Charlie Steiner comments on Paper: Constitutional AI: Harmlessness from AI Feedback (Anthropic)