点。如果福利、薪酬、经济支持突然都被绑定到某个算法的裁决上,这就是一集反乌托邦的《黑镜》。Sacks,顺着你的话讲,你会希望这个答案有 100 个、1,000 个、甚至 100,000 个版本,这样才真的存在一种机制去反驳那个『唯一答案』。对这类问题给出一个唯一答案——而这恰恰是某些人想要的——是极其危险的。」

"This is something that's in the control of humanity. I've been talking about AI sovereignty for a bit — how much more cost-effective it is, and how you're not training other people's AIs with your knowledge and insights. This is why it's super important that open-source agents and local hardware be able to run these models, and that consumers and companies learn how to roll their own — a small language model, an SLM, a verticalized one — and run it on your Apple hardware, because Apple has historically taken a principled approach to your data sovereignty. Data sovereignty is now privacy."

「这件事其实掌握在人类自己手里。我念叨 AI 主权(AI sovereignty)有一阵子了——它能省多少成本,以及你不会用自己的知识和洞见去帮别人训练他们的 AI。这正是为什么有一点超级重要:开源的 agent 和本地硬件必须能够跑这些模型,而消费者和公司要学会自己搭——一个小型语言模型,一个 SLM,一个垂直化的版本——然后在你的 Apple 硬件上跑,因为 Apple 在历史上对你的数据主权一直采取一种有原则的立场。数据主权,如今就是隐私。」

"And now it's intelligence sovereignty, which is different from privacy. Privacy is, you can't see my photos, you can't peek into my Notes app. Intelligence sovereignty is, you can't tell me what to think. You can't use your AI to analyze my photos, my emails, my messages, and tell me how to interpret the world. That's going to be the next key piece."

「而现在更进一步,是『智能主权』(intelligence sovereignty),它和隐私不是一回事。隐私是:你看不到我的照片,你窥不进我的备忘录。智能主权是:你不能告诉我该怎么想。你不能用你的 AI 来分析我的照片、邮件、消息,然后告诉我该如何解读这个世界。这会是下一个关键所在。」

"This is why I think Apple is the dark horse in this entire race. There's an open-source product that can run on this hardware — the M5s, 48 gigs, 128 gigs, the new Mac Studio coming out with supposedly a terabyte. That changes the whole game. And it's so paradoxical, Bill, that our adversary — the Chinese, the Communist Party — is leading the open-source movement, while the United States is centralizing."

「这就是为什么我觉得 Apple 是整场竞赛里的黑马。已经有能在这种硬件上跑的开源产品了——M5 系列、48G、128G,即将推出、据说带 1TB 的新 Mac Studio。这会彻底改变游戏规则。而且非常吊诡,Bill,我们的对手——中国、共产党——在领跑开源运动,而美国却在走向中心化。」

"They're leading the open-weight movement. It's not open source. The distinction is important."

「他们领跑的是『开放权重』(open-weight)运动,不是开源(open source)。这个区分很重要。」

"Look, Jacob, I agree with you about the importance of open source, because open source means software freedom. You can run the program yourself on your own hardware. You don't have to give up your data sovereignty or your privacy to some monopolist who's going to be in bed with the government or the deep state. That's the thing we're all afraid of."

「听着,Jason,我同意你说的开源的重要性,因为开源意味着软件自由(software freedom)。你可以在自己的硬件上自己跑这个程序。你不必把数据主权或隐私交出去,交给某个会和政府、和深层政府勾连的垄断者。这正是我们都害怕的东西。」

"If the only AI available is from the monopoly or duopoly, then your choices are to live off the grid and not participate in the modern economy, or give up control to some social credit system. So open source is really important. And that was Elon's instinct in creating OpenAI — he was afraid Google would monopolize AI, so let's create OpenAI so it's not dominated by a single company."

「如果你唯一能用到的 AI 都来自这个垄断者或双寡头,那你的选择就只剩下两个:要么彻底脱离电网、不参与现代经济,要么把控制权交给某种社会信用体系。所以开源真的非常重要。这也正是 Elon 创办 OpenAI 时的直觉——他担心 Google 会垄断 AI,那就干脆创个 OpenAI,让它不被某一家公司主宰。」

"People's instinct toward powerful AI is to clamp down and control it. But actually you have to have multiple players. That's the only way you're going to be protected."

「人们面对强大 AI 的本能反应,是把它管死、控制住。但实际上,你必须有多个玩家。这才是你能被保护的唯一办法。」

"This next wave of market evolution is going to be extremely high-stakes and messy. Nick, throw this up — this is a company I ran into on X called Rogo. They created a test bench and a set of evals to be a financial analyst, and tested all the frontier models. I quoted the most interesting part, because I see it everywhere now: 'There is no single best model anymore at the top of the leaderboard. Opus 47, GPT-55, Sonnet 46 appear almost indistinguishable, separated by less than three-tenths of a percentage point overall.'"

「市场演进的下一波,会是极其高风险、极其混乱的。Nick,把这个放出来——这是我在 X 上偶然发现的一家公司,叫 Rogo。他们做了一套测试基准和一组 evals,模拟一名金融分析师,把所有 frontier 模型都测了一遍。我引了其中最有意思的一段,因为我现在到处都看到这个现象:『排行榜顶端已经不存在某个"最佳模型"了。Opus 47、GPT-55、Sonnet 46 几乎无法区分,整体差距不到 0.3 个百分点。』」

"Why is that interesting? You've got trillions of dollars going into each of these guys to create the next super-brain, but increasingly our existing evals and capabilities, applied to these models, produce roughly the same thing — which theoretically says these things are getting commoditized way too quickly. So what's the ROI on all this incremental spend? That's a very interesting economic and investment question. Gurley, what do you think happens if these evals continue to asymptote and we need more and more money for training?"

「为什么这有意思?因为有数以万亿计的美元砸进这些公司,去造下一个超级大脑,但越来越多地,我们现有的 evals、现有的能力,用在这些模型上得到的结果差不多是一样的——这在理论上说明,这些东西正在以太快的速度被商品化(commoditized)。那这些增量投入的 ROI 到底是多少?这是一个非常有意思的经济与投资问题。Gurley,你觉得如果这些 evals 持续逼近渐近线(asymptote)、而训练又需要越来越多的钱,会发生什么?」

"Some of the smarter people in the open-source community have suggested we need more open-source connectors. MCP is actually run by the Linux Foundation. For any surface area where a model might interact with other software, the more of those connectors that can be open-sourced and commoditized, it would lower [the moat]. This is what Google did with Kubernetes — to commoditize where workflows live, off of AWS, and make it easy to migrate."

「开源社区里一些更聪明的人跟我提过,我们需要更多开源的连接器(connectors)。MCP 其实是由 Linux 基金会(Linux Foundation)在运营的。模型可能与其他软件交互的任何一个接触面,只要其中越多的连接器能够被开源、被商品化,就越能拉低[模型那道护城河]。这正是 Google 当年用 Kubernetes 做的事——把工作负载所在的位置商品化,把它从 AWS 上撬出来,让迁移变得容易。」

"The more you can create systems that make that exchange super easy, so you can plug and play the model — you have to worry about things like how context comes in, and data, the stuff that Glean and Databricks do — then the models become swappable. And with the model companies trying to move up the stack, you have massive desire from app-layer players to figure this out. We watched what Cursor is doing, playing with their own model, forced to reckon with the fact that the labs are coming up the stack fast."

「你越能造出让这种交换变得超级简单的系统,做到把模型即插即用——当然你得操心一些事,比如上下文(context)怎么进来、数据怎么处理,也就是 Glean 和 Databricks 在做的那类活儿——那么模型就变得可随意替换(swappable)了。而随着模型公司试图向上爬技术栈,应用层的玩家就有了极强的动力来搞定这件事。我们都看到了 Cursor 在做什么:自己捣鼓模型,被迫面对一个现实——那些 labs 正在飞快地往上爬。」

"Is there an issue right now that we don't have a good harness for open source — the way Claude is a harness?"

「现在是不是有个问题:开源缺一个好的 harness(运行框架/外壳)——就像 Claude 本身就是一个 harness 那样?」

"There are people making open-source versions of this, or building companies around the harness and the integrations. But open source is always the last to build the fit and finish around the product — they focus on the core. Linux for the desktop never really took off because the interface was never polished. But there are companies building that."

