RLHF

Reinforcement Learning from Human Feedback: training a model to prefer answers that human reviewers rated as better.

Why it matters

RLHF is a big part of why modern assistants feel helpful and polite rather than raw.