LLMs show partial metacognitive ability in medical diagnosis, study finds

A new study published on arXiv explores the metacognitive abilities of large language models (LLMs) in medical reasoning, specifically focusing on diagnosing conditions like Alzheimer-type neurocognitive disorder (AT-NCD) and depression-related cognitive impairment (DRCI). Researchers developed a clinical benchmark using synthetic vignettes that varied evidence strength and completeness. In a pilot test with GPT-4.1 nano, the model demonstrated high diagnostic accuracy (93.5%) and a confidence level that generally tracked evidence quality and uncertainty, showing partial metacognitive sensitivity. However, the model exhibited errors in cases with moderate, conflicting evidence, where it shifted towards a less accurate diagnosis while maintaining high confidence, indicating localized calibration failures. AI

IMPACT This research suggests LLMs may be useful in medical diagnostics, but highlights the need for careful calibration and evaluation of confidence levels, especially in complex cases.

RANK_REASON The cluster contains an academic paper detailing research findings on LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Alzheimer-type neurocognitive disorder
• arXiv
• GPT-4.1 nano
• Hugging Face
• large language models

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →