A Ten-Minute AI Test Changes How We Should Use It
A multi-institution study across math and reading tasks found that after 10 minutes of AI-assisted work, users who lost access to AI performed worse than those who never used it; using AI for hints or clarifications yielded no impairment, while asking it to solve problems did. This shifts the question from whether to use AI to how you use it—prefer facts, direction, or sanity checks over outsourcing thinking.