LLM Judges Miss What AI Scribes Leave Out of Clinical Notes
LLM Judges Verify Presence, Not Absence: Omission Blindness in AI Clinical Notes
A new benchmark reveals a critical blind spot in how AI scribes are evaluated: LLM judges reliably catch added or altered content in clinical notes but fail to detect omissions—the dominant error in ambient AI scribes. Across eight judge designs, detection of omissions barely exceeded chance, while restructuring the task—listing facts from the transcript, then checking the note—recovered performance. A per-fact pipeline flagged missing facts at 2.7% false alarms, while a single-call prompt detected more (36.9% vs 24.6%) at 6.2% false alarms and a tenth of the cost. Physician validation favored the pipeline on disagreements. The benchmark, prompts, and judgements are released.
LLM judges verify presence, not absence: they reliably flag added or altered content but miss omissions, the dominant error in AI-scribed clinical notes.
- Planktonne
The fact that even the abstract is very clearly AI-generated does not fill me with confidence in the rigour of the research.
- velrim
Not sure about the paper but the results make sense, we see it in PDF extraction too. Fields that aren't in the document are being made up 11 to 40% of the time depending on the API.
The whole thing is nasty partly because it isn't just AI problem. When checking the human labels that we used in our evals 40 out of 142 answer keys claiming absence were wrong. Tricky one.
- udik2
May be AI written. Used my AI to get the gist of it. One thing that I believe is LLMs need to be considered a pure play tool and can be guided to point out misses as well. The reason is simple. To find whats missing, one needs to ask the right questions on how to judge and when that is there.