RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback https://arxiv.org/abs/2309.00267