AI vs. Human Judgment of Content Moderation:LLM-as-a-Judge and Ethics-Based Response Refusals GPT-4o와 라마 모델, 윤리적 거부 응답에 32%포인트…