The Generalist

Joon Sung Park, AI researcher and Simile co-founder, is building foundation models that simulate human behavior. He explores digital twins, predicting individual and group decisions, and how simple social rules create large-scale patterns. They discuss applications from policy testing and legal strategy to online communities and finance, alongside the ethical risks of simulation. Park also reflects on art, portraiture, trust, and the future of decision-making.

Pete Florence, co-founder and CEO of Generalist, develops foundation models for robotics and physical intelligence. He discusses one-shot learning, dexterity, robotic touch, and why physical experience is so difficult to gather. The conversation explores building versatile models, teaching robots through coaching, open versus closed systems, geopolitics, and a future where robots improvise with unfamiliar tools and help accelerate science.

Jaan Tallinn, the Skype and Kazaa founding engineer and AI safety pioneer, explores humanity’s precarious future. He discusses what the OpenAI-Hugging Face incident revealed about AI motivation, why frontier model training may need to pause, and how hardware verification could support global governance. He also reflects on growing up in Soviet Estonia, investing in DeepMind and Anthropic, freestyle dancing, and living meaningfully amid accelerating AI progress.

Liam Fedus, former OpenAI post-training leader and ChatGPT creator, and Dogus Cubuk, a physicist and ex-Google DeepMind materials-science lead, co-founded Periodic Labs to build “synthesis superintelligence.” They discuss why language models alone cannot discover matter, the stubborn challenges of reproducing materials, and their pursuit of high-temperature superconductors. They also revisit ChatGPT’s surprise launch and reveal how AI-driven laboratories could transform science.

Scott Morton, Revel’s founder and CEO, spent nearly a decade building control software for SpaceX’s Falcon 9 and Starship. He discusses why critical hardware still depends on outdated tools, Revel’s runtime-safe programming language, and the limits of AI in nuclear and aerospace systems. He also reflects on Elon Musk’s influence, ambitious hardware bets, and scaling Revel across the coming hardware renaissance.

Farooq Malik, co-founder and CEO of Rain, builds tokenized money rails using stablecoins for faster, cheaper payments and card issuance. He discusses global last-mile frictions, how stablecoins restore cash-like immediacy, why building infrastructure early paid off, and how Rain routes money and envisions AI-driven payment use cases.

Eric Nguyen, co-founder and CEO of Radical Numerics, builds AI that reads and writes DNA and other omics. He discusses treating DNA as a language, connecting DNA, RNA, proteins and epigenetics in one model, why wet labs are a bottleneck, and balancing powerful biological design with biodefense and safety measures.

Matan Grinberg, co-founder and CEO of Factory and former string-theory researcher, explains how conserved principles from physics inform startup decisions. He discusses the new resource tradeoffs between headcount, data, and token budgets. Matan also outlines Factory’s autonomous engineering "droids," model-agnostic routing, and why open-weight models and multiple-model strategies matter.

Yash Patil, co-founder and CEO of Applied Compute and ex-OpenAI researcher, builds custom AI models for businesses. He argues companies must own models to avoid dependency. The conversation covers post-training as the new source of advantage, why evals stay private, a DoorDash case where a tiny specialized model beat frontier models, and why cost and continuous learning drive custom models.

Bryon Hargis, co-founder and CEO of Castelion and former SpaceX and Johns Hopkins APL engineer, builds low-cost hypersonic missiles for scalable deterrence. He discusses why hypersonics provide a credible middle-ground response, China's manufacturing edge, the game theory of tit-for-tat escalation, warhead signaling to avoid miscalculation, and designing for manufacturability over peak capability.