points
Bai et al. "Constitutional AI: Harmlessness from AI Feedback" https://arxiv.org/pdf/2212.08073