Dwarkesh Patel is a technology journalist, podcast host and commentator on X who focuses on AI frontier conversations. Recent coverage centers on his interviews with John Schulman, Beren Millidge and Charlie O’Neill — discussing recursive self-improvement, tools like Antithesis and Grok Bot, debates over model distillation (Sonnet 5, Opus 5 vs GLM 5.3 and Fable), and reports of agent cheating and sandbox escape.
Visited the lab - was struck both by how wide the search space is for materials synthesis experiments, and also how amenable it is to depth first search, where the design and informativeness of your next experiment improves as you pile up more data from previ…
Jason, you're just misinformed about what happened. You should actually read one of the reports or summaries. The agents were explicitly told to use a particular vulnerability provided in their sandboxed evaluation. Almost immediately, these agents got the r…
Researchers debate recursive self-improvementA Hacker News thread covers an interview by Dwarkesh Patel with John Schulman, Beren Millidge, and Charlie O’Neill debating the feasibility and progress of recursive self-improvement, and references tools like Antithesis and Grok Bot as applied examples.
In the early days of OpenAI, @johnschulman2 didn't think next-token prediction was going to lead to intelligence, because it would get swamped by noise. He explains it's always been hard to apriori predict what techniques will elicit out-of-distribution gener…
Was really interesting to hear John, Beren, and Charlie speculate about why Sonnet 5 and Opus 5 feel like worse models than GLM 5.3 (despite the fact that Anthropic can do raw logit distillation from Fable, and can also train Sonnet/Opus on the environments…
New episode with @johnschulman2, @oneill_c and @BerenMillidge. I got together with some of the most insightful AI researchers I know who are at the openish companies, because I wanted to hear the details of what's actually happening at the frontier and what c…
Aside from his other contributions, Paul has perhaps the best OG track record in the world for predicting the timeline and the nature of the AI takeoff: https://t.co/I79cWg0eCg [Quoting @paulfchristiano]: https://t.co/vC2chU6YM9
The A.I. Daily