A new study finds heavy reliance on generative AI splits students into clear winners and losers, and the split shows up early in the semester, not at the end.
Researchers surveyed 118 students across 12 AI-related courses at one institution and sorted them into four usage clusters. High-use students reported the most academic benefit. Light users reported the least. Two moderate-use groups landed in between, differing mainly in how much benefit they felt they got for the effort they put in. Usage patterns also varied by free versus premium tool access, single versus multiple-tool habits, and the policy an individual instructor set for that course.
The regression results are the real story. Leaning on AI early in a course predicted more academic benefit, but also more negative impact, and that negative effect got stronger the better a student was at evaluating AI output. In other words, knowing how to judge an AI's answer does not protect a student who starts depending on it too soon; it can make the eventual cost worse.
It's a sharper version of the decades old "calculator in class" debate, except this tool can also write the essay, not just do the math, and it grades its own homework along the way.