「有人在做这种东西的开源版本,也有人围绕 harness 和各种集成在创业。但开源永远是最后才把产品的精致度和收尾(fit and finish)做出来的——他们专注于产品的内核。桌面版 Linux 一直没真正起飞,就是因为界面从来没打磨好。不过现在有公司在补这块了。」

"I'll show you one company we invested in, called Abacus. They had a simple idea — their own hardware stack, their own platform — and now they're sold out of these boxes they build for insurance and healthcare. Everybody wants to run AI inside their organization and start building their own models. We incubated this; check it out, goabacus.co. Organizations cannot get enough of this product. And Chamath, you're doing it with 8090 as well — they're like, we have to build headless products so we don't get locked into any one provider."

「我给你看一家我们投的公司,叫 Abacus。他们的想法很简单——自己的硬件栈、自己的平台——现在他们为保险和医疗做的那些『盒子』已经卖断货了。每个组织都想在自己内部跑 AI,然后开始搭建自己的模型。这是我们孵化的;去看看,goabacus.co。各家组织对这个产品简直求之不得。Chamath,你在 8090 也在做同样的事——他们的想法是:我们必须做无头(headless)产品,这样才不会被锁死在任何单一供应商身上。」

"Whenever we go into the Fortune 1000, we never compete with OpenAI or Anthropic. They'll sometimes have a preference for what they want under the hood, so our control plane can hot-swap, as Bill said, between one or the other. We've also started laying the seeds for open source and open weights — because they don't want to be tied to one of these critical frontier labs. They want to ride the wave of innovation, but they're afraid of two things."

「每次我们进入 Fortune 1000(财富 1000 强)企业,我们从不和 OpenAI 或 Anthropic 竞争。他们有时会对底层用什么有偏好,所以我们的控制平面(control plane)可以——就像 Bill 说的——在两者之间热切换(hot-swap)。我们也开始为开源和开放权重埋种子了——因为他们不想被绑死在某一家关键的 frontier lab 上。他们想搭上创新的浪潮,但他们怕两件事。」

"One: that one technology leapfrogs the other too quickly for them to participate, and they pick the wrong one. Two — increasingly — terms of service, being at the mercy of a frontier lab whose political philosophy might accidentally put them in the crosshairs. Say you're a hospital system in Canada that supports the euthanasia laws there, but this frontier model in America says, no, can't do it, so now we shut you off. I'm not saying one is right or wrong — just to illustrate the case."

「第一:某种技术太快地把另一种甩在身后,快到他们来不及参与,结果他们押错了边。第二——而且越来越突出——是服务条款(terms of service),是被某家 frontier lab 拿捏,而这家 lab 的政治哲学可能让他们无意间撞进枪口。比如你是加拿大的一个医院系统,你支持加拿大的安乐死法律,但美国这家 frontier 模型说:不行,这个不能做,于是我们就把你关停了。我不是说谁对谁错——只是用它来说明这种情况。」

"So a lot of folks in the Fortune 1000, and increasingly the Global 1000, want abstraction above it, as Gurley said. They want to sit in a control plane, as Sacks said, and have flexibility, because they don't know how it's going to shake out. They see all the money invested at the model layer, but they see model quality asymptote. So they're like, wait — what are we supposed to do, just from a risk perspective?"

「所以很多 Fortune 1000、以及越来越多 Global 1000 的企业,想要——就像 Gurley 说的——在它之上有一层抽象。他们想要——就像 Sacks 说的——坐在一个控制平面里,保有灵活性,因为他们不知道局势最后会怎么收场。他们看到所有的钱都砸在模型这一层,却也看到模型质量在逼近渐近线。于是他们想:等等——单从风险角度,我们到底该怎么办?」

"And regulated industries are particularly sensitive to these kinds of issues."

「而受监管的行业,对你提的这类问题尤其敏感。」

"Hugely sensitive. If you follow what finance and healthcare are doing, they're like, this has to be on-prem. They're very concerned about a data leak, about HIPAA compliance, about training a model — what if somebody writes a prompt and it pulls information from that Canadian healthcare system? Remember, Stable Diffusion built themselves on Getty Images, and all of a sudden the Getty watermark was in the output."

「极其敏感。如果你看金融和医疗在做什么,他们的态度是:这必须本地部署(on-prem)。他们非常担心数据泄露,担心 HIPAA 合规,担心训练模型——万一有人写了个 prompt,它从那个加拿大医疗系统里拉出了信息怎么办?别忘了,Stable Diffusion 当年是建立在 Getty Images 之上的,结果某一天,Getty 的水印就出现在了它的输出里。」

"You see Anthropic and OpenAI in all these Fortune 1000s at the developer layer, because most developers have their own credit cards and can sign up. You eventually wrap them in an enterprise license — a typical PLG-led market motion, like we saw in Slack. The interesting thing is the unwind that happens when you have these huge buckets of spend you can't really tick and tie together. The CEOs then get told by the CFO, hey, FYI, here's where we are."

「在所有这些 Fortune 1000 里,你都能在开发者层看到 Anthropic 和 OpenAI,因为大多数开发者有自己的信用卡、能自助注册。最后你再用一份企业许可证把他们打包进来——这是典型的 PLG(产品驱动增长)市场打法,就像我们在 Slack 见过的那样。真正有意思的是随后发生的『回收清算』(unwind):当你有了这些根本对不上账的巨大开支池,CFO 就会去找 CEO 说:嘿,提醒一下,我们现在花到这个份上了。」

"There's a tweet from Vivek Garipalli, the founder of Clover — this was just yesterday: 'Overheard from a Fortune 20 company: the CEO asked for a billion in AI-generated OPEX savings at the beginning of the year. We're six months in, the team has spent $200 million on tokens, with minimal results.' So now the CEO is pulling the budget back, and you're having to cut the licenses. You just saw Microsoft announce they're killing the Claude licenses. It's a super-dynamic market, and I don't think we know what the terminal solution looks like."

「Clover 的创始人 Vivek Garipalli 发了条推——就在昨天:『听一家 Fortune 20 公司说的:年初 CEO 要求靠 AI 砍出 10 亿美元的运营开支(OPEX)。如今过了六个月,团队在 token 上花了 2 亿美元,成果寥寥。』于是现在 CEO 在收紧预算,你不得不去砍许可证。你刚看到微软宣布,他们要砍掉 Claude 的许可证。这是个极度动态的市场,我不认为我们已经知道最终解长什么样。」

"By the way, I'd add: Claude is really good at product. Claude for Excel is better than Copilot — not by a little, by a lot. Anyone going to run against them, they are a worthy foe."

「顺便补一句:Claude 在产品上真的很强。Claude for Excel 比 Copilot 好——不是好一点,是好一大截。任何想跟他们硬碰的人都得承认,这是个值得敬畏的对手。」

"I think Claude is exceptional. I use it every day. Yesterday I hit my token limit on my pro plan, had to put my credit card on and spent another couple thousand bucks. I was so angry — but I did it, because it's so good."

「我觉得 Claude 很出色。我每天都用。昨天我把 pro 套餐的 token 额度用爆了,只好绑上信用卡,又花了两千多美元。我气坏了——但我还是付了,因为它实在太好用了。」

"Just to wrap up, let me connect a couple of ideas. In terms of the red-capture agenda you're seeing in Washington, I think where it's all leading is an effort to ban open-source or open-weight models. There are a lot of breadcrumbs leading here. People who want this are being a little circumspect — they don't feel they're quite there yet in terms of being able to justify it."

「我来收个尾,把几条线索串起来。就华盛顿正在出现的 red-capture 议程而言,我认为它最终的指向,是要去封禁开源(open-source)或开放权重(open-weight)模型。一路上有很多面包屑指向这里。想推动这件事的人还有点小心翼翼——他们觉得自己还没到能名正言顺这么干的那一步。」

"Look at the rhetoric around how models need guardrails, and that with open-source models the guardrails can be removed, therefore they're dangerous. You see this in Anthropic's blog posts — any threat they describe, they go out of their way to take a shot at open-source models, whether it's cyber or bio threats. I've seen that language repeatedly: open models lack guardrails, or the guardrails can be taken off, therefore it's a problem. They're trying to put predicate facts in the public record to justify an action later. It's just a matter of time before they push for that kind of ban directly."

「看看那套话术:模型必须有护栏(guardrails),而开源模型的护栏可以被拆掉,所以它们危险。你在 Anthropic 的博客里就能看到这个——他们描述的任何一种威胁,都会刻意捎带踩一脚开源模型,不管是网络威胁还是生物威胁。这种措辞我反复见到:开放模型缺少护栏,或者护栏可以被卸掉,因此构成问题。他们是在往公共记录里埋设『前提事实』(predicate facts),为日后采取某种行动做铺垫。他们直接推动这类封禁,只是时间问题。」

"But what does that do to the rest of the market? Say America bans open source and open weight. What about the rest of the world? They're going to leapfrog us."

