tl;dr
RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback