RLAIF (Reinforcement Learning from AI Feedback)
Concept
A training method where an AI model verifies and improves other AI outputs. It's considered distinct from RLHF and potentially works if verification is easier for the AI than generation.
Mentioned in 1 video
