Most students report using LLMs to get solutions more often than they believe is ideal for learning, and this gap widened after the lesson¶

Claim · All claims
Evidence · 1 study · causal r1 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? After the lesson, 75% of students reported using LLMs for getting solutions to problems more often than they believe ideal, up from 53% before the lesson. → Jake O'Brien 2026

Evidence¶

Jake O'Brien 2026¶

Jake O'Brien, Margaret Ellis, Alma Robinson, John Simonetti, and Naren Ramakrishnan. (2026). Skepticism vs. Convenience: Physics Students' Perceptions and Use of Large Language Models Before and After Instruction. https://arxiv.org/abs/2609.21037 doi:10.48550/arxiv.2609.21037

q2 · i? · causal · r1

Comparison of self-reported usage frequency against beliefs about ideal usage for learning, from pre- and post-survey multiple-choice items (Tables III and IV). The solution-generation gap grew from "75% of students, up from 53% in the pre-survey"; other purposes shifted minimally. Descriptive percentages only.

"After the lesson, a large portion of students report using LLMs for getting solutions to problems more often than they believe they should (75% of students, up from 53% in the pre-survey)."

Discussion¶

Related Claims¶

• First-year physics students overwhelmingly believe LLMs can solve introductory physics problems with high accuracy before instruction — related
• A reflective lesson on LLM functioning and educational impacts shifted physics students toward greater skepticism about LLM learning impacts — related
• Despite skepticism about learning benefits, physics students report using LLMs regularly, motivated primarily by convenience and deadline pressure — related
• Before instruction, physics students show skepticism about LLM learning impacts, agreeing that over-reliance risks dependence, false confidence, and reduced conceptual understanding — related
• AI assistance reduces persistence: persistence costs concentrate among learners who used AI for direct solutions, not hints — related