High-school students given GPT-4 access solved more problems in the moment but performed worse than controls once the tool was removed (review attribution)¶

Claim · All claims
Evidence · 1 study (2 entries) · design r1 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Students with GPT-4 access outperformed controls during use but underperformed them after tool removal, and randomised trials (N=1,222) add causal evidence plus reduced persistence. → Mahbub Ul Alam 2026 q2 i? Randomised controlled trials reported by the review (N=1,222) provide causal evidence for worse performance after AI assistance is removed, plus reduced persistence. → Mahbub Ul Alam 2026 (2)

Evidence¶

Mahbub Ul Alam 2026¶

Mahbub Ul Alam. (2026). PAUSE: A Privacy-Preserving Self-Reflection Tool for AI-Associated Cognitive Offloading. arXiv. https://arxiv.org/abs/2609.13155 doi:10.48550/arxiv.2609.13155

q2 · i? · design · r1

The review attributes to Bastani et al. (2025) the finding that students "given GPT-4 access solved more problems in the moment but performed worse than the control group once the tool was removed".

"Bastani et al. (2025) found that high-school students given GPT-4 access solved more problems in the moment but performed worse than the control group once the tool was removed."

Mahbub Ul Alam 2026 (2)¶

Mahbub Ul Alam. (2026). PAUSE: A Privacy-Preserving Self-Reflection Tool for AI-Associated Cognitive Offloading. arXiv. https://arxiv.org/abs/2609.13155

q2 · i? · design · r1

The review attributes to Liu et al. (2026) randomised controlled trials with N=1,222 providing "causal evidence for the same pattern" and reduced persistence. No effect size is printed.

"Liu et al. (2026) provide causal evidence for the same pattern in randomised controlled trials (N= 1,222 ), adding that AI assistance also reduced persistence."

Discussion¶

Related Claims¶

• How genAI is used, not who uses it, predicts whether independent performance declines or improves — the general claim this finding bears on, with the studies behind it
• Heavy reliance on generative AI can undermine learning: students with GPT-4 access performed better on practice problems but worse on exams once the tool was removed — possibly the same claim (merge candidate)
• AI-induced gains are not purely delegation: treated participants do not perform worse once AI is removed, and lower-education participants retain part of their gain, though a 0.200 SD gap re-emerges — reports the opposite
• Performance improvements from generative AI assistance diminished once the assistance was removed — related
• Students given AI assistance in peer feedback tended to rely on the AI-generated feedback and struggled to perform the task when the assistance was removed — related
• Actors internalized the AI's questioning style, self-generating similar questions after the tool was removed — related
• Access to a standard GPT-4 chat tutor during practice substantially improves practice performance in high school mathematics — a narrower finding that bears on this claim
• In a randomized experiment on fraction problems, participants who used an AI assistant solved fewer unassisted test problems and skipped more than control participants — related
• Students do not perceive the learning harm from copying GPT solutions and overestimate their exam performance — related