Training

RLHF

Reinforcement learning from human feedback optimizes a model using preferences derived from people.

Reinforcement learning from human feedback optimizes a model using preferences derived from people. Values shown in the worked visual are illustrative unless the labels state otherwise.

How it works

A preference model or related signal turns comparisons into a training objective. Policy optimization then shifts outputs toward preferred behavior while constraining drift.

A concrete example

Illustrative: reviewers prefer response A in 7 of 10 pairwise comparisons, producing preference data for training.

Do not confuse it with Supervised fine-tuning

SFT imitates target responses; RLHF optimizes a learned preference signal.

Study the full lesson