Summary
The episode works through four questions that students find difficult when they use AI to study. It examines why a smooth practice session with an assistant can overstate learning, drawing on Bastani and colleagues (2025) and a made-up calibration example, and it puts the Harvard tutoring trial by Kestin and colleagues (2025) in perspective by converting effect sizes into percentiles and weighing the trial's limits. In a harder worked example, a generated practice-quiz key cites a genuine quotation whose rare-outcome condition does not hold, and recomputing the odds ratio exposes the error. The episode also reads the survey evidence linking AI use to lower critical thinking with attention to what a cross-sectional design can show, and it closes with a disagreement about hints that ends in a shared rule for asking for help.