Schulman, Millidge, O'Neill Debate How Close AI Is to Recursive Self-Improvement
Dwarkesh Patel · youtube · 2026-09-12
Dwarkesh Patel hosts a deep discussion with John Schulman, Charlie O'Neill, and Beren Millidge on recursive self-improvement: steelmanning the skeptics, what drives Chinese labs, training automated AI researchers, long-horizon RL, the sim-to-real gap, and rapid-fire timelines.
Related event: Dwarkesh talks with Schulman and researchers on RL and AGI timelines(2 posts)→
More from AGI Musings
- Jensen Huang slams 'end of humanity' AI predictions as nonsense; Chollet maps the doomer camps — beffjezos · 2026-09-12
- Counterpoint to AI-credit debate: builders of AI that can answer hard prompts deserve credit — _aidan_clark_ · 2026-09-12
- AI research agent cracks Komlós conjecture, proves 3√(2πt) Beck-Fiala bound — basedjensen · 2026-09-12
- Cambridge's David Krueger: the real AI race is builders vs. those trying to stop them — DavidSKrueger · 2026-09-12
- If your AI work risks 800M lives, you have a duty to quit, argues Arpit — soumitrashukla9 · 2026-09-12
- Economists: AI could transform society yet show only 4-5% GDP growth — soumitrashukla9 · 2026-09-12