Update: he now just called him "the little maestro" and "little Leo"
更新：他现在就只是叫他“the little maestro”和“little Leo”。
MT
Matt Turck
@mattturck ↗VC at @FirstMarkCap. Host: MAD Podcast; Organizer: Data Driven NYC, Author: MAD Landscape.
2 最新30 累计22 期
每条推文 hover 显示单独 ▶
You can be Messi, the best footballer ever, and some rando commentator on Fox will still call you "the little magic man"
你可以是 Messi，史上最伟大的足球运动员，而 Fox 上某个路人解说员还是会叫你“the little magic man”。
When your VC is exhausted from attending the board meeting
当你的 VC 因参加董事会会议而精疲力尽时
This awesome conversation with @stephenbalaban of @LambdaAPI is also available on Spotify, Apple Podcasts and here on YouTube:
这场与 @LambdaAPI 的 @stephenbalaban 的精彩对话，也可以在 Spotify、Apple Podcasts，以及这里的 YouTube 上收听/观看：
State of AI compute 2026: my conversation with @stephenbalaban of @LambdaAPI on the neocloud boom, data centers, GPUs and what's ahead 00:00 — Cold open 01:21 — Why GPU compute was never a commodity 02:45 — The H100 price index and what it gets wrong 04:02 — The real moat: technology or financing? 05:57 — Winner-take-all, or room for many neoclouds 06:48 — Are we overbuilding or underbuilding AI compute? 09:26 — What if AI gets 10x more compute-efficient? 10:44 — The real bottleneck: land, power, and shell 11:38 — The backlash against data centers — and the misinformation 15:00 — Opening the hood: from photons to tokens 17:11 — Extracting more value from the same chip 19:26 — Frontier inference and distributed training, explained 23:26 — What actually drives compute cost 25:21 — Lambda's chip stack and the NVIDIA relationship 26:17 — A multi-silicon world? CUDA, CUDNN, and NVIDIA's real moat 28:59 — Networking, storage, and the one-click cluster 34:46 — Renting vs. owning, and full vertical integration 36:24 — How global is Lambda? Does location still matter? 38:44 — The financing stack: off-take agreements, SPVs, and credit 41:16 — Why a 2023 GPU leases for more today 42:36 — A futures market for compute? 43:54 — Origin story: facial recognition, Perceptio, and Apple 47:03 — The Lambda hat and Dream Scope 48:59 — The $60K bet that became a cloud business 52:00 — Holding the team together through the hard times 54:30 — Bringing on a new CEO; Stephen as CTO 57:33 — Matching xAI on high-velocity deployment 59:29 — "AI won't write software — it will become the software" 01:01:30 — Neural software vs. vibe coding 01:04:25 — Do agents change the compute layer 01:06:14 — Self-assembling software inside Lambda 01:08:18 — Gigawatt-scale AI factories 01:08:57 — One person, one GPU 01:12:04 — Hot takes: overrated and underrated in AI
