SpaceX's $2T Case, Nvidia's Shock Selloff, America Turns on AI, Trump Pulls AI Order, Bond Crisis?

All right, everybody. 好，各位。 Welcome back to the number one podcast in the world. 欢迎回到全球第一播客。 It's the Allin podcast, episode 274. 这里是 All-In Podcast，第 274 期。 Sachs is out today, but we're very lucky to have Gavin Baker from Treaties Management joining us. Sacks 今天缺席，但我们非常幸运，Gavin Baker 来自 Atreides Management，今天加入我们。 The spicy takes must flow. 火药味十足的观点必须照常开炮。 Welcome back to the program. 欢迎回来。 Besty Gavin, Bestie Gavin， thanks for having me. 感谢邀请。 Always love it. 每次都很开心。 It's been a huge week in tech. 科技圈这周太热闹了。 We can start with the SpaceX and OpenAI IPOs. 可以从 SpaceX 和 OpenAI 的 IPO 聊起。 We've got Andre Karpathy joining Anthropic. Andrej Karpathy 加入了 Anthropic。 Nvidia crushing it. Nvidia 势头强劲。 So many different places to go, but I think we'll start with Andre Carpathy joining Anthropic. 话题太多，但我想先从 Karpathy 加入 Anthropic 说起。 Carpathy is only 39 years old. Karpathy 才 39 岁。 He's already a legend in the tech industry if you don't know him. 如果你不了解他，他已经是科技圈的传奇人物了。 I believe he's also coming to liquidity. 我相信他也会来 LAUNCH。 Yet, 还没有吧， he's going to keynote on Monday morning. 他周一上午会做主题演讲。 Oh, fantastic. 太棒了。 No, Tuesday. 不，是周二。 Tuesday. 周二。 Tuesday. 周二。 Day two. 第二天。 I think he's keing. 我觉得他会做主题演讲。 Okay. 好。 As is Gavin. Gavin 也会去。 Gavin will be there. Gavin 也会在场。 Founding member of 创始成员之 Gavin anchoring day two as Gavin 作为压轴嘉宾主持第二天 excellent. 很好。 Yeah, this is Gavin's second appearance at 这是 Gavin 第二次出席 Look at those two bookmarks. 看看这两位书签。 Andre Carpathy and Gavin Baker. Andrej Karpathy 和 Gavin Baker。 Oh, yeah. 哦，对。 You know, 你知道， liquidity pulls in the stars. 流动性吸引明星。 Obviously, Andre was a founding member of OpenAI. Karpathy 是 OpenAI 的创始成员之一。 He led the self-driving team. 他领导了自动驾驶团队。 Also, hold on. 等一下。 Gavin is going to help us judge the best ideas section as well. Gavin 也会帮我们评判最佳创意环节。 Excellent. 太好了。 I don't know if you know that, Gavin, but you're a judge. Gavin，你可能还不知道，你是评委。 You're a judge. 你是评委。 I'm up for anything, man. 什么都行，我随时奉陪。 I'm easy. 我好说话。 Yes. 是的。 Kurpathy also coined the term vibe coding. Karpathy 还创造了「vibe coding」这个词。 He recently built auto research. 他最近做了个自动研究工具。 I think we talked about that here a bit. 我们节目里也聊过。 That's an open source training tool. 这是个开源训练工具。 It helps AI models improve themselves by running five-minute experiments. 它让 AI 模型通过运行五分钟实验来自我提升。 That got over 82,000 stars on GitHub. 在 GitHub 上获得了超过 82,000 个 star。 He did that like as a weekend experiment and all these civilians started building their own uh recursive LLMs. 他就当个周末实验做了，结果一堆人开始自己搭递归 LLM。 Really inspiring. 真的很励志。 And the Andre Karpathy skills is a tool based on his set of principles for claude code. Karpathy skills 是基于他为 Claude Code 制定的一套原则的工具。 And somebody just released that. 有人刚把它发布出来了。 And so that's just pretty crazy when you think about it. 这么想想其实挺疯狂的。 He's going to be in charge of a new pre-training team at Anthropic. 他将负责领导 Anthropic 新组建的预训练团队。 The focus obviously being recursive self-improvement. 重点显然是递归自我改进。 In other words, they're going to have Claude improve itself. 换句话说，他们要让Claude自我改进。 And they've already talked a little bit about AI improving AI over at anthropic. Anthropic早就谈过AI改进AI这件事。 What's your take on this? 你怎么看？ Is this uh super important in 2026? 2026年这件事超级重要吗？ Obviously, Karpathy is super well respected. Karpathy显然是业界顶尖人才，极受尊重。 is obviously uh you know one of the true talents in the space but hey we're in we're in a different inning than we were say 10 years ago when he was at Tesla or five years ago when he co-founded open AI 他无疑是这个领域真正的大才，但话说回来，现在已经是不同局面了不是他在Tesla那会儿的10年前，也不是他联合创立OpenAI的5年前。 you know what's interesting if you go back to like Google the culture of Google which they got right was the singular technical talents 说个有意思的，回头看Google，他们做对的一件事就是把顶尖的单一技术天才单独拎出来， there they were singled out and they were called Google fellows 这些人被专门筛选出来，称为Google Fellow。 I don't know if you guys remember this like Ahmed Single 不知道你们还记不记得，像Ahmed Singhal， Shredar Ramaswami Sridhar Ramaswamy， Jeff Dean these guys are stars. Jeff Dean，这些人都是巨星。 