GLOSSARY TERM
What is Reward Model?
A neural network trained to predict human preference scores for AI outputs.
A reward model evaluates the quality, safety, or helpfulness of a generative response. It acts as the automated adjudicator in reinforcement learning pipelines, translating subjective human values into computable gradient updates.
Automate Quality Scoring
Implement robust reward pipelines to safeguard internal private models.