This research evaluates how large language models respond to complex cancer-related questions from real patients, with special attention to questions containing false presuppositions.

Key findings:

• Researchers developed Cancer-Myth, a dataset of authentic patient questions reviewed by oncologists
• LLMs frequently fail to identify and correct false assumptions in patient questions
• The study revealed significant gaps in how AI handles nuanced medical inquiries with potentially harmful misconceptions
• Results highlight the need for improved AI safety mechanisms before deployment in medical contexts

This research is critical for healthcare as patients increasingly turn to AI for medical information, making the accurate handling of misconceptions a patient safety concern.

Cancer-Myth: Evaluating AI Chatbot on Patient Questions with False Presuppositions