Clinician use of language models diverges from how the models are evaluated
Study of 127,833 clinician queries finds LLM benchmarks poorly match real clinical assistant use.
Researchers analyzed 127,833 queries from 6,342 physicians, advanced practice providers, and nurses across 35 specialties sent to an institutional LLM assistant over eight months. Using the clinician-validated RCQ-Map framework, they found documentation and administration (36.2%) and knowledge retrieval (28.9%) accounted for nearly two-thirds of use, while diagnosis was 3.7%, and more than a third of queries could not be answered well as posed. Applied to 58 public benchmarks assembled as the Clinical AI Benchmark Atlas, the median benchmark contained no documentation requests and shared only 31% of the real-use task mix. Benchmarks designed to resemble clinical practice were no closer to actual use than those cited in frontier model reports.
- 127,833 queries from 6,342 clinicians in 35 specialties were analyzed.
- Documentation (36.2%) and knowledge retrieval (28.9%) dominated; diagnosis was 3.7%.
- Over one-third of queries were not answerable well as written.
- Median of 58 benchmarks shared only 31% of real task mix.
- Authors argue clinical AI evaluations should match actual clinician work.
From the original article
1 min read at huggingface.co
Large language model (LLM) assistants are being deployed to clinicians across health systems, and judgments about their readiness rest largely on benchmark scores, most of them derived from examination questions or curated cases. A benchmark predicts performance in deployment only to the extent that its items resemble real use, yet whether benchmarks reflect the work these systems receive has rarely been measured.
By Krithik Vishwanath, Haitong Lin, Anton Alyakin · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.