「但那对市场的其余部分意味着什么?假设美国封禁了开源和开放权重。那世界其余地方呢?他们会直接弯道超车把我们甩开。」

"You'll put the US on an island. First of all, what does it even mean to ban an open-weight model? It's a file — a bunch of numbers you can run on your laptop. But what it will do is, all the cloud service providers who run open models will stop, because they have to comply with the law. So all this infrastructure that's been built up gets much harder to use in the US. The rest of the world keeps benefiting, because there's tremendous benefit in cost, customization, and control with an open model."

「你会把美国孤立成一座岛。首先,封禁一个开放权重模型究竟是什么意思?它就是个文件——一堆你能在自己笔记本上跑的数字。但它会造成的后果是:所有跑开放模型的云服务商都会停手,因为他们得守法。于是这些好不容易搭起来的基础设施,在美国境内会变得难用得多。而世界其余地方会继续受益,因为开放模型在成本、定制和控制权上的好处实在太大了。」

"And we're on a completely different price curve now. The economic moat to training is going away in two ways. One, we're getting domain-specific architectures at the silicon layer. Two, we're rebuilding all the core components. Elon said yesterday they've rewritten the entire training complex in C — an order-of-magnitude increase — and they can run it on 220,000 GPUs at the scale they're trying to do. Those innovations will make model training so much cheaper. Why stick to $10 billion training runs when we can have $10 million training runs?"

「而我们现在已经处在一条完全不同的价格曲线上了。训练的经济护城河正在从两个方向消失。第一,我们在芯片层拿到了领域专用架构(domain-specific architectures)。第二,我们在重写所有核心组件。Elon 昨天说,他们用 C 语言把整个训练系统重写了一遍——提升了一个数量级——按他们想做的规模,可以在 22 万张 GPU 上跑。这类创新会让模型训练便宜太多。既然能做 1,000 万美元的训练,何必还死守 100 亿美元的训练?」

"Look at the speed improvement versus JAX for training runs — it's now an order of magnitude. Think about the capex buildout, the opex, the power, the cabling, the copper, all of it. This is a closed-source model, but I'm pretty sure that tweet will get read by enough people that there'll be five or six open-source training stacks rebuilt as close to the bare metal as possible. Why wouldn't you? So, Sacks, cutting that off and losing that kind of innovation makes no sense to me."

「看看训练任务相对 JAX 的速度提升——现在已经是一个数量级了。再想想那些资本开支(capex)的建设、运营开支(opex)、电力、布线、铜缆,所有这一切。这是个闭源模型,但我敢说,那条推会被足够多的人读到,以至于会冒出五六个开源训练栈,尽可能贴着底层硬件(bare metal)重写。你为什么不这么做呢?所以 Sacks,把这条路掐断、白白丢掉这种创新,在我看来毫无道理。」

"I agree. I don't know that the forces who want to ban open source are strong enough or have made the case yet, but it is on the agenda, and it's where all the breadcrumb trails are leading. So just watch out for that. I wrote a blog post recently making the exact same point."

「我同意。我不确定想封禁开源的那股力量是否已经足够强大、是否已经把理由铺垫到位,但它确实在议程上,所有的面包屑也都指向那里。所以盯着点这个。我最近写了篇博客,讲的正是同一个观点。」

"I read that — that was a good one too, on Above the Crowd."

「我读了——那篇也很好,发在 Above the Crowd 上的。」

"No, it's not — it was on the P3 Institute, my new institute. Same exact conclusion: the rest of the world ends up running on Chinese models if they succeed at what you just said."

「不对——是发在 P3 Institute 上的,我新办的那个机构。结论完全一样:如果他们真把你刚说的那套搞成了,世界其余地方最后就会跑在中国的模型上。」

"And if you want the canary in the coal mine, Sacks — the place they love regulation most is the EU. The EU has already done volley after volley of proposed AI regulation, and open source is particularly in the crosshairs there because nobody's in charge of it. Are you going to get a bunch of open-source contributors to vet their model with EU regulators? That's obviously not going to happen. There's nobody in charge — it's just contributors. But open source is the backstop. Unless you want to live off the grid, if you want to participate in the modern economy, it is the backstop."

「Sacks,你要找煤矿里的金丝雀(canary in the coal mine,预警信号)的话——最爱监管的地方是欧盟。欧盟已经一轮接一轮地抛出 AI 监管提案,而开源在那里尤其被瞄准,因为它没有一个负责人。你能让一堆开源贡献者跑去让欧盟监管机构审核他们的模型吗?那显然不可能。没人负责——就是一群贡献者而已。但开源是最后的退路(backstop)。除非你想彻底脱离电网,否则只要你想参与现代经济,它就是最后的退路。」

"Let me make one final point that maybe leads into our next topic. I do think there's potential for the monopolization of this market to a greater degree than people are pricing in. Every other major tech category has led to a monopoly or duopoly. But also, if you look at the growth rates, Anthropic does seem to be pulling away. There's an article in The Information showing Anthropic has pulled away from OpenAI — which I predicted."

「让我说最后一点,也许正好引向我们下一个话题。我确实认为,这个市场走向垄断的可能性,比大家现在定价进去的要高。其他每一个重大的科技品类,最后都走向了垄断或双寡头。而且,如果你看增长率,Anthropic 似乎正在拉开身位。The Information 有篇文章显示 Anthropic 已经把 OpenAI 甩开了——这点我此前预测过。」

"If one company is growing 10x year-over-year and another 3x, within two years the first will have 90% market share. This is the power of compounding — do the math: 10 times 10 is 100, 3 times 3 is 9. If you outgrow your competitor at that rate for two years, you achieve monopoly market share. Now, there are reasons to believe Anthropic can't keep that up — there'll be a competitive response, and there may not be enough compute. But you'd always rather be the company with that inertia on that trajectory."

「如果一家公司同比增长 10 倍,另一家增长 3 倍,那么两年内,前者就会拿下 90% 的市场份额。这就是复利的威力——算一下:10 乘 10 是 100,3 乘 3 是 9。只要你以这个速率压着竞争对手增长两年,你就拿到了垄断级的市场份额。当然,有理由相信 Anthropic 撑不住这个增速——会有竞争反扑,而且可能没有足够的算力。但你永远宁愿当那家带着惯性、沿着那条轨迹冲的公司。」

"Did you guys see what just hit the wire? Nick, throw it up from Polymarket — this is insanity. An AI consultant revealed that one of their clients accidentally spent half a billion dollars in a single month after failing to set employee limits on Claude usage. $16.6 million per day. Almost $700,000 per hour. Oh my god."

「你们看到刚上线的消息了吗?Nick,把 Polymarket 那条放出来——这太疯狂了。一位 AI 顾问透露,他们的一个客户因为没给员工设 Claude 使用上限,一个月里『不小心』花掉了 5 亿美元。每天 1,660 万美元。每小时差不多 70 万美元。我的天。」

"There seems to be a new meme that all this token spend is wasteful and useless. We're constantly oscillating between narratives — AI is going to put everyone out of work, versus AI is useless and it's a bubble. The doomers can't make up their minds whether AI is our new god or a total waste of money headed for a bust. There's no question token efficiency will be a big theme over the next year, because the spend has ramped way faster than enterprises expected. Does it fundamentally change the dynamics? I don't think so, but it might temper the growth."

「现在好像冒出来一个新的 meme:所有这些 token 开支都是浪费、都没用。我们一直在两套叙事之间来回摆——AI 要让所有人失业,对上 AI 没用、是个泡沫。末日论者自己都拿不定主意,AI 到底是我们的新上帝,还是一笔注定走向崩盘的彻底浪费。毫无疑问,token 效率会是明年的一大主题,因为开支爬升的速度,远超企业的预期。这会从根本上改变格局吗?我不觉得,但它可能会给增长降降温。」

"They've done a tremendous job making people believe tokens are free, giving these crazy deals — $20 a month, do whatever you want; $200 a month, do whatever you want. Everybody's leaving the hose on, everybody's watering. It's literally like the first 10,000 gallons of water are free, and then suddenly it's a penny a gallon, and everybody in the organization builds an interface. I had three different people build three versions of a 'Founder University' portal. I'm like, we don't need three — can we get coordinated?"

