Medical AI is weakest at clinical decision support and admin workflows — exactly where hospitals need it most
jonc101x · x · 2026-10-28
Part 9/14 of Stanford's medical AI benchmark thread: these benchmarks grew out of MedHELM, where clinical decision support and administration/workflow were among the weakest areas of medical AI performance — ironically exactly where health systems are looking for the most help. The cited tweet gives HealthAdminBench numbers: Claude Opus 4.6 at 36.3% end-to-end, GPT-5.4 at 82.8% subtasks but only 26.7% full workflows.
Related event: HealthAdminBench shows AI agents struggle on healthcare admin workflows(2 posts)→
More from Research
- Sweeping every Qwen3-8B neuron, rated for weirdness by Haiku — Sauers_ · 2026-10-28
- Follow-up repo machine-verifies Hamiltonian cycle results on top of OpenAI/Anthropic ETH refutation claims — xennygrimmato_ · 2026-10-28
- DiscoTrace wins EMNLP 2026 Best Paper: LLM answers lack humans' community-specific rhetorical diversity — TuhinChakr · 2026-10-28
- New arXiv paper frames world model representation as reachability, admission, assignment — SoloGen · 2026-10-28
- OpenAI shares internal frontier model math results, Terence Tao responds in meme'd thread — krisspy39 · 2026-10-28
- Divergence Labs and Anthropic use Claude to discover antibody therapy for food allergy — anshulkundaje · 2026-10-28