How RL training
fits together.
Some methods choose the feedback. Others decide how to update the model. They can work together.
Where feedback comes from
These methods differ in how they decide which answers deserve a higher score.
RLHF
People show what “better” means.
Learn from human preferences
Reinforcement Learning from Human Feedback RLVRPractice with an answer checker.
Learn from checkable results
Reinforcement Learning with Verifiable Rewards RLCDLearn from deliberately contrasting examples.
Learn from contrasting prompts
Reinforcement Learning from Contrastive Distillation RLAIFAn AI helps judge the practice.
Learn from AI-generated feedback
Reinforcement Learning from AI FeedbackHow the model updates
These algorithms use scores to change the model. They do not decide what is correct.