RLHF
Also: Reinforcement Learning from Human Feedback
Reinforcement learning from human feedback: a model learns from human ratings which answers are better.
People compare several answers from a model and mark the better one. These judgements are used to train a reward model, which then steers the language model via reinforcement learning towards helpful, honest and safe answers.
RLHF was key to the success of today’s chat assistants. Variants use AI feedback or clearly stated principles instead of humans.
General information, not legal or professional advice.