John’s AI Research Podcast

EP13: Jensen Huang Declares AGI on Lex Fridman, Google AI Co-Scientist rewrites drug discovery, DIVA-GRPO adaptive training at ICLR 2026, rewritable DNA hard drives, and swarm intelligence lessons for multi-agent AI systems. [CN]

EP13: Jensen Huang Declares AGI on Lex Fridman, Google AI Co-Scientist rewrites drug discovery, DIVA-GRPO adaptive training at ICLR 2026, rewritable DNA hard drives, and swarm intelligence lessons for multi-agent AI systems. [EN]

深度剖析OpenClaw开源AI Agent网关,二十五万星GitHub项目。架构设计、生产部署和生态系统。

Deep dive into OpenClaw, the open-source multi-platform AI Agent gateway with 250K+ GitHub stars. Architecture, production deployment, and ecosystem.

Claude Code 生产级使用完全指南，涵盖工作流、最佳实践和高级技巧。

A comprehensive guide to using Claude Code effectively in production environments, covering workflows, best practices, and advanced techniques.

解读开创性的 Transformer 论文，这篇论文彻底革新了自然语言处理领域，奠定了现代 AI 的基础。

Breaking down the landmark Transformer paper that revolutionized NLP and laid the foundation for modern AI.

深度剖析 Manus 自主 AI 智能体，探索其如何突破 AI 系统自主完成任务的边界。

A deep dive into Manus, the autonomous AI agent pushing the boundaries of what AI systems can accomplish independently.

深入解析 LangGraph 的架构设计，探讨如何构建具有持久记忆和复杂工作流的有状态多智能体 AI 系统。

Exploring LangGraph's architecture for building stateful, multi-agent AI systems with persistent memory and complex workflows.

一期特别教程专辑，系统拆解 AI 短视频全流程：整体设计、分镜头、剧本写作、AI视频生成（可灵/Sora/Runway）、无版权音乐、声音克隆配音、剪辑发布。包含完整的 Manus/Agent 自动化提示词。 博客完整图文版：https://johngai.com

A comprehensive tutorial episode covering the entire AI short video production pipeline: planning, storyboarding, scripting, AI video generation (Kling/Sora/Runway), royalty-free music, AI voice cloning, editing, and publishing. Includes complete Manus/Agent prompts for automation. Blog post with full details and prompts: https://johngai.com

谷歌Titans MIRAS重新定义AI长期记忆。小米MiMo-V2-Pro（猎人阿尔法）在OpenRouter上匿名测试一万亿token。苹果×谷歌Siri-Gemini联姻重塑隐私AI。MangroveGS以80%准确率预测癌症转移。Meta Omnilingual MT覆盖1600种语言。

Google Titans MIRAS redefines AI long-term memory. Xiaomi's MiMo-V2-Pro (Hunter Alpha) stealth-tested 1T tokens on OpenRouter. Apple×Google Siri-Gemini partnership reshapes private AI. MangroveGS predicts cancer metastasis at 80% accuracy. Meta Omnilingual MT covers 1600 languages.

第五期晚间版：多智能体辩论中的不确定性分解（认识论与偶然性之别），二零二六年 MCP 与 A2A 协议格局，Lex Fridman 第四九零期解析，太平洋海山百余个新物种发现，以及十五分钟城市理论与实践的张力。

EP05 Evening Edition: Multi-agent debate uncertainty decomposition (epistemic vs aleatoric), MCP vs A2A protocol landscape 2026, Lex Fridman #490 State of AI breakdown, Pacific seamount 100 new species discovery, and the 15-minute city theory vs practice.

EP03 — 二零二六年三月二十日。本期话题：中东冲突第三周与霍尔木兹海峡石油咽喉；美联储维持利率百分之三点五至三点七五不变，道指暴跌七百六十八点，黄金周跌百分之十；滞胀经济学迷你课；稳定币成为应急货币（市值两千四百亿美元）；全球开斋节庆典；量子几何隐藏结构发现及阿贝尔奖授予格尔德·法尔廷斯。

EP03 — March 20, 2026. Topics: Middle East conflict Week 3 & Strait of Hormuz oil chokepoint; Fed holds rates at 3.5-3.75%, Dow plunges 768 points, gold crashes 10%; Stagflation mini-lesson; Stablecoins as emergency currency ($240B market cap); Eid al-Fitr celebrations worldwide; Hidden quantum geometry discovery & Abel Prize for Gerd Faltings.

