Limitations of large language models in clinical problem-solving arising from inflexible reasoning
5.5
来源：
Nature
关键字：
EEG
发布时间：
2025-11-12 07:34
摘要：
The study investigates the limitations of large language models (LLMs) in clinical problem-solving, revealing their poor performance in scenarios requiring flexible reasoning. The introduction of the Medical Abstraction and Reasoning Corpus Question and Answer (mARC-QA) benchmark aims to assess these limitations. Results show that LLMs often lack commonsense medical reasoning and exhibit overconfidence in their outputs, highlighting the need for caution in clinical applications. The findings underscore the importance of developing rigorous benchmarks to evaluate LLM reasoning capabilities in medical contexts.
原文：
查看原文
价值分投票
评分标准
新闻价值分采用0-10分制，综合考虑新闻的真实性、重要性、时效性、影响力等多个维度。
评分越高，表示该新闻的价值越大，越值得关注。
价值维度分析
domain_focus
0.5
business_impact
0.0
scientific_rigor
1.5
timeliness_innovation
1.5
investment_perspective
2.0
market_value_relevance
0.0
team_institution_background
0.5
technical_barrier_competition
0.0
关键证据
LLMs perform poorly compared to physicians on mARC-QA.
The study introduces a new benchmark to assess clinical reasoning.
Findings indicate significant gaps in LLM reasoning capabilities.
真实性检查
否
AI评分总结
The study investigates the limitations of large language models (LLMs) in clinical problem-solving, revealing their poor performance in scenarios requiring flexible reasoning. The introduction of the Medical Abstraction and Reasoning Corpus Question and Answer (mARC-QA) benchmark aims to assess these limitations. Results show that LLMs often lack commonsense medical reasoning and exhibit overconfidence in their outputs, highlighting the need for caution in clinical applications. The findings underscore the importance of developing rigorous benchmarks to evaluate LLM reasoning capabilities in medical contexts.