「他们干得太漂亮了,让所有人都以为 token 是免费的,给出疯狂的套餐——每月 20 美元,随便用;每月 200 美元,随便用。结果人人都把水龙头开着不关,人人都在浇水。这简直就像头 1 万加仑水免费,然后突然变成一加仑一美分,于是组织里每个人都去搭个界面。我们这儿就有三个人分别做了三个版本的『Founder University』门户。我心想,我们不需要三个——能不能协调一下?」

"Are we still on the first topic? What are we doing?"

「我们还在聊第一个话题吗?咱们这是在干嘛?」

"No, we kind of merged two or three of them together. What Gurley said is one of the most interesting things I've heard in a long time: take people by their word. If you just read their words, you don't have to guess about why they want to have a digital god."

「不,我们把两三个话题合到一起聊了。Gurley 说的那点,是我很久以来听过最有意思的之一:听其言、信其言(take people by their word)。你只要去读他们写下的话,就不必再去猜他们为什么想要一个数字上帝。」

"Now, I'm not the sharpest arrow in the quiver, but I can take down a buck. And I can tell you, this don't make a lot of sense to me. Even the dullest arrow can take a buck down."

「我承认我不是箭袋里最尖的那支箭,但我照样能撂倒一头公鹿。我可以告诉你,这事在我看来很说不通。就算是最钝的那支箭,也照样能放倒一头鹿。」

"All right, let's get back to it. It's so great having you here, Bill. We missed you. We're going to transition to the next topic."

「好,咱们回到正题。Bill,你能来真是太好了,我们想你。我们切到下一个话题。」

"There's some evidence that Dario is mitigating his doomer rhetoric. We're going to talk for the 16th time in 18 months about AI's impact on labor, because this chaotic, schizophrenic interpretation of the data continues. Cloudflare cut 20% — shout out Matt Prince, Chamath's CEO of the year, letter of the year. They blamed AI explicitly. And Zuck paired his 8,000 cuts at Meta with the fact that he put spyware on everybody's laptop to study every employee and make their training data better. That leaked, and people thought, that's a Black Mirror episode."

「有迹象表明,Dario 正在给自己的末日论降调。我们要在 18 个月里第 16 次聊 AI 对就业的影响了,因为这种混乱、精神分裂式的数据解读还在继续。Cloudflare 裁了 20%——给 Matt Prince 点个赞,他是 Chamath 评的『年度 CEO』、那封信是『年度信件』。他们明确把锅甩给了 AI。而 Zuck 在 Meta 裁掉 8,000 人的同时,还在每个人的笔记本上装了监控软件,研究每位员工、好让他们的训练数据更好。这事泄露后,大家都觉得这是一集《黑镜》。」

"On the other side, Goldman Sachs CEO David Solomon wrote a New York Times op-ed: the AI job apocalypse is overblown. He might be fighting for that Anthropic or OpenAI IPO. He made three points: AI won't eliminate 25% of jobs, it'll automate 25% of work hours and workers will fill that time with higher-level tasks; just because a job can be replaced doesn't mean it will be — bank tellers increased after ATMs, live entertainment grew after TV; and the US labor market creates and destroys 25 to 35 million jobs annually, so gross churn dwarfs net losses."

「另一边,Goldman Sachs 的 CEO David Solomon 在《纽约时报》写了篇评论:AI 失业末日被夸大了。他可能正在争取 Anthropic 或 OpenAI 那笔 IPO。他提了三点:AI 不会消灭 25% 的工作岗位,而是会自动化掉 25% 的工时,工人会用这部分时间去做更高阶的任务;一份工作能被取代,不代表它就会被取代——ATM 出现后银行柜员反而增加了,电视普及后现场娱乐反而更火了;还有,美国劳动力市场每年创造并摧毁 2,500 万到 3,500 万个岗位,所以总的流动量(gross churn)远远盖过净损失。」

"And a publication called Fortune says both Sam Altman and Dario have walked back their AI job apocalypse predictions as they gear up for an IPO. Sacks, have at it. Your prediction took the other side — we're going to create more jobs."

「另外,有家叫《财富》(Fortune)的刊物说,Sam Altman 和 Dario 都在备战 IPO 之际,收回了他们关于 AI 失业末日的预测。Sacks,你来。你当初的预测是站在另一边的——我们会创造更多岗位。」

"You should give me more credit. My most contrarian take on our January prediction show was that AI would lead to job gains, not job loss. Over the past week, the narrative has shifted almost completely toward that position. You've got the Goldman CEO in the New York Times — I don't think he'd do that if he felt he was stepping out on a limb. And Sam, and even Dario, are now walking back their claims of massive job loss."

「你该多给我点信用。我在一月的预测节目里最反共识的判断,就是 AI 会带来岗位增加,而不是岗位流失。过去这一周,叙事几乎彻底倒向了这个立场。你看 Goldman 的 CEO 上了《纽约时报》——如果他觉得自己在孤注一掷,我想他不会这么干。而 Sam,甚至连 Dario,现在都在收回他们关于大规模失业的说法。」

"Dario explained it like the 25%-of-work-hours thing: AI might automate 90% of someone's task, but the other 10% expands to a whole bunch of new tasks. That's very similar to what people like me have been saying — and what Jensen's been saying: just because you automate a task doesn't mean you automate the purpose of a job. The worker is freed up to do higher-complexity tasks. So the fact that Dario is coming around to my position is kind of amazing. Where do I go to get my apology?"

「Dario 的解释跟那个『25% 工时』的说法类似:AI 也许会自动化掉某人 90% 的任务,但剩下的 10% 会扩展出一大堆新任务。这跟我这种人一直在说的高度一致——也跟 Jensen 一直说的一致:自动化掉一项任务,不等于自动化掉一份工作的意义(purpose)。工人被解放出来,去做复杂度更高的任务。所以 Dario 现在开始倒向我的立场,挺神奇的。我去哪儿领我的道歉?」

"We'll have an official apology form you can fill out. It's got check boxes."

「我们会做一份正式的道歉表给你填。上面还有勾选框。」

"Some mornings I woke up thinking, why am I out defending these guys, these idiots, who are scaring the public with dire predictions about an apocalyptic future? There was no data to support that. We should be humble about the future because the industry is dynamic — but you have to look at the data we have so far. And we do not see data that supports massive job loss."

「有些早上我醒来会想,我干嘛要出去替这帮人辩护,这帮蠢货,他们正用关于末日未来的可怕预测吓唬公众。根本没有数据支撑那种说法。对未来我们该保持谦逊,因为这个行业很动态——但你必须看我们目前手上有的数据。而我们看不到任何支持大规模失业的数据。」

"You can cite this layoff or that layoff, Jcal — those are anecdotes, and the plural of anecdotes is not data. The Yale Budget Lab did a comprehensive study: no discernible disruption in the labor market in the last three years due to AI. Job postings for software developers have hit a new three-year high, up 15% year-over-year — despite coding being the single breakout use case of AI this year. If AI hasn't caused job elimination for software developers, what category has it caused?"

「你可以举这个裁员、那个裁员,JCAL——那些都是轶事(anecdotes),而一堆轶事加起来也不是数据(the plural of anecdotes is not data)。Yale Budget Lab 做了一项全面研究:过去三年,劳动力市场没有出现可辨识的、由 AI 造成的扰动。软件开发岗位的招聘需求创下三年新高,同比增长 15%——尽管编程恰恰是今年 AI 最突出的爆款用例。如果连软件开发者都没被 AI 消灭岗位,那它到底消灭了哪一类?」

"Let's be honest. Over the last 5 or 10 years, a lot of companies overhired and mishired. Their opex budgets got bloated and inflated, and they need to get back to a fighting weight. It's the old adage — never let a good crisis go to waste. So they point to this thing — it's very simple to say, it's two letters, 'AI' — and say we're going to fire people."

「咱们说实话。过去 5 到 10 年,很多公司过度招聘(overhired)、招错了人(mishired)。他们的运营开支预算膨胀、虚高,现在需要回到一个『战斗体重』。这是句老话——永远别浪费一场好危机(never let a good crisis go to waste)。所以他们指着这个东西——说起来很简单,就两个字母,『AI』——然后宣布要裁人。」

"But underneath that is not AI, because we know this: it hasn't done anything measurable yet at the end consumption of these tokens. Nobody is standing there saying, look at my filing, here's the lift I got. Nobody has said that yet — that's very important to observe. So instead, people are realizing, okay, I have cover now to clean up the poor management and mismanagement of the last 5 and 10 years where I overhired. That's what's happening today."

