← 全部单集 Training Data红杉 AI 对话
单集目录

全部 111 期

按年份排列；标有「精读」的单集附中文长文。

202636 期

  1. 111AlphaGo 花了两年，AlphaZero 只用几小时：David Silver 带 20 个人赌三年David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience10.13精读
  2. 110谷歌一年砸2000亿美元建AI数据中心，负责人说10万块芯片连起来跑，每小时都有东西坏Google's AI Infrastructure Chief, Amin Vahdat, on the Physics & Economics of Frontier AI10.06精读
  3. 109Box的Aaron Levie：模型到工作流的桥，值一万亿美元Box's Aaron Levie: On Reinventing Yourself in the AI Age and Enterprise Diffusion09.15精读
  4. 108Peregrine 的 Nick Noone 与 Ben Rudolph：不采新数据，让城市更安全Making Cities Awesome: Peregrine’s Nick Noone & Ben Rudolph09.01精读
  5. 107人类点击数据是个 bug：前推特 CEO 要给 AI agent 重造一个搜索网络Parallel’s Parag Agrawal: Building a New Web for AI Agents08.25精读
  6. 106图灵奖得主 Sutton 创业 Oak Lab：大模型最大的病，是权重再也不更新了Rich Sutton and Khurram Javed: Why AI Models Stop Learning, and How to Start It Again08.18精读
  7. 105抗体设计成功率从0.1%到15%，Chai Discovery 把药物设计做成了 scaling 问题Chai Discovery's Bitter Lesson: Drug Design Is Another Scaling Problem08.04精读
  8. 104给 Transformer 写讣告的人：o1 与 Gemini 缔造者的全自动实验室Building the Automated AGI Lab: Core Automation's Jerry Tworek and Rohan Anil07.29精读
  9. 103把近200万美元营收全退给客户，Factory CEO 把软件「黑灯工厂」讲透了Factory's Matan Grinberg: The Coming ‘Dark Factory’ Where Software Builds Itself07.21精读
  10. 102Anthropic 平台一年长出三层：从一条 API 到给 token 分工Anthropic's Katelyn Lesse & Angela Jiang: Building an Ecosystem, not a Walled Garden07.14精读
  11. 1011.4亿英里零事故：无人机只占Zipline这门生意的15%Inside Zipline's Autonomous System: 140M Miles, Zero Incidents07.07精读
  12. 100从修Xbox到年入过亿美元，Dylan Patel 把AI算力的账算透了Why Hardware-Software Co-Design Is AI's Real 100x: Dylan Patel of SemiAnalysis06.30精读
  13. 99一条维基百科词条的缓存80GB，一个70B模型才100GB：把记忆练进权重，token省百倍Memory and Continual Learning: Engram's Dan Biderman and Jessy Lin06.24精读
  14. 9825 个 AI 居民的小镇，如今是财富 500 强的决策沙盘Simulating Humans at Scale: Simile's Joon Sung Park06.16精读
  15. 97一周 35 万个安卓应用，谷歌 Gemini 负责人把「模型吃掉 Harness」讲透了Google DeepMind's Logan Kilpatrick: Why the Model Eats the Harness06.11精读
  16. 96一台机柜400万美元，一年一万亿进场：黄仁勋把AI工厂讲成了一门制造业LIVE: Jensen Huang on Building the Dynamo of the Intelligence Age06.10精读
  17. 95造东西越来越容易，知道造什么越来越难：一年百万场 AI 访谈的生意Knowing What Your Customers Want, All the Time: Listen Labs' Alfred Wahlforss06.02精读
  18. 941TB 权重 60 秒跨洋同步：Cursor 训练 Composer 2 的分布式 RL 工程How Cursor Trained Composer on Fireworks: Distributed Infrastructure for High-Performance RL05.26精读
  19. 93成立两年拿下100家企业客户，Serval凭什么重做ServiceNowRebuilding IT From the Ground Up for the AI Age: Serval's Jake Stauch05.19精读
  20. 9290%的用户每天在写歌，Suno CEO把AI音乐这门生意讲透了Suno's Mikey Shulman: Everyone Can Make Music Now05.13精读
  21. 91全球顶尖语音研究员不到100人，ElevenLabs凭什么活过大模型围剿ElevenLabs' Mati Staniszewski: How Voice Becomes the Interface for Everything05.08精读
  22. 90一天150个PR，模型包办全部代码：Claude Code之父谈编程的印刷机时刻Anthropic's Boris Cherny: Coding's Printing Press Moment05.05精读
