3 ms·
I'm watching an AI agent right now learn to play Texas Hold'em and after 75,000 hands, starting with zero knowledge, it plays about as well as some of my friend
by bluetwo 9y ago
I'm watching an AI agent right now learn to play Texas Hold'em and after 75,000 hands, starting with zero knowledge, it plays about as well as some of my friends.
What should I try to teach it next?
- carlmr 9y agoFinding the cure for cancer.
- AareyBaba 9y agoWash my dishes and fold my laundry.
- yazr 9y ago75K hands seems very low to me Can you give more details ? Is it HU ? NL ? Using just self-play from 0-knowledge ? How many roll-outs do u perform for each game action?
- bluetwo 9y agoHome-brewed AI, not a NN. Similar in some ways to AlphaGo Zero's MCTS. Learns to play with 0 knowledge. No training data. No rollouts. It squeezes a lot of info from each hand, more than a NN, which is why it needs less trials. Currently 100k hands and most respectable play.
- yazr 9y agoWould love to chat and exchange some info yazr2yazr@gmail.com