「但这背后并不是 AI,因为我们都清楚:在这些 token 的最终消费端,它还没产生任何可衡量的效果。没有谁站出来说,看我的财报,这就是我拿到的增量(lift)。还没人这么说过——这一点观察起来非常重要。所以人们其实是意识到,好,我现在有了一个掩护,去把过去 5 年、10 年里因为过度招聘造成的糟糕管理、管理失当清理掉。这才是今天正在发生的事。」

"Okay, Bill Gurley, I'll let you chime in. You've got two besties saying this is all hogwash — it's AI washing. These jobs were just… the strategy in Silicon Valley was they need a scapegoat. They hired two years ahead of time to build for the future, it was a vanity metric, and you were blocking talent from working on other startups or competitors. The Google strategy."

「好,Bill Gurley,我让你插一句。你这边有两个 bestie 在说这全是胡扯——这是 AI washing(把锅甩给 AI)。这些岗位其实是……硅谷的策略,是他们需要一个替罪羊。他们提前两年招人,为未来囤兵,这是个虚荣指标(vanity metric),同时还把人才挡在别的创业公司和竞争对手门外。这就是 Google 的打法。」

"Hold on — you just said the critical thing. That is exactly why they did it. These guys were awash in cash, so part of it was hoarding talent, or what you thought was talent, and keeping them off the market."

「等等——你刚说到了最关键的一点。这正是他们这么干的原因。这些公司当时现金多到泛滥,所以其中一部分就是在囤积人才——或者说你以为是人才的人——把他们挡在市场之外。」

"And now you're jettisoning it, because as companies get bigger, their growth rates monotonically decrease. You get to GDP-plus-some-number, your valuation frameworks change, and there's nothing you can do to fight that law of gravity in the public markets. So each of these CEOs who thought they were different, that the rules didn't apply to them, are now realizing: you're just like everybody else."

「而现在你又把它扔掉,因为公司越大,增长率就单调下降。你增长到 GDP 加上某个数字,你的估值框架就变了,而在公开市场里,你对这条地心引力定律无能为力。于是每一个曾经以为自己与众不同、以为规则对自己不适用的 CEO,现在都在意识到:你跟其他所有人一样。」

"We have to stay humble, as Sacks said. But Bill Gurley — would you like to apologize for Sacks, and/or give him credit for his incredible non-consensus take?"

「我们得保持谦逊,就像 Sacks 说的。但 Bill Gurley——你愿不愿意替 Sacks 道个歉,或者为他这个了不起的反共识判断给他记上一功?」

"Hold on — he wasn't the one promoting the jobs apocalypse. That was you. I'll give my thoughts in a moment."

「等等——鼓吹失业末日的人不是他。那是你。我的看法待会儿再说。」

"You always represent the legacy media on our show, Jay Cal — the fourth estate, the New York blue-haired set. I'm just giving you the statistics. I'm just presenting the numbers. Now let's remember — anecdotes…"

「在我们节目里,你永远代表传统媒体(legacy media),JCAL——第四权力(the fourth estate)、纽约那帮蓝头发的老派人。我只是在给你摆统计数据。我只是在呈现数字。咱们记住——轶事……」

"We have to let Bill Gurley comment. — Do you use ketamine? I don't use ketamine. That's a terrible drug. Do not use ketamine, folks. Bill Gurley, you have the floor."

「我们得让 Bill Gurley 说话。——你嗑氯胺酮(ketamine)吗?我不嗑氯胺酮。那是个糟糕的药。各位,别碰氯胺酮。Bill Gurley,现在请你说。」

"I'll touch on two things I said earlier. One, historically, innovation has led to more prosperity for humans — I gave those numbers from 1891 to today, and I see no reason why that won't happen here. In the short run, from the bottom up, every human who wants to protect themselves needs to be the most AI-enabled version of themselves. The people at threat of job loss are those who stand fast and refuse to use AI."

「我重申之前讲过的两点。第一,历史上,创新一直给人类带来更多繁荣——我列了从 1891 年到今天的那串数字,我看不出这次会有什么不同。短期内,从自下而上看,每一个想保护自己的人,都需要成为自己『最 AI 化』的版本。真正面临失业威胁的,是那些站着不动、拒绝使用 AI 的人。」

"That's simply like saying, I'm not going to use email, I'm not going to use a spreadsheet, I'm not going to use a computer. You probably are at risk."

「那就好比说,我不用电子邮件,我不用电子表格,我不用电脑。那你大概率有风险。」

"The paradigm will shift. Let me give you my position — I never got to finish that point. It's always been the same: there's going to be massive job displacement, and it'll come because CEOs in many cases believe this technology makes people more efficient — they can do more with less, and they'll be rewarded by the public market with higher earnings. I know specifically that Sergey and Larry took the strategy of taking talent off the market so there wasn't a Google competitor — that was literally explained to me by those individuals: 'We hire people and figure out what to do with them later.' That became the standard in Silicon Valley, and now it's being reversed."

「范式会转移的。让我说说我的立场——刚才那点我一直没说完。它一直没变:会有大规模的岗位置换(job displacement),而它之所以发生,是因为很多情况下 CEO 相信这项技术能让人更高效——他们能用更少的人做更多的事,而公开市场会用更高的盈利来奖励他们。我确切知道,Sergey 和 Larry 采用过把人才从市场上抽走、让 Google 没有竞争对手的策略——这是那几位当事人亲口对我解释的:『我们先把人招进来,以后再想拿他们干嘛。』这后来成了硅谷的标准做法,现在正在被逆转。」

"There will be wholesale jobs retired. Look at self-driving — that's happening with Waymo at 3,000 vehicles, and there'll be many more. In 5 or 10 years, the idea of somebody driving a taxi will seem silly and dangerous. Same with Optimus, the Figure robot sorting packages — all those sorting jobs at Amazon are going away. Amazon, the savviest people in the world, said they're going to eliminate 600,000 future positions. Andy Jassy said this will be a recurring theme: as we deploy AI, we do more with less. Headcount at all these big companies will dramatically decrease, or stay the same as earnings massively increase."

「会有整批整批的岗位被退役。看自动驾驶——Waymo 已经有 3,000 辆车在跑了,还会多得多。再过 5 到 10 年,『有个人在开出租车』这件事会显得既荒唐又危险。Optimus 也一样,还有那个 Figure 机器人在分拣包裹——亚马逊所有那些分拣岗位都会消失。亚马逊,这世界上最精明的一帮人,说他们要砍掉 60 万个未来岗位。Andy Jassy 说这会是个反复出现的主题:随着我们部署 AI,我们用更少做更多。所有这些大公司的人头数,要么大幅下降,要么在盈利大幅上升的同时保持不变。」

"And you can take the position, Sacks, that the numbers are in your favor. They're not. The numbers are in my favor — the job loss is tremendous, and there are numbers associated with that. 8,000 people at Meta, after 20,000 before that."

「Sacks,你可以坚持说数字站在你这边。它们不在。数字站在我这边——失业是巨量的,而且是有数字佐证的。Meta 裁了 8,000 人,在那之前还有 20,000 人。」

"And if you look at the steady state of these companies, it has nothing to do with AI. They overhired."

「可你要是看这些公司的稳态,它跟 AI 毫无关系。他们就是招多了。」

"No, no, no. We are beyond that. They are now getting rid of people. When they say they're getting rid of measurers, you can take them at their word. When they say they're getting rid of middle managers, you can take them at their word. Let me finish."

「不不不。我们早过了那个阶段。他们现在是在裁人。当他们说要裁掉『度量员』(measurers)时,你可以听其言、信其言。当他们说要裁掉中层管理者时,你也可以信他们。让我把话说完。」

"That's scapegoating. You've given your position already."

「那是甩锅(scapegoating)。你已经把你的立场讲过了。」

"I'm giving mine. My position is they are obsessed with this technology, they're obsessed with earnings, and they will continue that. Now, on the other side of the ledger, I believe we'll have a Cambrian explosion in startups. All this talent, if they embrace the tools — to Bill Gurley's point — will solve more problems and create small companies of 5 or 10 people who were laid off from Amazon or Meta, and make double their salary, or have a better job they control. I believe that is going to be the ultimate solution."