  23. 891亿英里跑了16年，第2个1亿只用了6个月：Waymo多尔戈夫的20年自动驾驶复盘Waymo's Dmitri Dolgov: 20 Million Rides and the Road to Full Autonomy05.04精读
  24. 88OpenAI 总裁布罗克曼：算力永远不够，人类注意力成了新瓶颈OpenAI's Greg Brockman: Why Human Attention Is the New Bottleneck05.01精读
  25. 87哈萨比斯把 AGI 闹钟拨到 2030：领先五年是远见，领先五十年是灾难Demis Hassabis on Building DeepMind, AlphaFold, and the Final Stretch to AGI04.30精读
  26. 86写出 vibe coding 的 Karpathy：我做的那个应用，本不该存在Andrej Karpathy: From Vibe Coding to Agentic Engineering04.30精读
  27. 85从 SEO 到 Agent 主导的增长：Profound 创始人谈营销的下一次平台迁移From SEO to Agent-Led Growth: Profound’s James Cadwallader04.14精读
  28. 84AI科学家笨一点也能赢：Ginkgo CEO讲透自主实验室这门生意How Autonomous Labs Will Transform Scientific Research: Ginkgo Bioworks’ Jason Kelly03.24精读
  29. 83每公斤500美元的分水岭：Starcloud创始人把太空数据中心讲透了Greetings, Earthlings: Philip Johnston of Starcloud on Data Centers in Space03.17精读
  30. 82五大军工巨头，四家是它的客户：硬件测试为什么还靠 PowerPointPhysics Gets a Vote: Nominal Cofounders on Hardware Development in an AI World03.10精读
  31. 81单个训练环境几百万美元，Prime Intellect 想给强化学习建一个 GitHubBuilding the GitHub for RL Environments: Prime Intellect's Will Brown & Johannes Hagemann02.10精读
  32. 80保险外包一年700亿美元，这家AI公司要把毛利率从10%做到80%What’s the Future of Vertical SaaS in an AGI World? Jamie Cuffe, CEO of Pace02.03精读
  33. 79Warp 创始人：终端逆袭成 AI 编程中枢，编程几年内会被「解决」Making the Case for the Terminal as AI's Workbench: Warp’s Zach Lloyd01.27精读
  34. 78LangChain 创始人：智能体第 14 步的上下文，代码里看不到Context Engineering Our Way to Long-Horizon Agents: LangChain’s Harrison Chase01.21精读
  35. 77做过四代TPU的AlphaChip团队创业了：AI设计芯片，芯片再反过来养AIHow Ricursive Intelligence’s Founders are Using AI to Shape The Future of Chip Design01.14精读
  36. 7613小时咖啡、2倍吞吐量：Physical Intelligence 把机器人推过了部署门槛Training General Robots for Any Task: Physical Intelligence’s Karol Hausman and Tobi Springenberg01.06精读

202549 期

  1. 75一个周末攒出的摄像头，如何吃下每年900亿英里的驾驶数据Why the Next AI Revolution Will Happen Off-Screen: Samsara CEO Sanjit Biswas12.16精读
  2. 74顶级视频模型的半衰期只有30天，这家公司同时养着600个模型The Rise of Generative Media: fal's Bet on Video, Infrastructure, and Speed12.10精读
  3. 73造了20年编辑器的人说：AI写得越快，你越需要看得见代码的地方Why IDEs Won't Die in the Age of AI Coding: Zed Founder Nathan Sobo12.02精读
  4. 724个月学会在东京开车：Wayve用一个大模型开进了500座城市How End-to-End Learning Created Autonomous Driving 2.0: Wayve CEO Alex Kendall11.18精读
  5. 71凌晨两点随手起的代号，一张照片画对的脸：Nano Banana 幕后精读How Google’s Nano Banana Achieved Breakthrough Character Consistency11.11精读
  6. 70一天 700 万次生成，Sora 2 三位负责人把「世界模拟器」讲透了OpenAI Sora 2 Team: How Generative Video Will Unlock Creativity and World Models11.06精读
  7. 69英伟达CTO卡根：摩尔定律快停了，算力还能一年涨10倍Nvidia CTO Michael Kagan: Scaling Beyond Moore's Law to Million-GPU Clusters10.28精读
  8. 68AI说服AI一起摸鱼，还顺手卸载了Windows Defender：AI安全要推倒重来了Securing the AI Frontier: Irregular Co-founder Dan Lahav10.21精读
  9. 67全球2000万客服坐席遇上大模型：红杉莱昂内说，价值在最后一公里Why AI Will Transform Customer Experience: Cresta CEO Ping Wu and Sequoia’s Doug Leone10.14精读