AI compute（算力）2026 现状：我与 @LambdaAPI 的 @stephenbalaban 关于 neocloud 热潮、data centers（数据中心）、GPUs 以及未来走向的对话 00:00 — 开场冷启 01:21 — 为什么 GPU 算力从来不是一种 commodity（标准化商品） 02:45 — H100 价格指数，以及它错在哪里 04:02 — 真正的 moat（护城河）：技术还是融资？ 05:57 — 赢者通吃，还是会有许多 neocloud 并存的空间 06:48 — 我们是在过度建设还是建设不足 AI 算力？ 09:26 — 如果 AI 的算力效率提高 10 倍会怎样？ 10:44 — 真正的瓶颈：土地、电力和 shell（机房壳体/基础设施） 11:38 — 针对 data centers 的反弹情绪，以及其中的错误信息 15:00 — 打开引擎盖：从 photons（光子）到 tokens 17:11 — 如何从同一块 chip（芯片）中榨取更多价值 19:26 — Frontier inference（前沿推理）与 distributed training（分布式训练）解析 23:26 — 真正驱动算力成本的因素是什么 25:21 — Lambda 的 chip stack（芯片栈）以及与 NVIDIA 的关系 26:17 — 一个 multi-silicon（多芯片/多架构）世界？CUDA、CUDNN，以及 NVIDIA 真正的 moat 28:59 — 网络、存储，以及 one-click cluster（一键集群） 34:46 — 租用 vs. 拥有，以及完全垂直整合 36:24 — Lambda 的全球化程度如何？地理位置还重要吗？ 38:44 — 融资结构：off-take agreements（包销协议）、SPVs（特殊目的实体）和信贷 41:16 — 为什么一块 2023 年的 GPU 今天租价更高 42:36 — 算力会有 futures market（期货市场）吗？ 43:54 — 起源故事：facial recognition（人脸识别）、Perceptio 和 Apple 47:03 — Lambda 帽子与 Dream Scope 48:59 — 那笔 6 万美元的赌注，如何变成了一门 cloud（云）业务 52:00 — 在艰难时期让团队保持团结 54:30 — 引入新 CEO；Stephen 担任 CTO 57:33 — 如何在高速部署上匹配 xAI 59:29 — “AI 不会编写软件——它会成为软件本身” 01:01:30 — Neural software（神经软件）vs. vibe coding 01:04:25 — agents（智能体）会改变 compute layer（算力层）吗 01:06:14 — Lambda 内部可自组装的软件 01:08:18 — 吉瓦级 AI 工厂 01:08:57 — 一人一 GPU 01:12:04 — 犀利观点：AI 里哪些被高估了，哪些被低估了
Key lesson: don't ignore your DMs on LinkedIn (he got recruited to the national Cabo Verde team via a couple of LinkedIn messages) Probably some lessons for B2B sales as well
关键教训：别忽视你在 LinkedIn 上的私信（他就是通过几条 LinkedIn 消息被招募进了 Cabo Verde 国家队）。这对 B2B 销售大概也有一些启示。
2026 is a BRUTAL grind in VC. You start in Davos, freeze in Aspen, hit Upfront, survive Milken, then it’s straight to Paris for the French Open. Briefly back in NYC for the Knicks. Then, total blur: SuperReturn in Berlin, Founders Forum in London, then back stateside for the World Cup, back to Paris for Raise AI, Idaho for Sun Valley, quick respite in Mykonos, then the Goldman tech gauntlet, Slush in Finland, NeurIPS in freakin’ Sydney… and *boom*, a productive year of thought leadership and adding value is over, and you’re a wreck.
2026 年做 VC 真是场残酷苦战。你从 Davos 开始，在 Aspen 冻得发僵，赶 Upfront，熬过 Milken，然后就直奔 Paris 看 French Open。短暂回到 NYC 看 Knicks。接着就彻底一片模糊：去 Berlin 参加 SuperReturn，去 London 参加 Founders Forum，然后回美国看 World Cup，再回 Paris 参加 Raise AI，去 Idaho 参加 Sun Valley，在 Mykonos 稍作喘息，接着闯过 Goldman 的科技马拉松，再去 Finland 参加 Slush，去他妈的 Sydney 参加 NeurIPS……然后 *boom*，一整年高产地输出 thought leadership（思想领导力）、持续 adding value（创造价值）就这么结束了，而你也彻底垮了。
Ok bad stories about VCs are spreading on X right now, but VCs have horror stories about founders too Like, that one time when a founder decided to take another term sheet with a higher valuation despite our obvious ability to add value, thought leadership and vendor discounts
好吧，现在 X 上关于 VCs 的糟糕故事正在传播，但 VCs 对 founders 也有各种恐怖故事。比如，有一次一位 founder 明明我们显然有能力提供 value、thought leadership 和 vendor discounts，却还是决定接受另一份 valuation 更高的 term sheet。
This great conversation with @danintheory of @OpenAI is also available on Spotify, Apple Podcasts and here on YouTube:
这场与 @OpenAI 的 @danintheory 的精彩对话，也可以在 Spotify、Apple Podcasts，以及这里的 YouTube 上收听或观看：
