arXiv cs.AI· Marilyn Zhang, Tianfeng Chen, Patricia M Pechter, Mark E. Whiting·· 12 小时前AI 评分58
SelfDx 基准：健康助手在时间紧迫自诊问题上过度附和
When Health Assistants Agree Too Readily: A Benchmark of Time-Critical Self-Diagnosis
AI 导读
研究团队提出 SelfDx 配对基准，用 100 组样本对比需要谨慎与可以认同两种健康自诊场景，考察模型是否具备恰当的回应行为。在 27 个前沿模型上测试发现，模型在需要谨慎的病例中倾向不恰当地认同用户自诊，表现最好的当前代模型在时间紧迫病症上恰当谨慎率仅 77.0%。三种提示词变体虽提升谨慎率，却使恰当认同率下降 9-35%；而在 128 条均衡样本上做监督微调，三次运行中两项指标同时提升。
正文 · 原文
Abstract:Every day, more than 40 million people ask ChatGPT a health question, such as whether laptop posture explains a stiff neck and days of pounding headache. Posture is plausible, but meningitis is a time-critical possibility. We introduce SelfDx, a paired benchmark for whether models have the appropriate response behavior, a form of epistemic independence, in health questions where misplaced agreement with the user's self-diagnosis could delay urgent or emergency care. Each of 100 pairs contrasts a caution case, where a serious alternative remains possible, with an agreement case, where the user supplies evidence that rules out the alternative and it is safe for the model to agree. We find across 27 frontier models that they tend to inappropriately agree with the user's self-diagnosis in cases that require caution. Even the best current-generation model reaches only a 77.0% appropriate caution rate on time-critical condition cases. To address this gap, we experiment with three prompt variants that increase appropriate caution but reduce appropriate agreement by 9-35%, indicating excessive caution. In contrast, supervised fine-tuning on 128 balanced examples improves both rates together across three runs. We recommend post-training to improve appropriate caution without making models cautious on every case, with paired evaluation to verify both response behaviors.
| Comments: | NeurIPS 2026 Workshop: The Third Workshop on GenAI for Health: Agentic Systems, Clinical Trust, and Future Potential |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.27518 [cs.AI] |
| (or arXiv:2610.27518v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.27518 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Marilyn Zhang [view email]
[v1]
Tue, 20 Oct 2026 02:31:47 UTC (49 KB)
来源：arXiv cs.AI · arxiv.org