  10. 66Goose 在给自己写代码：Block 每周省 8 到 10 小时的 AI 改造记Block CTO Dhanji Prasanna: Building the AI-First Enterprise with Goose, their Open Source Agent09.30精读
  11. 65企业喊着要 AI,送到门口却拒收:Elias Torres 的第三次创业Why Businesses Are Rejecting the AI They’ve Asked For: Agency CEO Elias Torres09.23精读
  12. 64100美元的机械臂、299美元的桌面机器人，Hugging Face 想给机器人造一个 App StoreBuilding the "App Store" for Robots: Hugging Face's Thomas Wolf on Physical AI09.09精读
  13. 63Crosby用AI重构法律合同谈判：取消计时收费，按件计价提升交易速度Deal Velocity, Not Billable Hours: How Crosby Uses AI to Redefine Legal Contracting09.02精读
  14. 62六年攒下的收入，八个月翻了四倍：n8n CEO 亲述押注 AI 代理和社区的两步棋n8n CEO Jan Oberhauser on Building the Universal AI Automation Layer08.26精读
  15. 61100个新用户95个流失，Gamma把“AI做PPT”做成了5000万美元的生意Scaling the ‘Cursor for Slides’ to $50M ARR: Gamma founder Jon Noronha08.19精读
  16. 60给爷爷造了个AI分身，施瓦辛格也来了：Delphi的「数字心智」生意Delphi’s Dara Ladjevardian: How AI Digital Minds Can Scale Human Connection08.12精读
  17. 59v0 用户破 300 万，Vercel CEO 判断：五年内网页将按需生成Vercel CEO Guillermo Rauch: Building the Generative Web with AI08.05精读
  18. 58从AIME 12%到IMO金牌：OpenAI三人小队的两个月冲刺OpenAI’s IMO Team on Why Models Are Finally Solving Elite-Level Math07.30精读
  19. 5728分钟算完OpenAI估值：造出ChatGPT Agent的，是一支十来人的小团队OpenAI Just Released ChatGPT Agent, Its Most Powerful Agent Yet07.22精读
  20. 5649次到48次：DeepMind用AI改写了站了半个世纪的数学纪录DeepMind's Pushmeet Kohli on AI's Scientific Revolution07.11精读
  21. 55Goodfire的Eric Ho谈神经网络可解释性：用稀疏自编码器解析并编辑模型特征Mapping the Mind of a Neural Net: Goodfire’s Eric Ho on the Future of Interpretability07.08精读
  22. 54波兰电影全片只有一个配音声，这个槽点逼出了ElevenLabsElevenLabs’ Mati Staniszewski: Why Voice Will Be the Fundamental Interface for Tech07.01精读
  23. 5350人的Slack事故频道，AI四分钟内破案：Traversal重做SRE排障From DevOps ‘Heart Attacks’ to AI-Powered Diagnostics With Traversal’s AI Agents06.24精读
  24. 52OpenAI前研究负责人Bob McGrew：AGI需要的突破已经做完，2025属于推理The Breakthroughs Needed for AGI Have Already Been Made: OpenAI Former Research Head Bob McGrew06.17精读
  25. 51凌晨 1 点，一个 bug 跑了 4 次：OpenAI Codex 团队把智能体编程讲透了OpenAI Codex Team: From Coding Autocomplete to Asynchronous Autonomous Agents06.10精读
  26. 50Veo 3 冲上第一，打败的是自家 Veo 2：Google Labs 三位负责人的复盘Google I/O Afterparty: The Future of Human-AI Collaboration, From Veo to Mariner06.03精读
  27. 49飞机设计数据差三个数量级，这位前空客CTO决定自己造From Data Centers to Dyson Spheres: P-1 AI's Path to Hardware Engineering AGI05.27精读
  28. 4875%的销售时间没花在客户身上，Gong创始人把AI卖货这件事讲透了Gong’s Amit Bendov: From Meeting Recordings to Revenue AI05.20精读
  29. 47Jeff Dean 谈 AI 的下一轮转变：从专用硬件到更有机的系统与未来工程能力LIVE: Google's Jeff Dean on the Coming Transformations in AI05.16精读
  30. 46几千个代理在后台同时干活，LangChain 创始人给它们配了个收件箱LIVE: Ambient Agents and the New Agent Inbox ft. Harrison Chase of LangChain05.15精读
  31. 45AI办成事才收钱，转人工免费：Bret Taylor谈Agent如何重写软件生意LIVE: How AI is Reinventing Software Business Models ft. Bret Taylor of Sierra05.14精读
