DarioAmodei comments on Learning from Human Preferences—from OpenAI (including Christiano, Amodei & Legg)