Bridging Human and LLMJudgments: Understanding and Narrowing the Gap

Large language models are increasingly used as judges (LLM-as-a-judge) to evaluate model outputs at scale, but their assessments often diverge systematically from human judgments.