NEW blog local-llm 1h ago · 
qiita-llm
LLM-as-a-Judgeを作る前に、まず人がログを読む重要性 This article argues that before building an LLM-as-a-Judge automated evaluator, humans sho…
AI要約 LLM-as-a-Judgeで自動評価を組む前に、まず人間が実際のログを読み込み、失敗パターンや評価軸を把握すべきだという主張。人手による分析がなければ、評価器自体の妥当性も担保できないと指摘する。
EN This article argues that before building an LLM-as-a-Judge automated evaluator, humans should first read actual logs to understand failure modes and evaluation criteria, since without manual analysis the judge itself cannot be validated.
og