
An AI system that simulates a panel of physicians reached the correct diagnosis in about 80% of test cases, compared with 36% to 47% for generalist physicians, though researchers caution the results come from a controlled, text-only benchmark rather than real-world practice.








