RLHF (Reinforcement Learning from Human Feedback)

A training technique where humans rate AI outputs, and the model learns to produce responses humans prefer. It's why modern AI feels more helpful and less robotic than earlier models.