A conversation with John Schulman on the first year LLMs could have been useful, building research teams, and where RL goes from here. 00:20 - Speedrunning ChatGPT 09:22 - Archetypes of research managers 11:56 - Was OpenAI inspired by Bell Labs? 16:54 - The absence of value functions 18:23... - Continual learning 21:09 - Brittle generalization 24:05 - Co-training generators and verifiers, GANs 27:06 - John’s personal use of AI for research 28:54 - Day in the life 33:01 - Slowdowns in consequential ML ideas 36:21 - "Peer review" within the labs 39:19 - Distribution shift in researchers 43:33 - Future of RL 45:33 - Will the labs coordinate if the world needs them to? 44:46 - Forecasting ills in AGI and engineering 47:53 - Thinking Machinesshow more

@johnschulman2 let's goooo

@johnschulman2 very cool

@tszzl @johnschulman2 What’s an LLM?

@johnschulman2 One thing that stands out in this conversation is that access is no longer the advantage. We all have similar LLMs, papers, and tools. The real difference is how you think, the questions you ask, and how you structure the work. AI amplifies judgment. It doesn’t replace it.

@johnschulman2 Much chiller version of John Collison

@johnschulman2 Michael observed an absence; value assessment requires further investigation, analysis.

@johnschulman2 well this is really cool.

@johnschulman2 nice one!

@johnschulman2 I did not expect this cross over

@johnschulman2 Great topic selection!

@johnschulman2 The Bell Labs comparison is fascinating, most research labs today optimize for publications while Bell optimized for breakthroughs that took decades to commercialize.

@johnschulman2 Excited to dive into these insights, especially on the future of RL. What were some surprising takeaways from the chat?

@johnschulman2 Very cool

@johnschulman2 Great thanks for that

@johnschulman2 Respect to Schuman, but Bell labs? There's levels to the game

@johnschulman2 I made a Gemini and Claude research team

@johnschulman2 Hell yeah, been waiting for the sequel to your Patrick Collison interview

@johnschulman2 @grok 帮我翻译这个视频内容并生成总结

@johnschulman2 Two normal innovating gentleman who have left their mark in technology history.

@johnschulman2 Have we confirmed Berkeley is an incredible powerhouse?

@johnschulman2 Amazing chat, incredibly insightful.

@johnschulman2 Merry Christmas ! All the best for you and your family.

@johnschulman2 Thank you for this!

@johnschulman2 Great watch! The “archetypes of research managers” + “peer review within the labs” sections are pure gold. Curious what you think changes most when research teams start using AI daily: hiring profile, review culture, or how ideas get selected?

@johnschulman2 the future of RL starts @ 41:40

@johnschulman2 The maturation of AI has shifted the required skillset from exploratory research taste towards strong software engineering, as much of the work now involves scaling ideas and building on existing infrastructure. Full episode notes here!

@johnschulman2 "i use cursor a lot and claude code" - we have the same tools!