Training novices to think, or giving them LLMs? Evidence from an RCT
· The paper specifies August 2026 without a publication day; 15 August is used only for sorting.
Summary
A randomized experiment with 1,053 first-year university students compares causal-reasoning instruction and access to ChatGPT on a business-advice task. ChatGPT improves rated output, while reasoning instruction changes how students develop and test explanations. Better assisted answers do not by themselves demonstrate that students have acquired lasting independent skills.
Why it matters
The experiment separates a productivity question—whether AI helps produce better work—from a human-capital question—whether people learn to do better work themselves. That distinction matters for training and entry-level jobs.
What to keep in mind
The study uses 13 classroom clusters at one university and a short task with ChatGPT Edu running GPT-4o. It does not establish retained skills or effects on workplace productivity. Some authors are affiliated with OpenAI; results need not generalize to other students, tasks or models.