What Is RLCD? Reinforcement Learning from Contrast Distillation

What it is

A plain-language explainer on RLCD, a way of aligning language models by learning from contrasting outputs rather than human ratings alone.

Good background reading if you want to understand how the models you use are actually steered.

Added Sep 17, 2026 · 0 opens