Matt Turck talks with Richard Socher on recursive self-improvement and AI-driven science
mattturck · x · 2026-09-11
Investor Matt Turck sits down with Richard Socher (founder of you.com / Recursive AI) for a 70-minute conversation on AI building AI. Key threads: why scientific progress is slowing and whether AI can re-integrate fragmented human knowledge; how LLMs learn biology and proteins; next-token prediction as a world model; whether AI can generate truly original ideas; simulations and verifiers on the path to superhuman AI; why hallucinations can drive discovery; AI accelerating drug discovery and possibly curing cancer; Socher's 'Eureka Machine' four pillars, virtual cells, self-driving robotic labs, agent swarms; and the compute and scientific-data bottlenecks that constrain it all.
Related event: Matt Turck Talks Richard Socher on AI Building AI(2 posts)→
More from AGI Musings
- Betting $1M vs $10M that AI won't cause human extinction by 2030 — robleclerc · 2026-09-11
- If we'd built AI in the 1800s, people would've said 'we need ASI to solve diabetes' — zetalyrae · 2026-09-11
- The tragedy of expertise: AI automating junior work may hollow out tomorrow's seniors — IgorCarron · 2026-09-11
- Peter Diamandis: AI and new instruments will break 'settled science' next decade — PeterDiamandis · 2026-09-11
- Lila Sciences bets open-endedness is core infrastructure for AI-driven autonomous science — tw_killian · 2026-09-11
- If AI is an existential risk, why spend trillions on data centers? — PAstynome · 2026-09-11