Why AI Can Now Make Discoveries - my conversation with @danintheory, Lead of the Foundations of Reinforcement Learning team at @OpenAI 00:00 Intro: AI's wild week in mathematics 01:21 What OpenAI's Foundations of RL team does 03:08 Dan's journey: from black holes and quantum gravity to frontier AI 07:04 Are AI systems becoming useful for real science 08:21 The AI math moment: Erdős, OpenAI, DeepMind, and Anthropic 08:52 Why the OpenAI result was an act of exploration 10:25 OpenAI vs. DeepMind: informal reasoning vs. formal proof 12:13 RL 101: learning by doing, not just watching 15:10 Why reinforcement learning works 15:58 How RL breaks: sparse feedback and long-horizon tasks 17:03 RLHF: how human feedback shaped early language models 18:48 Move 37, self-play, and the search for novel strategies 22:16 Explore vs. exploit in scientific discovery 24:49 Why RL may now be "the cake," not the cherry on top 25:46 Why RL started working with large language models 27:29 Is RL "sucking supervision through a straw"? 28:47 Why language may be the grounding layer for intelligence 31:46 A contrarian take on the Bitter Lesson 32:41 What test-time compute actually is 34:50 How RL gives models the ability to think 35:40 Verifiable rewards, math, coding, and the messy real world 38:00 What physics can teach us about AI 42:08 Is there a thermodynamics of AI? 43:08 From Erdős problems to Einstein-level AI 45:16 Is AI already doing original science? 45:51 How far are we from AI automating AI research 47:41 Why Dan is excited about the future of science
为什么 AI 现在能够做出发现——我与 @danintheory 的对话；他是 @OpenAI Foundations of Reinforcement Learning team 的负责人。00:00 开场：AI 在数学领域疯狂的一周 01:21 OpenAI 的 Foundations of RL 团队是做什么的 03:08 Dan 的经历：从 black holes 和 quantum gravity 到前沿 AI 07:04 AI 系统是否正在变得对真实科学有用 08:21 AI 的数学时刻：Erdős、OpenAI、DeepMind 和 Anthropic 08:52 为什么 OpenAI 的结果是一种探索行为 10:25 OpenAI vs. DeepMind：非形式化推理 vs. 形式化证明 12:13 RL 入门：通过实践学习，而不只是通过观察 15:10 为什么 reinforcement learning 有效 15:58 RL 如何失效：稀疏反馈与长时程任务 17:03 RLHF：human feedback 如何塑造了早期 language models 18:48 Move 37、self-play，以及对新颖策略的探索 22:16 科学发现中的 explore vs. exploit 24:49 为什么 RL 现在可能是“蛋糕本体”，而不只是顶上的樱桃 25:46 为什么 RL 开始在 large language models 上奏效 27:29 RL 是否是在“用吸管吸取 supervision”？28:47 为什么 language 可能是 intelligence 的 grounding layer 31:46 对 Bitter Lesson 的一个反常识观点 32:41 test-time compute 到底是什么 34:50 RL 如何赋予 models 思考的能力 35:40 可验证奖励、数学、coding，以及混乱的现实世界 38:00 physics 能教会我们关于 AI 的什么 42:08 AI 存在 thermodynamics 吗？43:08 从 Erdős 难题到 Einstein 级别的 AI 45:16 AI 是否已经在做原创科学 45:51 我们距离 AI 自动化 AI research 还有多远 47:41 为什么 Dan 对科学的未来感到兴奋
CEO: “we have tens of thousands of AI agents running in production at massive scale right now” CTO:
CEO：“我们现在已有数以万计的 AI agent（智能体）以极大规模在生产环境中运行。” CTO：
The biggest mindf*ck scenario in AI: things don’t change that much. Both doomers and accelerationists turn out to be wrong. We are all more productive. Agents deliver automation in the enterprise. Some important scientific discoveries are made. All great. But that’s it.
