IBM and Nvidia move quantum processors into supercomputing workflows
New Slurm integration, a Fermilab co-design platform and GPU-accelerated error-correction tools aim to make quantum processors usable inside established HPC environments.
AI / NEWS BRIEF
OpenAI has released healthcare-focused computer-use capabilities across its API and clinical products, with configurable approvals, audit logs and new performance results that remain provider-reported.
OpenAI says healthcare organizations, clinicians and developers can now use computer-use agents through the OpenAI API, ChatGPT for Healthcare and ChatGPT for Clinicians. The agents navigate software by reading screens, clicking and typing, allowing them to move information among electronic health records, payer portals and other systems. Deployments can limit application access, require human approval for selected actions and retain audit logs. OpenAI says protected-health-information workflows require eligible services covered by a business associate agreement plus the required security and retention settings.
OpenAI reports that physicians rated 99% of 325 reviewed GPT-6 Astra clinical tasks safe, while 93% of 646 assessments called the output very or extremely helpful. On Stanford's synthetic HealthAdminBench environments, OpenAI reports Astra earned 94.3% of rubric points and Luna earned 91.8% at a stated cost of 8.5 cents per task for Luna. HealthAdminBench contains 135 expert-defined tasks across four simulated interfaces. These are provider-run results on bounded test environments, not evidence that autonomous agents are safe across live clinical systems.
This is a consequential shift from answering medical questions to taking steps inside operational systems. Human review, access controls and auditability reduce risk but do not eliminate errors caused by ambiguous records, interface changes or missing context. The practical standard should be end-to-end completion, detection of unsafe uncertainty and recovery from failed actions—not benchmark rubric points alone. Independent evaluation is needed before the reported figures can be treated as general clinical or administrative reliability.
New Slurm integration, a Fermilab co-design platform and GPU-accelerated error-correction tools aim to make quantum processors usable inside established HPC environments.
Atlas moves beyond a model demonstration into public tools for reconstructing, shaping and exporting explorable 3D scenes, with developer endpoints intended for creative, simulation and robotics workflows.