AI Glossary

Reinforcement learning from AI feedback

Also called RLAIF

What does “Reinforcement learning from AI feedback” mean?

Reinforcement learning from AI feedback: Works much like RLHF, except that most preference judgments come from an AI model instead of human raters, often guided by a written set of principles. It makes feedback cheaper to collect at scale, but the results are only as good as the model doing the judging.

Reinforcement learning from AI feedback visual guide: Works much like RLHF, except that most preference judgments come from an AI model instead of human raters, often guided by a written set of principles. It makes feedback cheaper to collect at scale, but the results are only as good as the model doing the judging.
Visual guide Open full size (new tab)

What is an example of “Reinforcement learning from AI feedback”?

Anthropic's Constitutional AI approach uses AI feedback, guided by a list of principles, to train models toward less harmful behavior.

Keep learning

See also

Reinforcement learning from human feedbackTermReward modelTermSynthetic dataTermDirect preference optimizationTerm

Same topic

RegularizationTermReinforcement learning from verifiable rewardsTermQLoRATermPre-trainingTermScaling lawsTerm