Daily Briefing EP03 - March 20, 2026. Fed holds rates, Middle East conflict reshapes markets, stablecoins as emergency currency, Eid al-Fitr, quantum geometry discovery.

Daily Briefing EP03 - March 20, 2026. Fed holds rates, Middle East conflict reshapes markets, stablecoins as emergency currency, Eid al-Fitr, quantum geometry discovery.

第四期：当测试来驾驶Agent，混合架构来改写规则。深度解读TDAD论文（回归率降低70%）、OlMo Hybrid混合架构（49%训练数据节省）、MCP 2026企业级路线图、金缮哲学与Agent系统设计、莱维飞行随机游走。

EP04: When Tests Drive the Agent, and Hybrids Rewrite the Rules. Deep dive into TDAD (test-driven agentic development reducing regressions by 70%), OlMo Hybrid (49% data savings via GDN architecture), MCP 2026 roadmap, Kintsugi philosophy for agent systems, and Lévy flight random walks.

第三期：当 Agent 学会扩展，真菌网络早已给出答案 深度解读 Google 首个多 Agent 系统量化扩展法则（180种配置实测）、Linux Foundation 成立 Agentic AI Foundation（MCP/goose/AGENTS.md三大项目）、Latent Space 播客 Box CEO Aaron Levie 论企业 Agent 治理，以及来自菌根真菌网络和博弈论机制设计的跨领域启发。 核心要点：(1) 加 Agent 不是万能药——顺序推理任务性能下降39-70%；(2) Agent 互操作进入标准化时代；(3) 真菌网络的"主干路线"模式提供了超越"集中vs分散"的第三条路。

EP03: When Agents Learn to Scale, and Fungi Show Them How Deep dive into Google's first quantitative multi-agent scaling laws (180 configurations tested), the Agentic AI Foundation (AAIF) and its three foundational projects (MCP, goose, AGENTS.md), Aaron Levie's insights on enterprise agent governance from Latent Space podcast, plus cross-domain connections from mycorrhizal fungal networks and game-theoretic mechanism design. Key takeaways: (1) More agents is NOT a silver bullet - sequential tasks lose 39-70% performance; (2) Agent interoperability enters its standards era with AAIF; (3) Nature's fungal "trunk routes" offer a third path beyond centralized vs decentralized architectures.

Daily Briefing Episode 2. Dubai International Airport hit by drone strike on fuel depot; USS Ford carrier fire during Iran war Day 18; Mediterranean 2026 death toll reaches 682; Brent crude above 00; Bitcoin surges past 5,000; Kubernetes Ingress NGINX retirement and Gateway API migration; Boltzmann Brain paradox revisited. Five sections: World News, Economy, Cloud/DevOps/Crypto, Feature Story, Science.

每日简报第二期。迪拜国际机场遭无人机袭击燃料库起火；美伊战争第18天USS Ford航母火灾；地中海2026年初682人死亡/失踪创纪录；布伦特原油破100美元；BTC突破75000美元；Kubernetes Ingress NGINX退役与Gateway API迁移；玻尔兹曼大脑悖论新研究。五大板块：全球时事·经济金融·云原生运维·人文副刊·自然科学。

深入解读AI领域最深刻的发现——规模定律（Scaling Laws）。从Kaplan 2020年的开创性论文，到涌现能力（Emergent Abilities）的争议，再到2025-2026年推理时计算（Test-Time Scaling）的新方向。探讨智能是否只是规模的自然产物，以及Rich Sutton「苦涩的教训」对AI发展的哲学启示。关键概念：Scaling Laws / Chinchilla / Emergent Abilities / The Bitter Lesson / Test-Time Scaling。

西方哲学史上最伟大的悲观主义者，第一个读懂东方智慧的欧洲哲学家。从富商之子到孤独天才，叔本华的意志哲学、欲望钟摆、与佛教四圣谛的惊人对应，以及他对尼采、弗洛伊德、维特根斯坦的深远影响。最后用刺猬困境收尾——两百年后依然精准得令人心酸。| johngai.com

When multiple aligned AI agents share a real environment, game theory takes over. A 38-researcher study reveals local alignment ≠ global stability. Plus: OpenClaw (Agent OS) vs Context7 (minimal tool surface) design philosophies, Don Tapscott on Identic AI as your cognitive double, and cross-domain insights from termite stigmergy and Michelin kitchen brigade systems. | johngai.com