⑥難易度制御性 ⑦難易度境界[6] ベンチマーク レベル ⑧有効性[7] ⑨頑健性 ⑩効率性 [1] Wu et al,. Unigen: A unified framework for textual dataset generation using large language models. CoRR, abs/2406.18966. [5]Vivek et al., Anchor points: Benchmarking models with much fewer examples, EACL2024 [2] Perez et al,. Discovering language model behaviors with model-written evaluations. ACL 2023, [6]Patel et al., How to get your llm to generate challenging problems for evaluation. arXiv preprint arXiv:2502.14678. [3] Montahaei et al., Jointly measuring diversity and quality in text generation models. CoRR, abs/1904.03971. [4]Chan et al., Scaling synthetic data creation with 1,000,000,000 personas. CoRR, abs/2406.20094. [7]Perlitz et al., Benchmark agreement testing done right: A guide for LLM benchmark evaluation. CoRR, abs/2407.13696. 7
⑥難易度制御性 ⑦難易度境界[6] ベンチマーク レベル ⑧有効性[7] ⑨頑健性 ⑩効率性 評価⽅法:LLM-as-a-Judgeによって 0-1 のスコアリング バイアスの相関分析: スコアにJudgeのReasoning⻑に起因するバイアスが存在 問題の難しさ・⽂⻑ 相関 Reasoning⻑を含む回帰分析 JudgeのReasoning⻑ 問題の難しさ・⽂⻑ 相関 ⾮有意 Judgeのスコア Judgeのスコア →本論⽂では,重回帰分析により「Reasoning⻑によるスコア変動」を推定し,その影響を除去 [1] Wu et al., Unigen: A unified framework for textual dataset generation using large language models. CoRR, abs/2406.18966. [5] Vivek et al., Anchor points: Benchmarking models with much fewer examples, EACL 2024 [2] Perez et al., Discovering language model behaviors with model-written evaluations. ACL 2023, [6] Patel et al., How to get your llm to generate challenging problems for evaluation. arXiv preprint arXiv:2502.14678. [3] Montahaei et al., Jointly measuring diversity and quality in text generation models. CoRR, abs/1904.03971. [4] Chan et al., Scaling synthetic data creation with 1,000,000,000 personas. CoRR, abs/2406.20094. [7] Perlitz et al., Benchmark agreement testing done right: A guide for LLM benchmark evaluation. CoRR, abs/2407.13696. 8