Buzzworthy developments of the past few days.

• ChatGPT has flunked a board-certification practice test. Specifically it was a self-assessment exam with 300 multiple-choice questions designed to prep gastroenterologists pursuing bona fides from the American Board of Internal Medicine. Publishing the results May 22, Arvind Trindade, MD, of Northwell Health and co-authors dryly dismiss ChatGPT: “We don’t recommend its use for medical education in gastroenterology in its current form.” The study is behind a paywall here. MedPage Today covers it in some detail here.
   
• Big Tech players like Google and Amazon have been far more successful than AI startups at driving healthcare AI into clinical settings. To find out exactly why, researchers spoke with heads of early-stage AI companies as well as execs at tech-forward investment firms. The investigators found several drags on startups’ efforts. Fragmented procurement processes at hospitals and health systems landed near the top. Azizi Seixas, PhD, of the University of Miami, senior author of the study behind the findings, says that, to make gains, “we need to add an innovation and technology infrastructure to better incorporate and implement these [startup] solutions into healthcare.” Study here, UMiami coverage here.
   
• Epic is partnering with several leading healthcare systems to bring generative AI of the large-language kind to patient service at scale. The technology rollout will start with administrative communications and, if it goes well, may expand to clinical areas. UNC Health in North Carolina is lead-piloting the system. UC-San Diego Health, University of Wisconsin Health and Stanford Medicine are planning to closely follow suit. UNC news item.
   
• AI models and machine-learning algorithms have identified small proteins that could help target and treat glaucoma and macular degeneration. The advance in drug-delivery prediction is being innovated at Johns Hopkins with an assist from the University of Maryland. Johns Hopkins news item here.
   
• Intelligent Medical Objects (Chicago) has snapped up Melax Tech (Houston) with an eye on refining quality in healthcare data. The combined might of the two will equip IMO’s care and research clients with natural language processing and other AI flavors so they can build solutions from word-based data. IMO also will now turn its attention to offering AI capabilities to payers, pharma operations and life-sciences companies. Announcement.
   
• Oral healthcare AI outfit Kells (New York) is partnering with Liberty Dental Plan (Irvine, Calif.) to bring AI-based services to underserved communities in the Northeast. Kells says the arrangement will help Liberty use AI to create risk profiles, identify treatment needs and coordinate follow-up care. Announcement here.
   
• Researchers have used AI to create a tool for rendering sign language as text. The resulting product may help solve any number of challenges faced by deaf people. Coverage by New Atlas here.
   
• Verge Genomics of South San Francisco is tapping Emerald Innovations of Cambridge, Mass., to test an AI-sourced drug designed to counterpunch amyotrophic lateral sclerosis (ALS, aka Lou Gehrig’s disease). Researchers will place Emerald’s wireless digital health sensors in study participants’ homes to continuously measure key neurological functions affected by ALS, according to a Verge news release. The sensors grab and relay data on sleeping, breathing and walking. From this data will come AI-aided assessments of changes in patients being treated with increasing doses of a promising drug.
   
• January AI (Menlo Park, Calif.) has debuted an app equipped with generative AI that tells people when their glucose levels are likely to rise or fall. It does so without input from a glucose monitor, basing its predictions on planned activities as well as anticipated foods. The company is selling the app straight to consumers for $288 for the first month plus $28 a month from then on. Details.  
   
• Microsoft is priming the public for a June release of Copilot, its new AI assistant for Windows 11. The feature will be much more capable than Cortana, the cutesy chatbot it makes obsolete. Or so the company hints in a May 23 blog post directed at software developers. “Once open, the Windows Copilot side bar stays consistent across your apps, programs and windows, always available to act as your personal assistant,” writes Microsoft chief product officer Panos Panay. “It makes every user a power user.” Good coverage by tech journalist Chris Velazco of the Washington Post here.