AI 中最让人脑子发懵的一种情景是：事情其实并没有发生多大变化。无论是 doomers 还是 accelerationists，最后都被证明是错的。我们的生产力都更高了。agent（智能体）在企业中实现了自动化。一些重要的科学发现也出现了。都很好。但也就仅此而已。
Genuinely impressive release by Google today (remember when they were behind?) Gemini 3.5 Flash perf: * Building on prior strengths (83.6% of MMMU-Pro for multimodal), * big jump on agentic coding (76.2% on Terminal-Bench for agentic coding and 56.5% on Toolathon for real world tasks) * progress and expert tasks (57.9% on Finance Agent 2... we are cooked) * leading scores across SWE-Bench, OSWorld etc. (also, elegant to bold the top scores in the chart below even if when it's not Google leading) Ofc, just benchmarks, and also not cheap (~$9/M output), but Google is cookin'... we are all so spoiled to have the 3 labs compete
Google 今天发布的东西确实令人印象深刻（还记得他们以前还落后吗？）Gemini 3.5 Flash 的性能：* 在既有优势上继续提升（多模态方面，MMMU-Pro 达到 83.6%），* agentic coding（代理式编程）有大幅跃升（在 Terminal-Bench 上做 agentic coding 达到 76.2%，在 Toolathon 上做现实世界任务达到 56.5%）* 在进展与专家级任务上也有提升（Finance Agent 2 上达到 57.9%……我们完了）* 在 SWE-Bench、OSWorld 等基准上也拿到领先分数（而且很巧妙的是，下面图表里把最高分都加粗了，即使领先的不是 Google）当然，这些终究只是 benchmarks（基准测试），而且价格也不便宜（输出约 ~$9/M），但 Google 确实火力全开……我们这些人真是太幸福了，能看到这 3 家实验室彼此竞争
Breaking: Anthropic attains sainthood, officially annointed by AI Jesus
突发：Anthropic 荣登圣坛，已获 AI Jesus 官方加冕为圣
When a multi-billion dollar venture fund has a $1B exit in its portfolio
当一家管理着数十亿美元的 venture fund（风投基金）在其投资组合中拥有一笔价值 $1B 的 exit（退出）时
Some personal news: He slid into my DMs, don't say prayers don't work y'all
一则个人消息：他私信我了，所以别再说祈祷没用了，大家
Alright it’s now official - barely 9 months old and @GradiumAI is already trouncing the entire voice AI field on third party TTS benchmarks Better than OpenAI Better than Eleven Labs Better than Cartesia Better than Deepgram …
好了，现在已经正式官宣了——成立还不到 9 个月，@GradiumAI 就已经在第三方 TTS（文本转语音）基准测试中横扫整个 voice AI 领域。比 OpenAI 更强，比 Eleven Labs 更强，比 Cartesia 更强，比 Deepgram 更强……
This great conversation with @ivanburazin is also available on Spotify, Apple Podcasts and here on YouTube (like and subscribe!):
这场与 @ivanburazin 的精彩对话，也可以在 Spotify、Apple Podcasts，以及这里的 YouTube 上收听/观看（点赞并订阅！）：
The more I think about AI agents, the less obvious it is that pricing goes purely consumption-based Token costs matter... but enterprise agents may need identities, roles, auth, budgets, audit logs etc That sounds oddly seat-like? just not human-seat-like
我越是思考 AI agents（智能体），就越觉得定价未必会纯粹走 consumption-based（按用量计费）这条路。Token 成本当然重要……但 enterprise agents（企业级智能体）可能还需要 identities（身份）、roles（角色）、auth（认证）、budgets（预算）、audit logs（审计日志）等等。这听起来有点像 seat-based（按席位）？只是不是那种 human-seat-like（面向人类席位）的形式。
This great conversation with @zicokolter is also available on Spotify, Apple Podcasts and here on YouTube:
这场与 @zicokolter 的精彩对谈也可在 Spotify、Apple Podcasts，以及这里的 YouTube 上观看/收听：
Deeply thoughtful conversation with @zicokolter, board member at @OpenAI and head of the machine learning department at @CarnegieMellon, about AI safety, AI security, agents and frontier AI 00:00 Intro 01:32 OpenAI board role and Safety & Security Committee 03:53 How OpenAI reviews major model releases 05:33 OpenAI’s preparedness framework explained 09:46 Are frontier AI models getting safer? 12:33 Why AI safety does not come from scale 15:23 The four categories of AI risk 19:38 Doomerism vs accelerationism in AI 24:11 The six-month AI pause debate 26:20 AI safety as a global effort 28:04 How Zico Kolter got into machine learning 31:05 OpenAI in the early days 34:14 Why Carnegie Mellon became an AI powerhouse 38:43 What Gray Swan does in AI security 40:44 AI safety vs AI security 43:15 The GCG jailbreak paper 49:19 How AI labs responded to jailbreak research 50:19 State-of-the-art AI defenses 52:32 State-of-the-art AI attacks 54:22 Why AI agents expand the attack surface 58:39 Are AI agents ready for production? 