The Rundown AI is reporting that a Harvard study published in the journal Science found OpenAI’s o1-preview model diagnosed patients more accurately than two attending emergency room physicians. It said the study involved 76 real ER cases where the AI model provided the correct diagnosis 67.1% of the time, while the two physicians achieved accuracy rates of 55.3% and 50.0%.
The Rundown AI said the model used only raw electronic health-record text to make its determinations. In one specific instance, the AI identified a rare flesh-eating infection in a transplant patient between 12 and 24 hours before the treating physician discovered it. The report noted that physician reviewers who scored the results were unable to distinguish between the diagnoses provided by the human doctors and those generated by the AI.
The Rundown AI is also reporting that the Pentagon has added eight artificial intelligence companies to its classified networks to accelerate the military's transition into an "AI-first fighting force." It said the companies included in the agreement are SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, AWS, and Oracle.
The report said Anthropic was excluded from the list due to a "supply-chain risk label," despite the Department of Defense seeking access to Anthropic’s Mythos model. It noted that Reflection, one of the newly added partners, received $2 billion in funding from 1789 Capital, a fund backed by Donald Trump Jr.
Additionally, The Rundown AI said Maryland has enacted the first U.S. ban on AI-driven grocery pricing, which prohibits stores from using personal data to mark up prices. It also reported that a Chinese court ruled that companies cannot legally justify firing employees simply to replace them with AI.