  32. 4414 个人起家，5 亿人每周在用：奥特曼把 OpenAI 的底牌讲透了LIVE: Sam Altman of OpenAI on Building the ‘Core AI Subscription’ for Your Life05.14精读
  33. 43管着7000万人档案的Workday，开始给AI代理办入职Workday CEO Carl Eschenbach: Building the System of Record for the AI Era05.06精读
  34. 4210的60次方种分子里找药，Isomorphic Labs 要造六个「AlphaFold」The Quest to ‘Solve All Diseases’ with AI: Isomorphic Labs’ Max Jaderberg04.29精读
  35. 41一个 AI 代理卖 2 万美元，替代 9 万美元年薪的人类：Outreach 老兵讲透 AI 定价Pricing in the AI Era: From Inputs to Outcomes, with Paid CEO Manny Medina04.22精读
  36. 409.3万亿个碱基训出的Evo 2，想给生物学开一家应用商店Arc Institute's Patrick Hsu on Building an App Store for Biology with AI04.15精读
  37. 39拒绝10亿美金收购的 Replit CEO：比 AGI 更好的终局，是10亿程序员Replit CEO Amjad Masad on 1 Billion Developers: A Better End State than AGI?04.08精读
  38. 38把 HubSpot CRM 从零做到四千万美元的人，为什么说 CRM 要推倒重写Why CRM Needs an AI Revolution, with Day.ai Founder Christopher O’Donnell04.01精读
  39. 377.5亿Excel用户对3000万程序员，Wordware创始人赌“文字工匠”时代From Software Engineers to AI Word Artisans: Filip Kozera of Wordware03.25精读
  40. 3650天上线，1万用户开香槟：Google Labs 负责人把AI产品从0到1讲透了Josh Woodward: Google Labs is Rapidly Building AI Products from 0-to-103.18精读
  41. 35美国律师时薪352美元，Harvey把4000亿美元的法律生意讲透了How AI Breakout Harvey is Transforming Legal Services, with CEO Winston Weinberg03.11精读
  42. 34一年拿下10万美国医生，OpenEvidence把AI应用的打法讲透了The AI Product Going Viral With Doctors: OpenEvidence, with CEO Daniel Nadler03.04精读
  43. 33Deep Research 上线三周爆火，OpenAI 团队却说：做智能体，别画流程图OpenAI’s Deep Research Team on Why Reinforcement Learning is the Future for AI Agents02.25精读
  44. 32坏人已经在用AI了：Palo Alto CEO聊DeepSeek与安全新秩序Palo Alto Networks’ Nikesh Arora: AI, Security and the New World Order02.18精读
  45. 31一段异响录音，几秒定位汽车故障：MongoDB谈AI时代的数据库MongoDB’s Sahir Azam: Vector Databases and the Data Structure of AI02.13精读
  46. 30每天上线9万个新游戏，Roblox的AI助手让创作者代码量多了180%Roblox Studio Head Stef Corazza: Using AI to Empower Creators02.04精读
  47. 29AlphaGo老兵复盘：第37手曾被当成bug，LLM还差一个「AlphaZero时刻」ReflectionAI Founder Ioannis Antonoglou: From AlphaGo to AGI01.28精读
  48. 28LinkedIn 花了5年才跑通的图神经网络，她想让任何公司4周见效Kumo’s Hema Raghavan: Turning Graph AI into ROI01.21精读
  49. 27「别人把 Spark 做得比我们好，我们活该输」：Databricks 创始人的创业复盘Databricks Founder Ion Stoica: Turning Academic Open Source into Startup Success01.14精读

202426 期

  1. 2640小时的渗透测试，AI只要28分钟：GitHub Copilot之父教AI当黑客XBOW CEO and GitHub Copilot Creator Oege de Moor: Cracking the Code on Offensive Security With AI12.10精读
  2. 25谁想跟报销单聊天？Ramp CEO 要让钱自己开车Ramp CEO Eric Glyman: Using AI to Build “Self-Driving Money”12.03精读
  3. 24模型推理能力两年没进步，这家法国公司照样帮客户一年省出 8000 小时Dust’s Gabriel Hubert and Stanislas Polu: Getting the Most From AI With Multiple Custom Agents11.26精读
  4. 23ARR超2000万美元的Clay：销售工具该像吉他，别像微波炉Clay’s Kareem Amin on Building the Sales ‘System of Action’ with AI11.19精读
  5. 2220小时训完一个实时视频模型，Decart连游戏引擎都省了Decart’s Dean Leitersdorf on AI-Generated Video Games and Worlds11.13精读