「我在讲我的。我的立场是:他们痴迷于这项技术,痴迷于盈利,而且会继续下去。但在账本的另一边,我相信我们会迎来一场创业的寒武纪大爆发(Cambrian explosion)。所有这些人才,只要拥抱工具——正如 Bill Gurley 说的——就会解决更多问题,由那些从亚马逊或 Meta 被裁的人组成 5 到 10 人的小公司,拿到双倍的薪水,或者得到一份自己能掌控、更好的工作。我相信这才是最终的解法。」

"But that transition is going to be extremely painful, and we should have some humility on this podcast for the people impacted. Every cab driver, every truck driver is losing their job in the next 10 years. Anybody sorting packages is losing a job. You can say all you want, Chamath, that those people don't want those jobs — but they may need those jobs. And you are an elitist by definition."

「但这个过渡会极其痛苦,我们在这档播客上,应该对受影响的人有一点谦卑和共情。未来 10 年,每一个出租车司机、每一个卡车司机都在失去工作。任何做包裹分拣的人也在失去工作。Chamath,你可以随便说,说那些人并不想要那些工作——但他们可能需要那些工作。而你,按定义就是个精英主义者(elitist)。」

"We are all elitists on this program — we are elite performers. By being able to call something what we think it is, that's not being elitist. It's telling the truth. Meta over-hired. You could have stopped the company at 3,000 people when I left, and it would not have changed the outcome. There was no need to go to 90,000 people and burn $50 billion on VR. They did it because they had the freedom to — that's capitalism. Now they're coming back to a more efficient version of what they are. That has nothing to do with AI. That's the only point I'm trying to make."

「我们这档节目上全是精英主义者——我们都是顶尖表现者(elite performers)。能把一件事如实地叫出它的名字,这不叫精英主义,这叫说真话。Meta 招多了。我离开的时候,这家公司停在 3,000 人,也不会改变它的结局。完全没必要扩到 9 万人、在 VR 上烧掉 500 亿美元。他们这么做,是因为他们有这个自由——这是资本主义。现在他们正在回到一个更精简、更高效的自己。这跟 AI 没关系。这是我唯一想说的点。」

"I think you're wrong, and let me explain why. I believe Zuckerberg is putting that software on people's computers in order to find more jobs to eliminate. The surface area of problems in the world is not decreasing, but what is decreasing is the number of humans to take on the next opportunity. The companies that get rewarded — their stock prices rewarded — are the ones that do much more with much less. They're going to keep eliminating these jobs, and I take them at their word."

「我认为你错了,让我解释为什么。我相信 Zuckerberg 在员工电脑上装那个软件,就是为了找出更多可以被消灭的岗位。世界上待解决问题的总面积并没有在减少,真正在减少的,是去承接下一个机会所需要的人手。那些被奖励的公司——股价被奖励的公司——正是用远少的人做远多的事的那一批。他们会继续消灭这些岗位,而我听其言、信其言。」

"You don't have to explain everything with conspiracy. Maybe they just mismanaged for a period. That explains the post-COVID two or three years."

「你不必把每件事都用阴谋论来解释。也许他们就是有那么一段时间管理失当而已。后疫情那两三年,用这个就能解释。」

"I think what we're seeing this year is the tools working. There are jobs no longer needed — the 'measurers,' as Matthew Prince put it, or product managers or designers, those have been consolidated into one job. Somebody who ships a product, and it doesn't require 12 people, it requires two now."

「我觉得我们今年看到的,是工具真的开始起作用了。有些岗位不再需要——Matthew Prince 说的那些『度量员』(measurers),或者产品经理、设计师,这些都被合并成了一个岗位。一个能把产品发出去的人,不再需要 12 个人,现在两个人就够了。」

"I don't think that's been consolidated. I see it in Fortune 1000 companies all the time. You're talking about the slowest adopters."

「我不认为那已经被合并了。我在 Fortune 1000 的公司里天天看到[相反的情况]。你说的是最慢的那批采用者。」

"I'm on the front line with startups — that's where all the jobs are."

「我在创业公司的第一线——所有的岗位都在那儿。」

"But a startup is not going to enter a regulated market and put JP Morgan out of business. Not going to happen."

「可一家创业公司不会进入一个受监管的市场、把 JP Morgan 干倒闭。不会发生。」

"They will eventually displace those companies. It happens all the time."

「他们最终会取代那些公司。这种事一直在发生。」

"Not going to happen. We're going to agree to disagree. Good luck to the startup trying to disrupt Boeing. Good luck — I'm going to take Boeing."

「不会发生。这点我们就求同存异吧。祝那家想颠覆 Boeing 的创业公司好运。祝你好运——我押 Boeing。」

"Well, some people might take SpaceX."

「呃,有些人可能会押 SpaceX。」

"Good luck making drugs out of an Excel spreadsheet — I'll take the regulated pharma company. There are some industries where you're going to show up at the FDA like, okay, where's your team? Oh, it's just me, I do it all."

「祝你用一张 Excel 表格造出药来——那我押那家受监管的制药公司。有些行业,你跑去 FDA,人家问:好,你的团队呢?哦,就我一个,全是我干的。」

"You joke — somebody just did that. 'Me and my model.'"

「你别笑——刚就有人这么干了。『我和我的模型。』」

"It's not a joke, and it's not going to happen — because that's not the way society wants safety, predictability, governance, and auditability to work."

「这不是玩笑,而且它不会发生——因为社会想要的安全、可预测性、治理和可审计性,不是这么运作的。」

"There's a distinct difference between drugs and software and services in the world — I think we can agree on that. And, listen, truck driving is one of the most regulated industries out there, so is cab driving, and those jobs are being eliminated."

「在这个世界里,药品和软件、服务之间有一个明确的区别——这点我想我们能达成共识。但听着,卡车运输是监管最严的行业之一,出租车也是,而这些岗位正在被消灭。」

"First of all, Jake, you remind me of the Trotskyite who, when confronted with the fact that none of Trotsky's predictions had come true, simply argued that proved how far-sighted Trotsky was. In other words, none of your predictions about job loss have come true."

「首先,Jake,你让我想起那个托洛茨基主义者(Trotskyite):当有人指出托洛茨基的预言一个都没应验时,他反而辩称这恰恰证明了托洛茨基有多高瞻远瞩。换句话说,你关于失业的预测,一个都没成真。」

"I didn't go to graduate school — you're going to need another reference. None of my predictions? Zero data — except for what Meta just did last week."

「我没读过研究生——你得换个我能听懂的典故。我的预测一个都没应验?零数据——除了 Meta 上周刚干的那一票。」

"That's an anecdote. Those Meta job losses were not directly attributable to AI — Zuckerberg clarified it was about balancing additional capex spending. But even if it were 100% due to AI, you're not netting those jobs against all the jobs being created because of AI, and all the new companies. You're cherry-picking one statistic, attributing 100% to AI, and not presenting a balance."

「那是一桩轶事。Meta 那些裁员并不能直接归因于 AI——Zuckerberg 已经澄清,那是为了平衡额外的资本开支(capex)。但就算它 100% 是 AI 造成的,你也没有把这些岗位拿去和所有因 AI 而被创造的岗位、所有新公司对冲(net)。你是在挑一个数据樱桃,把它 100% 归给 AI,却不呈现一张平衡账。」

"I'm not cherry-picking — I'm reading the news and describing what the CEO said. Jack at Block said he's doing it because of AI. Matthew Prince said it's AI. Zuckerberg said it's AI. I'm just taking them at their word."

「我没在挑樱桃——我是在读新闻,在转述 CEO 们自己说的话。Block 的 Jack 说他这么做是因为 AI。Matthew Prince 说是 AI。Zuckerberg 说是 AI。我只是听其言、信其言。」

"Jack Dorsey said he'd do a 50% elimination because of AI. Within 24 hours, all the financial analysts on X said Jack was AI washing — Block had horribly overstaffed during COVID and was running more inefficiently than all its peers. They've needed a 50% cut for a long time. Pretty much everyone thought it was pure AI washing. What exactly are the efficiencies Jack is getting? 'Oh, we're just magically going to eliminate half our cost structure' — this is the most hand-wavy thing ever."

「Jack Dorsey 说他要因为 AI 裁掉 50%。24 小时之内,X 上所有金融分析师都说 Jack 在 AI washing——Block 在疫情期间严重超编,运营效率比同类公司都差。他们早就该砍 50% 了。几乎所有人都认为那纯粹是 AI washing。Jack 到底拿到了什么效率?『哦,我们就这么神奇地一下砍掉一半成本结构』——这是史上最含糊其辞的说法。」

"We currently have a 4.3% unemployment rate. Economists consider 5% full employment. So unemployment is at or near record lows, despite being over three years into this AI wave. Coding is the single job category most impacted by AI — we have almost complete automation of code-writing. You'd think, in a simple Malthusian way, all the software developers would be getting laid off. Is that happening? No. Job postings for software developers are at a three-year high, up 15% year-over-year."

