Researchers affiliated with Lille University Hospital and other institutions in France have published a paper titled "Hallucination by proxy in LLM-assisted differential diagnosis." This study investigates how fictitious disease names, presented plausibly by large language models (LLMs), influence physicians' differential diagnosis processes. The study reports that 44% of medical residents trusted non-existent disease names generated by LLMs. Additionally, the research examined the differences between physicians who were able to identify the fictitious names and those who were not.
French University Hospital Study Finds 44% of Residents Trusted Fake Disease Names Generated by LLMs
This article is a translation. Read the Japanese original
Source: AIが生成した“存在しない病名”、研修医の44%が信用 見抜けた医師との差は? 仏大学病院が検証 (ITmedia AI+, 2026-08-31)