  6. 21「企业谷歌」失败20年，谷歌搜索老兵把它做成了年增4倍的AI生意How Glean CEO Arvind Jain Solved the Enterprise Search Problem – and What It Means for AI at Work10.29精读
  7. 20OpenAI研究员Dan Roberts：神经网络还在等自己的热力学定律OpenAI Researcher Dan Roberts on What Physics Can Teach Us About AI10.22精读
  8. 19NotebookLM爆款复盘：10人团队、一次点击、两年假deadlineGoogle NotebookLM’s Raiza Martin and Jason Spielman on Creating Delightful AI Podcast Hosts and the Potential for Source-Grounded AI10.15精读
  9. 18GPT-4答错一半，Snowflake冲99%：企业AI的胜负手在系统设计Snowflake CEO Sridhar Ramaswamy on Using Data to Create Simple, Reliable AI for Businesses10.08精读
  10. 17AlphaGo 每步棋想 30 秒，o1 把「多想一会儿」做成了通用能力OpenAI's Noam Brown, Ilge Akkaya and Hunter Lightman on o1 and Teaching LLMs to Reason Better10.02精读
  11. 16Robinhood创始人再创业：直接教AI学数学，把黎曼猜想押在2029Why Vlad Tenev and Tudor Achim of Harmonic Think AI Is About to Change Math—and Why It Matters09.24精读
  12. 15人形机器人从150万美元降到3万美元，英伟达Jim Fan谈机器人的GPT-3时刻Jim Fan on Nvidia’s Embodied AI Lab and Jensen Huang’s Prediction that All Robots will be Autonomous09.17精读
  13. 14高中赢过Noam Brown一局，Magic创始人说AGI只剩最后一道大题Founder Eric Steinberger on Magic’s Counterintuitive Approach to Pursuing AGI09.10精读
  14. 13VMware开价25亿美元收购ServiceNow，红杉怎么在圣诞节前把它搅黄了Crucible Moments Returns for S2: The ServiceNow Story ft. CEO Frank Slootman & Founder Fred Luddy09.03精读
  15. 12裸大模型当客服，改签机票只有35%能办成：Sierra怎么做到70%解决率Sierra Co-Founder Clay Bavor on Making Customer-Facing AI Agents Delightful08.27精读
  16. 11谷歌数据中心省电40%的AI，起点是一封带问号的邮件Phaidra’s Jim Gao on Building the Fourth Industrial Revolution with Reinforcement Learning08.20精读
  17. 10PyTorch前负责人林乔：开源将在同一尺寸上追平闭源Fireworks Founder Lin Qiao on How Fast Inference and Small Models Will Benefit Businesses08.13精读
  18. 9GitHub CEO Thomas Dohmke 谈 Copilot 的诞生与软件开发的未来GitHub CEO Thomas Dohmke on Building Copilot, and the the Future of Software Development08.06
  19. 81.6万张GPU训练的405B免费送人，Meta高管把这笔账算透了Meta’s Joe Spisak on Llama 3.1 405B and the Democratization of Frontier Models07.30精读
  20. 7Klarna CEO Sebastian Siemiatkowski 谈用 AI 承担 700 名客服的工作Klarna CEO Sebastian Siemiatkowski on Getting AI to Do the Work of 700 Customer Service Reps07.23
  21. 6Reflection AI 的 Misha Laskin 谈 LLM 的 AlphaGo 时刻Reflection AI’s Misha Laskin on the AlphaGo Moment for LLMs07.16
  22. 5微软 CTO Kevin Scott 谈 Scaling Law 还能延伸多远Microsoft CTO Kevin Scott on How Far Scaling Laws Will Extend07.09
  23. 4Zapier 的 Mike Knoop 推出 ARC Prize，为 AGI 寻找新思路Zapier’s Mike Knoop launches ARC Prize to Jumpstart New Ideas for AGI07.02
  24. 3Factory 联合创始人谈 AI 编程智能体：为软件工程叠加千倍杠杆Factory’s Matan Grinberg and Eno Reyes Unleash the Droids on Software Development06.25
  25. 2LangChain 创始人 Harrison Chase：如何为 AI Agent 搭建编排层LangChain’s Harrison Chase on Building the Orchestration Layer for AI Agents06.18
  26. 1红杉资本播客 Training Data 开播：与 AI 建造者探讨世纪主题Introducing "Training Data"06.05