D. da Silva Fernandes, Robin Welsch, et al. · 2025 · Computers in Human Behavior Read the source In two experiments on logical reasoning items, ChatGPT raised scores by about 3 points out of 20 — and people overestimated their score by about 4 more. The most AI-literate were the most overconfident. Questions it bears on Offloading or atrophy Why we misjudge our speed with AI