EpochLex entry

Reinforcement Learning with Verifiable Rewards

ree-in-FORS-ment LUR-ning with VAIR-uh-fy-uh-bul ree-WORDZ

A post-training approach that uses reinforcement learning with reward signals derived from outcomes that can be checked automatically or objectively, such as whether a mathematical answer or program is correct.