Doximity Ask news and updates

Doximity launched DoseCheck, a rule based check that tests every dosing claim in a Doximity Ask answer against a structured drug database built from FDA labeling and professional guidelines, instead of asking a model to grade itself. In Doximity's test of nine widely used AI models on more than 8,000 real physician dosing questions, at least 2.3 percent of dosing claims were wrong and 43.8 percent of those errors were serious or potentially lethal. DoseCheck is free and live for all Ask users on mobile and desktop.

Our read

Dosing is where a confident wrong answer does the most harm, and a deterministic check against labeling is a stronger guard than a second model. Doximity says it lifts dosing accuracy above 99 percent; the testing is its own and has not been published.

Doximity published Bedside Bench, an open source clinical AI benchmark of 500 cases across 10 sub benchmarks, including drug safety, guideline adherence, landmark trials, hallucination and false premises, health equity and diagnostic safety. The 250 case training split and the grading rubrics are public on Hugging Face, and the test split is held out. Bedside Bench is one of the domain benchmarks in Fireworks' Specialized Intelligence Index, and Doximity states that Doximity Ask ranked first against frontier models when Fireworks graded it.

Our read

Buyers get a public rubric they can run against Doximity Ask and competing clinical reference tools. The ranking is stated without published scores and the benchmark was written by Doximity, so ask for the per benchmark numbers before relying on it.