Search

Type to search the site.

Reinforcement learning (RL)

Learning policies via rewards or scores from interacting with an environment: core to games, robotics, and some RLHF-style LLM tuning.

RL optimizes behaviors through trial signals: delayed rewards sparse, exploration vs. exploitation traps, simulations.

Gerald Tesauro’s TD-Gammon (1992) was an early landmark, teaching a neural network strong backgammon through temporal-difference learning and self-play. Modern game systems in the style of AlphaGo and LLM alignment sometimes use RL-style stages e.g., preference models RLHF with PPO alongside supervised learning logs.