RLHF

Outline the supervised fine-tuning, reward-modeling, and policy-optimization stages of RLHF and their main failure modes.