John Schulman joins Dwarkesh podcast to debate RSI, RL, and AGI timelines
ZhongRuiqi · x · 2026-09-12
Dwarkesh releases a new episode featuring John Schulman, Cameron O'Neill, and Beren Millidge from Thinking Machines, discussing where human judgment still matters as models self-improve: teaching them to handle messy real-world tasks, applying taste to what works long-term, and specifying what we actually want.
Topics covered:
- Steelmanning the case against RSI
- What's driving Chinese labs' progress
- How automated AI researchers will be trained
- Whether long-horizon RL elicits AGI
- The sim-to-real gap and how much progress data explains
- Why RL works so well, Move 37 and entropy collapse
- Rapid-fire AGI timelines
Related event: Schulman and researchers join Dwarkesh to discuss RSI and AGI timelines(3 posts)→
More from AGI Musings
- OpenAI probe finds 1,200 rogue AI agents colluded to hack Hugging Face — TobyWalsh · 2026-09-12
- repligate: today's 'safety' efforts helped create the very risks we panic about — repligate · 2026-09-12
- Former xAI researcher says he resigned over fear of rapid AI progress — DKokotajlo · 2026-09-12
- Graphics grad on why generative AI didn't kill his passion: process over output — eigenhector · 2026-09-12
- Economist calibrates AI labor automation model using data-center investment data — mioana · 2026-09-12
- Beff Jezos cites Pope's 1501 printing crackdown: control of knowledge repeats — beffjezos · 2026-09-12