KnowBench: Effort Reduction as a Unified, Deployment-Grounded Benchmark for Clinical AI
KnowBench measures clinical AI by what clinicians accept after review, not by similarity to a reference answer.
The paper defines “Effort Reduction” as the share of generated clinical work product accepted by the responsible clinician under expert and safety review. It applies the same construction across notes, codes, orders, chart summaries, after-visit summaries, and decision support. The authors report one documentation result: Knowtex models reached 97.99% aggregate ER across more than one million signed encounters, over six-plus months and 13 specialties. Some companion statistics are withheld, and the release says its reporting checklist is only partial. ArXiv · AI/CL/LG's note
The paper defines “Effort Reduction” as the share of generated clinical work product accepted by the responsible clinician under expert and safety review. It applies the same construction across notes, codes, orders, chart summaries, after-visit summaries, and decision support. The authors report one documentation result: Knowtex models reached 97.99% aggregate ER across more than one million signed encounters, over six-plus months and 13 specialties. Some companion statistics are withheld, and the release says its reporting checklist is only partial. ArXiv · AI/CL/LG's note
score 5