S3 E14 OpenAI Research Scientist Noam Brown on Solving Poker and Diplomacy with AI
What's in this episode:
00:00:00 Noam Brown
00:01:20 sponsors: Index Ventures and Weights and Biases
00:02:19 why should AI be concerned with playing games?
00:05:15 going from AI winning in Chess to AI winning in Poker
00:08:54 no single optimal action in Poker: the need for learning probabilistic strategies
00:13:20 the core algorithm: fictitious play / counter-factual regret minimization
00:15:00 analyzing the AI’s poker strategy
00:18:14 role of neural nets
00:19:59 role of search
00:29:32 how to run search in poker, what’s even the state
00:37:19 human poker strategies and modeling the other players
00:43:10 optimizing for winning the hand vs. for winning the tournament
00:47:01 are your bots playing online poker
00:51:15 Diplomacy
01:07:17 real world negotiations and strategy
01:11:10 planning, reasoning, self-play in large language models
01:19:03 initial interest in AI and career trajectory
01:23:37 ways to relax
Links:
/ polynoamial
noambrown.github.io/
scholar.google.com/citations?...
Sponsor Links:
www.indexventures.com/
wandb.com
Subscribe and listen to our podcast today:
Apple: bit.ly/44jx4aQ
Spotify: bit.ly/3JywnCo
Amazon: bit.ly/44kjN1N
Google: bit.ly/3NRw3kM
Acast: bit.ly/3JAQKPC
Host: Pieter Abbeel -- www.cs.berkeley.edu/~pabbeel
Production: Bo Obradovic