AI models struggle to accurately identify genetic conditions from patient-written descriptions.

NIH researchers discovered that AI models struggle to accurately identify genetic conditions from patient-written descriptions, despite accurately diagnosing from textbook-like descriptions. Testing 10 different large language models, including ChatGPT, the team found that accuracy ranged widely, dropping significantly when analyzing patient-written summaries. This highlights the need to improve AI tools for healthcare applications.

August 14, 2024
4 Articles

Further Reading