AdaRubric: LLMエージェント評価のためのタスク適応型ルーブリック
原題: AdaRubric: Task-Adaptive Rubrics for LLM Agent Evaluation
分析結果
- カテゴリ
- AI
- 重要度
- 69
- トレンドスコア
- 28
- 要約
- LLMを審査者として用いる評価は、固定されたルーブリックではエージェントのタスクを適切に評価できません。コードデバッグには正確性とエラーハンドリングが求められます。
- キーワード
arXiv:2603.21362v2 Announce Type: replace Abstract: LLM-as-Judge evaluation fails agent tasks because a fixed rubric cannot capture what matters for this task: code debugging demands Correctness and Error Handling; web navigation demands Goal Alignment and Action Efficiency. We present ADARUBRIC, which closes this gap by generating task-specific evaluation rubrics on the fly from task descriptions, scoring trajectories step-by-step with confidence-weighted per-dimension feedback, and filtering preference pairs with the novel DimensionAwareFilter - a provably necessary condition for preventing high-scoring dimensions from masking dimension-level failures. On WebArena and ToolBench, ADARUBRIC achieves Pearson r=0.79 human correlation (+0.16 over the best static baseline) with deployment-grade reliability (Krippendorff's $\alpha$=0.83). DPO agents trained on ADARUBRIC preference pairs gain +6.8 to +8.5 pp task success over Prometheus across three benchmarks; gains transfer to SWE-bench code repair (+4.9 pp) and accelerate PPO convergence by +6.6 pp at 5K steps - both without any rubric engineering. Code: https://github.com/alphadl/AdaRubrics. arXiv:2603.21362v2 Announce Type: replace Abstract: LLM-as-Judge evaluation fails agent tasks because a fixed rubric cannot capture what matters for this task: code debugging demands Correctness and Error Handling; web navigation demands Goal Alignment and Action Efficiency. We present ADARUBRIC, which closes this gap by generating task-specific evaluation rubrics on the fly from task descriptions, scoring trajectories step-by-step with confidence-weighted per-dimension feedback, and filtering preference pairs with the novel DimensionAwareFilter - a provably necessary condition for preventing high-scoring dimensions from masking dimension-level failures. On WebArena and ToolBench, ADARUBRIC achieves Pearson r=0.79 human correlation (+0.16 over the best static baseline) with deployment-grade reliability (Krippendorff's $\alpha$=0.83). DPO agents trained on ADARUBRIC preference pairs gain +6.8 to +8.5 pp task success over Prometheus across three benchmarks; gains transfer to SWE-bench code repair (+4.9 pp) and accelerate PPO convergence by +6.6 pp at 5K steps - both without any rubric engineering. Code: https://github.com/alphadl/AdaRubrics.