Newsletter drafts

Can better project performance conceal weaker learning?

Yes—in a specific educational experiment, it did. A peer-reviewed 2025 study tested GPT-4-based mathematics support in a Turkish high school. Students using the unrestricted interface performed better during assisted practice but scored 17% lower than the control group when AI access was removed. A tutor designed with teacher-informed guardrails largely eliminated that negative effect, without establishing a positive independent-exam effect. The intervention occurred in 2023; it was not a test of today’s private coding mentors. (pmc.ncbi.nlm.nih.gov)

Buying implication: Ask for both an assisted project milestone and a modest independent task. Do not turn this finding into a claim that all AI use harms learning—or that one branded tutoring mode guarantees understanding.

Sources cited on this page

  1. Generative AI without guardrails can harm learning: Evidence from high school mathematics - PMC ↗pmc.ncbi.nlm.nih.gov