And what's interesting is if you track what folks, particularly Jeff Dean, I guess, now because the other two aren't there anymore, but what they did inside of Google, it's like wave upon wave, they were at the foot of those waves. 有意思的是，如果你追踪这些人尤其是Jeff Dean，现在另外两个已经不在了他们在Google内部的轨迹，简直是一浪接一浪，他们每次都踩在浪头上。 What's interesting about Andre is he's been at the wave upon wave of AI. Andrej有意思的地方，就在于他也是这样一浪接一浪地紧跟AI每次浪潮。 He was probably the first person that really commercialized the Richard Sutton bitter lesson essay when he was leading FSD at Tesla which was really about the brute force computation and I remember him telling me this story I don't know if he said this publicly or not but where he spent a portion of his time I want to say a quarter of his time labeling data could you imagine like 201617 like handlabeling video data from Teslas 他很可能是第一个把Richard Sutton的bitter lesson文章真正商业化的人，那是他在Tesla主导FSD的时候，核心就是蛮力计算。我记得他跟我讲过一个故事，不知道他有没有公开说过，就是他花了相当一部分时间我猜大概四分之一的时间在亲手标注数据。你能想象吗，2016、2017年，亲手给Tesla的视频数据打标签。 So he did that then he co-founder of open AI he's a star and he's an exceptional human being and he's super curious and then what he's done as a kind of a free agent is also quite impressive. 他做完这些，又联合创立了OpenAI，是真正的巨星，也是极其出色的人，极度好奇，而他作为自由人所做的事同样相当亮眼。 So I think that this is a really important deal. 所以我觉得这次加入意义非常重大。 I think he's one of these really curious people that can be sent off and they'll just go and invent new things. 他就是那种极度好奇的人，放出去就会自己去发明新东西。 And I think this idea of recursive self-learning puts these models on a combination of overdrive and autopilot. 我认为递归自我学习这个想法，会让这些模型同时进入超速档和自动驾驶模式。 And so if you put those two things together, I think that you start to you can potentially live out this idea that there's an order of magnitude improvement on a yearly basis. 把这两点结合起来，我认为你就可以开始实现一个构想模型每年都有数量级级别的提升。 So like this new form of Moore's law. 有点像AI版摩尔定律。 So then the model quality just goes absolutely parabolically just like this straight up. 模型质量就会绝对地抛物线式直线拉升。 throwing a bunch of compute at the problem and these things learn really quick. 往问题里砸一堆算力，这些东西学得超快。 I think is the high order bit there. 我觉得这才是最关键的。 Gavin, what are you Gavin，你怎么 what's your take on 你怎么看 Anthropic's recent success and their massive hiring binge? Anthropic近期的成功和大规模招聘？ The success is extraordinary. 成功是毋庸置疑的， It's undeniable. 这无可辩驳。 I think the fact that they are now they were EBIT positive per the Wall Street Journal in the most recent quarter is a really important fact for kind of the whole AI narrative 我认为一个非常重要的事实是，据 Wall Street Journal 报道，他们上个季度已经实现 EBIT 转正，这对整个 AI 叙事来说意义重大， because now there's you know you could talk about circular funding 因为现在，你知道，有人说这是循环融资， you could talk about ROI and we could go look at the ROIC of the hyperscalers but if OpenAI and Enthropic are at call it a hundred billion dollars of ARR now with 80%ish gross margins on inference like the returns are there and then if we add in and they're growing really fast 可以聊聊ROI，也可以去看看超大规模厂商的ROIC，但如果OpenAI和Anthropic现在ARR已经到了大概1000亿美元，推理业务毛利率80%左右，回报就在那里，而且它们还在飞速增长， if we add in Gemini we add in cursor we add in XAI we add in open source 再加上Gemini、Cursor、xAI，再加上开源， you know it's it's not hard to see 200 300 400 billion of ARR at the end of this year at high market 不难想象今年底整个行业ARR能到2000亿、3000亿、4000亿，估值倍数还很高。 across all of the language models and you're talking specifically about the private language model companies maybe not Google which is including 横跨所有语言模型，而且你说的主要是私有语言模型公司，可能不包括Google， But I 但我 was excluding you know a lot of the returns to this GPU spend have come from 没算那部分，GPU投入带来的大量回报其实来自于 you know better recommener systems at Facebook and Google Facebook和Google更好的推荐系统， Amazon better ad targeting better ad measurement. Amazon更精准的广告定向和广告效果衡量。 Sure. 对。 So I was excluding that and just narrowing it to LLMs which I think strictest possible definition and it seems like there's going to be a really strong ROI this year even excluding what are still some of the most economically important and profitable use cases for GPUs and AI infrastructure. 所以我把那部分排除掉，只看LLM，这是最严格的定义，今年ROI看起来会很强劲，即便排除了GPU和AI基础设施里一些经济价值最高、最赚钱的应用场景。