Dwarkesh Podcast Hosts John Schulman and Frontier Researchers on RSI, RL, and AGI Timelines
On September 12, Dwarkesh Patel released a new podcast episode featuring three researchers at frontier (open-leaning) labs — Thinking Machines' John Schulman, Beren Millidge, and a third guest — discussing real progress inside frontier labs and what comes next.
Confirmed
- The core topic was recursive self-improvement (RSI), including a steelman of counterarguments and an assessment of how far RSI is from reality
- The guests discussed why RL works, recent progress in long-horizon RL, and RL as a path toward AGI
- They addressed the value of human judgment in an era of model self-improvement: how to teach models and maintain human oversight
- The conversation also covered Chinese lab progress, automated researchers, and AGI timelines
Unconfirmed
- The third guest's name varies across reposts (Charlie/Cameron/Cassandra O'Neill, Callum O'Neill, Colin O'Neill, Miles O'Neill), with inconsistent affiliation details, so their identity is not confirmed here
Why it matters
John Schulman, a core researcher at Thinking Machines and former OpenAI co-founder-level figure, offers a frontline perspective on RSI and RL as paths to AGI, making this a valuable reference for assessing AGI timelines and technical roadmaps.
2026-09-12 ~ 2026-09-13 · 7 related posts
Primary sources
- [source] Schulman, Millidge, O'Neill Debate How Close AI Is to Recursive Self-Improvement — Dwarkesh Patel · 2026-09-12
- Dwarkesh podcast: Schulman, O'Neill & Millidge on RSI, long-horizon RL and AGI timelines — saranormous · 2026-09-12
- John Schulman joins Dwarkesh podcast to debate RSI, RL, and AGI timelines — ZhongRuiqi · 2026-09-12
4 near-duplicate retellings: Scobleizer · _AndrewZhao · SonglinYang4 · _sholtodouglas