RLHF (2017)

Reinforcement Learning from Human Feedback, introduced by Christiano et al. (2017). Uses human preferences to define reward signals. Critical for aligning LLMs like ChatGPT and Claude.