John Schulman on dead ends, scaling RL, and building research institutions

https://is1-ssl.mzstatic.com/image/thumb/Podcasts221/v4/92/8d/46/928d46e9-4de2-f84c-f1b6-e37ebd9230c2/mza_10803927947408564186.jpg/600x600bb.jpg

Cursor

3 episodes

1 week ago

A collection of conversations from the team at Cursor.

All content for Cursor is the property of Cursor and is served directly from their servers with no modification, redirects, or rehosting. The podcast is not affiliated with or endorsed by Podjoint in any way.

A collection of conversations from the team at Cursor.

Technology

https://d3t3ozftmdmh3i.cloudfront.net/staging/podcast_uploaded_nologo/43785841/43785841-1766014295238-d5290ab05494b.jpg

John Schulman on dead ends, scaling RL, and building research institutions

Cursor

51 minutes 25 seconds

2 weeks ago

John Schulman on dead ends, scaling RL, and building research institutions

A conversation with John Schulman on the first year LLMs could have been useful, building research teams, and where RL goes from here.00:00 - Speedrunning ChatGPT09:22 - Archetypes of research managers11:56 - Was OpenAI inspired by Bell Labs?16:54 - The absence of value functions18:23 - Continual learning21:09 - Brittle generalization24:05 - Co-training generators and verifiers, GANs27:06 - John’s personal use of AI for research28:54 - Day in the life33:01 - Slowdowns in consequential ML ideas36:21 - "Peer review" within the labs39:19 - Distribution shift in researchers43:33 - Future of RL45:33 - Will the labs coordinate if the world needs them to?44:46 - Forecasting ills in AGI and engineering47:53 - Thinking Machines