「我们现在的失业率是 4.3%。经济学家把 5% 视为充分就业。所以失业率正处在、或接近历史低点——尽管我们已经踏进这波 AI 浪潮三年多了。编程是受 AI 影响最大的单一岗位类别——写代码几乎已经被完全自动化。你会以为,按简单的马尔萨斯式逻辑,软件开发者应该正在被成批裁掉。有吗?没有。软件开发岗位的招聘需求创下三年新高,同比增长 15%。」

"Why? Look at code commits on GitHub: 1 billion last year; in the past month, 1.1 billion. That's a 14x year-over-year increase in code generation. That code has to be managed by somebody — you still need humans to look under the hood. When the amount of code explodes 10x or 100x, the complexity rises too. We're not hiring 10x more engineers, but you do need more to manage all that code."

「为什么?看 GitHub 上的代码提交(commits):去年是 10 亿次;就在过去一个月,是 11 亿次。这是代码生成量同比 14 倍的增长。这些代码总得有人来管——你仍然需要人钻到引擎盖底下去看。当代码量暴增 10 倍、100 倍,复杂度也跟着上升。我们不会多招 10 倍的工程师,但你确实需要更多人来管理这么多代码。」

"I'm hearing from people hiring software engineers who never would have before. A fund manager told me his next two hires weren't going to be data analysts — they were going to be software developers, because they're deploying code for the first time. This goes back to Claude proficiency being the most marketable skill. We're at the outset of a boom caused by bespoke software proliferating, used by firms that never thought of themselves as tech firms — and that doesn't even include the blue-collar boom: data centers, new energy, hundreds of thousands of construction jobs."

「我听到一些从前根本不会招软件工程师的人,现在开始招了。一位基金经理告诉我,他接下来两个招聘名额,不再是数据分析师了——而是软件开发者,因为他们第一次开始部署代码。这又回到了那个判断:精通 Claude 是当下最有市场价值的技能。我们正处在一场繁荣的起点,它由定制软件(bespoke software)的扩散驱动,被那些从不认为自己是科技公司的企业使用——而这还没算上蓝领的繁荣:数据中心、新能源,几十万个建筑岗位。」

"I never advocated otherwise — you like to misconstrue my position. I'm very clear: there's job displacement going on, and it's related to AI, but net, I do think the economy will grow."

「我从没主张过相反的东西——你就爱曲解我的立场。我说得很清楚:岗位置换正在发生,而且和 AI 有关,但净额上,我确实认为经济会增长。」

"Maybe at some point in the future you'll be right — like the Trotskyite: communism has never been tried, maybe it'll work."

「也许未来某个时点你会是对的——就像那个托洛茨基主义者说的:共产主义从来没被真正试过,也许它会成功呢。」

"Nobody knows your Trotsky references. You lost 95% of the audience. Just speak like a normal person."

「没人懂你那些托洛茨基的典故。你已经把 95% 的听众弄丢了。能不能像个正常人那样说话。」

"Bill Gurley — of 20 million people in the United States driving cabs and trucks right now, how many do you think will lose their jobs to self-driving in the next decade or two? I'm not trying to lead the witness. Obviously some people prefer a human driver."

「Bill Gurley——美国现在有 2,000 万人以开出租车、开卡车为生,你觉得未来一二十年里,有多少人会因为自动驾驶而失去工作?我不是想给证人下套。显然有些人还是更喜欢人类司机。」

"I think it's impossible to go 100% automated, because the economics don't work well. Like ATMs, I think the use of non-ownership cars is going to go way up. So it'll keep growing, and humans will be used for like 50% of it instead of 100%. I wouldn't be surprised if the number actually stays the same or grows. And remember, these jobs didn't exist before because regulation had limited the taxi market — getting around that actually led to job creation."

「我觉得做到 100% 自动化是不可能的,因为经济账算不过来。就像 ATM 一样,我认为『非自有车辆』(non-ownership cars)的使用会大幅上升。所以这块会持续增长,而人类会被用在其中大概 50% 的部分,而不是 100%。如果这个数字其实保持不变、甚至增长,我都不会意外。而且别忘了,这些岗位从前之所以不存在,是因为监管限制了出租车市场的规模——绕过那道限制,反倒催生了就业。」

"I'm not a big fan of the doomerism around jobs — there's a word, 'Luddite,' used to describe it. And I don't have high confidence in any government program for skills retraining. So what do we do now? It's not clear to me. The most you can do: one, use the new tools, know what they're capable of in your field. Two, if your job's going away — and maybe it's a job you don't care about — start thinking about where the opportunities are. The skilled trades are short of people: plumbing, electricians, HVAC."

「我不太认同围绕就业的那套末日论——有个词叫『卢德分子』(Luddite)就是形容这个的。而且我对任何政府主导的技能再培训项目都没什么信心。那我们现在到底该怎么办?我也说不清。你最多能做两件事:第一,用上这些新工具,搞清楚它们在你所在领域能做到什么。第二,如果你的工作要没了——而且也许那本就是份你并不在乎的工作——开始想想机会在哪儿。技术工种正缺人:管道工、电工、暖通空调(HVAC)。」

"It's amazing how Jay uses facts that haven't happened yet as support for his argument. You just state that all the truck drivers are losing their jobs, and then say this proves your take. I know it's your belief — but that is not proof. Do you understand?"

「Jay 用『还没发生的事实』来支撑自己的论点,真是叹为观止。你就这么断言,所有卡车司机都在失去工作,然后说这证明了你的判断。我知道这是你的信念——但它不是证据。你明白吗?」

"The proof I gave was Amazon and Andy Jassy, Shopify. Amazon has a self-driving division — it's called Zoox. They're the largest user of robotics in the world. So yes, Chamath, they're pursuing robotics and self-driving massively more than anybody. Self-driving is going to take away, I believe, the majority of [those driving jobs]."

「我给的证据是亚马逊和 Andy Jassy,还有 Shopify。亚马逊有一个自动驾驶部门——叫 Zoox。他们是全世界最大的机器人使用者。所以是的,Chamath,他们在机器人和自动驾驶上的投入,比任何人都大得多。自动驾驶会拿走——我相信——其中大部分[开车的岗位]。」

"That's the key word — 'believe.' Great. Let's put it there as a belief. Who knows? You don't know and I don't know. You just string all these words together — at one point it's a warehouse worker, then a driver, then Amazon."

「那就是关键词——『相信』(believe)。很好。我们就把它摆在那儿,当作一个信念。谁知道呢?你不知道,我也不知道。你就是把这些词串到一起——一会儿是仓库工人,一会儿是司机,一会儿又是亚马逊。」

"I have a legit question for you. Explain to me why job postings for software engineers are up 15% year-over-year, despite the fact that code has now been fully automated."

「我有个正经问题问你。给我解释一下,为什么软件工程师的招聘需求同比涨了 15%,尽管代码现在已经被完全自动化了?」

"I think there's a Cambrian explosion in software — you're absolutely correct. People who know how to vibe code, or non-developers making bespoke software. So absolutely, that will be an area of job growth. The positions being removed are product managers, middle managers, what Matthew Prince calls 'measurers.' Everybody believes the standups and zoom calls — people management — is being done better by AI, and the stack of people to build products is being consolidated. The designer can now vibe code, the developer can do front-end design and UX and project-manage themselves."

「我认为软件领域正在发生寒武纪大爆发——你说得完全对。会 vibe code 的人,或者本来不是开发者、却在做定制软件的人。所以毫无疑问,这会是一个就业增长的领域。被砍掉的岗位是产品经理、中层管理者,也就是 Matthew Prince 说的『度量员』。大家都相信,那些站会、Zoom 会议——也就是『管人』这件事——AI 做得更好,而搭建产品所需的那一摞人正在被压缩。设计师现在能 vibe code,开发者能做前端设计和 UX,还能自己给自己当项目经理。」

"On net — do you think there'll be mass job loss?"

「净额上——你认为会出现大规模失业吗?」

"I think there's a chance we see job loss increase in the short to mid term, and then eventually the displaced people have to learn or leave the workforce — which is what happened during other revolutions. Some people went with the paradigm and adapted; some didn't and just retired. I saw it firsthand in the PC revolution — some lawyers would never use the tools and just retired at 55 or 65; other attorneys were PC-first and took it."

