Skip to content
Gone rogue · 1 source ·

ChatGPT Gets 88% of Mammogram Advice Right, Cites Journal Articles That Don't Exist

Ourtake

Roughly one answer in eight fell short, and the chatbot occasionally footnotes sources that were never written.

What happened

A study published in Radiology tested ChatGPT with 25 breast cancer screening questions, asking each three times. Three radiologists rated 88% of answers appropriate, but the chatbot gave one outdated answer about mammograms and COVID-19 vaccines, gave inconsistent answers to some questions, and sometimes made up fake journal articles. The researchers advised people to rely on doctors rather than the chatbot.

Source

Report a mistake