Reward model

A model that scores outputs by how well they match preferences. It is used in RLHF and to pick the best of several answers.

All terms, A to Z