OpenAI's latest model delivers powerful results but sometimes ignores simple directions, creating a tension between intelligence and control.
For many popular exams, recent score reports reflect not a surge in student mastery, but a quiet lowering of the bar.