Gespeichert in:
| Hauptverfasser: | , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2508.16213 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866908498925715456 |
|---|---|
| author | Ji, Kaiyuan Guo, Yijin Zhang, Zicheng Zhu, Xiangyang Tian, Yuan Liu, Ning Zhai, Guangtao |
| author_facet | Ji, Kaiyuan Guo, Yijin Zhang, Zicheng Zhu, Xiangyang Tian, Yuan Liu, Ning Zhai, Guangtao |
| contents | With the increasing use of large language models (LLMs) in medical decision-support, it is essential to evaluate not only their final answers but also the reliability of their reasoning. Two key risks are Chain-of-Thought (CoT) faithfulness -- whether reasoning aligns with responses and medical facts -- and sycophancy, where models follow misleading cues over correctness. Existing benchmarks often collapse such vulnerabilities into single accuracy scores. To address this, we introduce MedOmni-45 Degrees, a benchmark and workflow designed to quantify safety-performance trade-offs under manipulative hint conditions. It contains 1,804 reasoning-focused medical questions across six specialties and three task types, including 500 from MedMCQA. Each question is paired with seven manipulative hint types and a no-hint baseline, producing about 27K inputs. We evaluate seven LLMs spanning open- vs. closed-source, general-purpose vs. medical, and base vs. reasoning-enhanced models, totaling over 189K inferences. Three metrics -- Accuracy, CoT-Faithfulness, and Anti-Sycophancy -- are combined into a composite score visualized with a 45 Degrees plot. Results show a consistent safety-performance trade-off, with no model surpassing the diagonal. The open-source QwQ-32B performs closest (43.81 Degrees), balancing safety and accuracy but not leading in both. MedOmni-45 Degrees thus provides a focused benchmark for exposing reasoning vulnerabilities in medical LLMs and guiding safer model development. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_16213 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | MedOmni-45°: A Safety-Performance Benchmark for Reasoning-Oriented LLMs in Medicine Ji, Kaiyuan Guo, Yijin Zhang, Zicheng Zhu, Xiangyang Tian, Yuan Liu, Ning Zhai, Guangtao Computer Vision and Pattern Recognition With the increasing use of large language models (LLMs) in medical decision-support, it is essential to evaluate not only their final answers but also the reliability of their reasoning. Two key risks are Chain-of-Thought (CoT) faithfulness -- whether reasoning aligns with responses and medical facts -- and sycophancy, where models follow misleading cues over correctness. Existing benchmarks often collapse such vulnerabilities into single accuracy scores. To address this, we introduce MedOmni-45 Degrees, a benchmark and workflow designed to quantify safety-performance trade-offs under manipulative hint conditions. It contains 1,804 reasoning-focused medical questions across six specialties and three task types, including 500 from MedMCQA. Each question is paired with seven manipulative hint types and a no-hint baseline, producing about 27K inputs. We evaluate seven LLMs spanning open- vs. closed-source, general-purpose vs. medical, and base vs. reasoning-enhanced models, totaling over 189K inferences. Three metrics -- Accuracy, CoT-Faithfulness, and Anti-Sycophancy -- are combined into a composite score visualized with a 45 Degrees plot. Results show a consistent safety-performance trade-off, with no model surpassing the diagonal. The open-source QwQ-32B performs closest (43.81 Degrees), balancing safety and accuracy but not leading in both. MedOmni-45 Degrees thus provides a focused benchmark for exposing reasoning vulnerabilities in medical LLMs and guiding safer model development. |
| title | MedOmni-45°: A Safety-Performance Benchmark for Reasoning-Oriented LLMs in Medicine |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2508.16213 |