What abbreviation names the alignment technique that trains a reward model on human preference comparisons, then fine-tunes a language model against it with reinforcement learning?

RLHF

RLHF (Reinforcement Learning from Human Feedback) became the standard recipe for aligning chatbots: annotators rank model outputs, a reward model learns to predict those rankings, and the policy is optimized, typically with PPO, to maximize reward. It is credited with boosting helpfulness but blamed for reducing output diversity ("mode collapse").