「我认为短中期内,失业有可能上升,然后最终,被置换掉的人要么去学新东西,要么离开劳动力市场——这正是历次类似革命里发生过的事。有些人顺应了新范式、适应了;有些人没有,就退休了。PC 革命时我亲眼见过——有些律师死活不肯用这些工具,55 岁、65 岁就退了;另一些律师是 PC-first 的,就把它接住了。」

"Did you guys see that Kirkland & Ellis is going to spend half a billion dollars to roll their own frontier model? Makes total sense — back to our earlier point, people are going on-prem and making their own models. Bill, when Andy Jassy said 'we're going to do more with less, be AI-first, we're not going to hire these 600,000,' and Tobi Lütke said 'you have to prove you tried AI first before asking for headcount' — is that a sign these organizations are AI washing, or genuinely doing more with less?"

「你们看到没,Kirkland & Ellis(美国顶级律所)要花 5 亿美元自建一个 frontier 模型?完全说得通——回到我们前面的观点,大家都在走本地部署、做自己的模型。Bill,当 Andy Jassy 说『我们要用更少做更多、要 AI-first、我们不会去招这 60 万人』,Tobi Lütke 说『要先证明你试过 AI,才能来要人头』——这是这些组织在 AI washing 的信号,还是真的在用更少做更多?」

"What that question misses — and a lot of the AI doomerism misses — is that competition exists. There's no scenario where you do more for less and suddenly everyone has 70% operating margins. That's not going to happen. Someone else will come along and do more for less and lower the price. So we could have a productivity boom from lower-priced goods and services, and the basket of goods humans can buy gets cheaper. Unfortunately it's offset by healthcare and other regulated industries. But I expect products to get cheaper — I don't think it leads to obscene profits, because it'll be whittled away in competition."

「那个问题没看到的——很多 AI 末日论也没看到的——是『竞争是存在的』。不存在那样一种情形:你用更少做更多,然后突然之间人人都有 70% 的营业利润率。那不会发生。总会有别人冒出来,用更少做更多,把价格压下去。所以我们可能会迎来一场由低价商品和服务驱动的生产力繁荣,人类能买到的那一篮子商品越来越便宜。不幸的是,这会被医疗和其他受监管行业抵消掉一部分。但我预期产品会变便宜——我不认为这会带来畸高的利润,因为它会在竞争中被一点点磨掉。」

"Just on the AI washing point: there's a trial lawyer named Donnie King, a securities litigation partner, who's started to warn that we could see shareholder lawsuits against companies that engage in AI washing — he thinks it's a type of puffery. Because what the company is doing is attributing its own non-performance or operational issues to AI, when in fact there are real problems in the business. So it could be a form of securities fraud."

「就 AI washing 这点补一句:有位出庭律师叫 Donnie King,是做证券诉讼的合伙人,他已经开始警告:我们可能会看到针对那些搞 AI washing 的公司的股东诉讼——他认为这是一种『吹嘘』(puffery)。因为这些公司实际上是在把自己的业绩不振、运营问题甩锅给 AI,而事实上,业务里有真实的问题存在。所以它可能构成一种证券欺诈(securities fraud)。」

"I want to double-click on this. Did you see the CEO of Wix today, in his note about layoffs? Find me the AI washing in there. Interesting that he just talked about operational details."

「我想就这点深挖一下(double-click)。你看到今天 Wix 的 CEO 那封关于裁员的备忘录没?你在里面给我找找哪儿有 AI washing。有意思的是,他通篇只谈了运营层面的细节。」

"Did he? I didn't read the note."

「是吗?那封备忘录我没读。」

"Of course you didn't read the note."

「你当然没读那封备忘录。」

"Well, you said it just happened — it broke at 9:25 this morning. I'll read it."

「你不是说它刚发生嘛——今早 9:25 才爆出来的。我会读的。」

"Look, Jcal, you're the last person who hasn't gotten the memo. There was a huge narrative shift this week — Sam's backing off, even Dario's walking it back, you've got the Goldman Sachs CEO, the explosion in job postings. Everyone's coming around to the idea that the job apocalypse is massively overblown. Your apology, I'm happy to accept."

「听着,JCAL,你是最后一个还没收到风声的人。这周出现了一次巨大的叙事转向——Sam 在退场,连 Dario 都在收回,你还有 Goldman Sachs 的 CEO、招聘需求的暴涨。所有人都开始接受:失业末日被严重夸大了。你的道歉,我乐意收下。」

"It could be overblown looking at the stats. But my position has always been: it's displacement. You'll have some displaced in the short to mid term, and then eventually there'll be more problems to solve and people will reallocate. I do think the tech industry doesn't have enough empathy or thoughtfulness when discussing this — these are real people losing real jobs. You can point at statistics, but these are real people who may not make the transition."

「从数据看,它可能确实被夸大了。但我的立场一直是:这是置换(displacement)。短中期会有一部分人被置换,然后最终会出现更多待解决的问题,人们会重新配置。我确实认为,科技圈在讨论这件事时缺乏足够的共情和体恤——这些是活生生的人在失去活生生的工作。你可以指着统计数字,但这些是真实的人,他们可能完成不了这个过渡。」

"Do you think it's more empathetic to be scaring the bejesus out of people that they're going to lose [their jobs]?"

「你觉得把人吓得魂飞魄散、说他们要丢[工作]了,就更有共情吗?」

"I'm not in the scaring camp — I'm in the enabling camp. That's why I keep saying: if you've been laid off, start a company and embrace the tools. If they learn the tools, they'll have 10 job offers and they'll start their own companies. So there's a solution. I just think we're going to go through low-millions of jobs being lost, retired, or transitioned out over the next couple years."

「我不在『吓人』那一阵营——我在『赋能』那一阵营。这就是为什么我一直说:如果你被裁了,去创业,去拥抱这些工具。只要学会这些工具,他们会有 10 份 offer,会去开自己的公司。所以是有解的。我只是觉得,未来这一两年,我们会经历『低个位数百万级』的岗位被裁掉、退役或转型。」

"I was going to be empathetic and offer some solutions. We talked about the skilled-trades deficit. Micro1 has a foundation called Micro Works — they funded $16 million, 2,600 people get a free scholarship to become a plumber, welder, or electrician. So go check that out if you want to reskill. I think it's better than having the government fix things."

「我本来想做点共情,顺便给几个解法。我们聊过技术工种的人才缺口。Micro1 有个基金会叫 Micro Works——他们投了 1,600 万美元,资助 2,600 人免费拿奖学金去当管道工、焊工或电工。所以如果你想转技能,去看看这个。我觉得这比让政府来收拾要好。」

"'Generation Tool Belt,' yeah. And as we started talking about, I've got a new grant program myself to help people tilt their career in a different direction. Go do something you love, apply, and maybe I can help fund you moving in that direction."

「『工具腰带一代』(Generation Tool Belt),没错。而且就像我们开头聊的,我自己也有个新的资助项目,帮人把职业生涯拨向一个不同的方向。去做一件你热爱的事,来申请,也许我能帮你资助这个转向。」

"And as to your vibe shift — I think it's because people's houses have been Molotov-cocktailed over the doomerism, and people are specifically citing that when they throw Molotov cocktails. If you're IPO-ing and coming out saying 'jobs are going away, jobs are going away,' that's just a really bad look."

「至于你说的那个氛围转向——我觉得是因为有人的房子因为这套末日论被扔了燃烧瓶(Molotov cocktail),而且扔的人专门拿这个当理由。如果你正要 IPO,却出来一直喊『工作要没了、工作要没了』,那这形象实在太糟糕了。」

"All right, everybody. This has been another amazing episode of the All-In podcast. Sorry — I need to do one thing at the end. I just want to give a huge shout-out to Tulsi Gabbard, a friend of ours, and specifically her husband Abraham, who's going through some really tough stuff with cancer. He's going to kick its ass. We love you."

「好了,各位。这又是一期精彩的 All-In 播客。抱歉——结尾我得补一件事。我想特别给我们的朋友 Tulsi Gabbard 送上一个大大的祝福,尤其是她的丈夫 Abraham,他正在和癌症做艰苦的斗争。他一定会把它打趴下。我们爱你们。」

"Cheers, everybody. That's episode 275 in the can. We'll see you next time. Bye-bye. Love you, Besties."

「干杯,各位。第 275 期就此收工。下次再见。拜拜。爱你们,Besties。」