[UCLA RL-LLM] Chapter 1.5: AlphaGo, test-time compute, and expert iteration

Ernest Ryu · 50:17

AlphaGo and AlphaGo Zero show that two-player zero-sum games are minimax problems whose training is unstable, and that superhuman play comes from combining learning with Monte Carlo tree search (MCTS) and expert itera...

Read the full summary on tuber

Redirecting...