LLM Judges Verify Presence, Not Absence: Omission Blindness in AI Clinical Notes
Researchers found that large language models used for analyzing electronic health records tend to confirm information that is explicitly stated while overlooking data that is missing, a phenomenon called omission blindness. The study shows that these models may incorrectly assume absent findings are normal, potentially leading to inaccurate clinical interpretations. Strategies to improve detection of omitted information are discussed.