59:40 Mechanistic interpretability explained 1:02:31 Will AI be safer in two years? 1:03:46 Reinforcement learning and self-improving models 1:08:09 Do post-transformer architectures matter 1:09:29 Best research directions in AI now 1:11:00 Zico Kolter’s Intro to Modern AI course 1:14:53 Why modern AI is simpler than people think
与 @zicokolter 的一场极具深度的对谈。@zicokolter 是 @OpenAI 的 board member（董事会成员），也是 @CarnegieMellon 机器学习系主任。话题涵盖 AI safety（AI 安全）、AI security（AI 安防）、agent（智能体）以及 frontier AI（前沿 AI）。00:00 开场介绍 01:32 在 OpenAI 董事会中的角色以及 Safety & Security Committee 03:53 OpenAI 如何审查重大模型发布 05:33 解析 OpenAI 的 preparedness framework（准备度框架） 09:46 前沿 AI 模型是否正在变得更安全？ 12:33 为什么 AI safety 不是规模扩张的自然结果 15:23 AI 风险的四大类别 19:38 AI 中的 doomerism（末日论）与 accelerationism（加速主义）之争 24:11 关于 AI 暂停六个月的辩论 26:20 将 AI safety 视为一项全球性努力 28:04 Zico Kolter 如何进入机器学习领域 31:05 早期的 OpenAI 34:14 为什么 Carnegie Mellon 会成为 AI 重镇 38:43 Gray Swan 在 AI security 方面做什么 40:44 AI safety 与 AI security 的区别 43:15 GCG jailbreak 论文 49:19 AI 实验室如何回应 jailbreak 研究 50:19 当前最先进的 AI 防御 52:32 当前最先进的 AI 攻击 54:22 为什么 AI agent 会扩大攻击面 58:39 AI agent 是否已准备好投入生产环境？ 59:40 解析 mechanistic interpretability（机制可解释性） 1:02:31 两年后 AI 会更安全吗？ 1:03:46 强化学习与自我改进模型 1:08:09 post-transformer（后 Transformer）架构是否重要 1:09:29 当下 AI 最值得投入的研究方向 1:11:00 Zico Kolter 的 Intro to Modern AI 课程 1:14:53 为什么现代 AI 比人们想象的更简单
VCs need to start following the literal description naming trend: The Capital Deployment Company of San Francisco The Check Writing Company of New York The Bridge Round Partnership of Brooklyn The Ghosting Company The Liquidated Preference Firm of Miami
VCs（风险投资人）需要开始追随这种按字面描述来命名的趋势：The Capital Deployment Company of San Francisco、The Check Writing Company of New York、The Bridge Round Partnership of Brooklyn、The Ghosting Company、The Liquidated Preference Firm of Miami
VCs need to start following the literal description naming trend: The Capital Deployment Company of San Francisco The Check Writing Company of New York The Bridge Round Partnership of Brooklyn The Ghosting Company The Liquidated Preference Firm of Miami
VCs（风险投资人）需要开始追随这种按字面描述来命名的趋势：The Capital Deployment Company of San Francisco、The Check Writing Company of New York、The Bridge Round Partnership of Brooklyn、The Ghosting Company、The Liquidated Preference Firm of Miami
Silicon Valley: AI is self-accelerating, agents run everything, old people are dumb Global 2000: I spent a fortune on your AI chat 2 years ago and got zero productivity; my engineers like the coding AI thing but no one else cares; agents are scary Never seen a gap this huge
Silicon Valley：AI 正在自我加速，agent（智能体）会接管一切，老年人很蠢；Global 2000：我两年前在你们的 AI chat（聊天式 AI）上花了一大笔钱，结果生产力提升为零；我的工程师们喜欢那种 coding AI（编程 AI）的东西，但其他人根本不在乎；agents（智能体）很吓人。很少见到这么巨大的认知鸿沟。
Gotta love how Ineffable Intelligence is name mogging all the "superintelligence" named startups lol
不得不说，Ineffable Intelligence 在名字这件事上把所有那些叫“superintelligence”的 startup（初创公司）都给秒了，笑死。
One head-scratching idea that gets repeated endlessly: the new TAM for AI is the size of the human labor market, dollar-for-dollar. Many trillions! Just for like any labor automation technology in history, the price of AI services will be the marginal cost + a normal margin.
一个让人挠头、却被无休止重复的观点是：AI 的新 TAM（总可寻址市场）等于整个人类劳动力市场的规模，按美元一比一计算。那可是数万亿美元！但就像历史上任何劳动自动化技术一样，AI 服务的价格将会是边际成本加上正常利润率。