Skip to main content

Mapping the susceptibility of large language models to medical misinformation across clinical notes and social media: a cross-sectional benchmarking analysis

Share:

More than 3.4 million prompts were used to test how well LLMs could detect misinformation in clinical notes and social media posts. Overall, LLMs failed to detect misinformation 31.7% of the time, with some models performing better than others. Misinformation in discharge notes, which contains authoritative or formal language, was missed 46.1% of the time, whereas misinformation on social media was missed only 8.9% of the time.