AI Finds Hidden Errors in Medical Records, But Struggles at Scale
A new study on ArXiv shows that a two-stage Gemini AI pipeline can detect internal inconsistencies in 3,000 MIMIC-IV-Note discharge summaries, but recurring failure modes limit its reliability for automated clinical use.
