Human virologists have devised a rigorous test for AI, and the findings are astonishing: In the lab-based problem test, virology experts achieved an average accuracy rate of just 22.1% on a tailored subset of questions specific to their field. In contrast, the top-performing OpenAI model, o3, achieved an impressive accuracy rate of up to 43.8%, surpassing 94% of virologists on this particular subset of questions.
