Recursive Reward Model

Using AI to assist in generating feedback for another AI (RLHF)

Method

1. Train an AI assistant using human feedback (RLHF)
2. Use AI assistants on harder tasks, to train new models for those harder tasks