Quintin Pope comments on Large language models can provide “normative assumptions” for learning human preferences