GoKawiilNature correspondence flags AI-driven gap between take-home and closed-book exam scores
A letter published in Nature responds to an earlier essay noting a class scored 96% on a take-home midterm but only 48.6% on a closed-book final. The authors argue this disparity isn't primarily about students cheating or inadequate proctoring, but reveals that many university assessments test tasks AI tools can now complete faster and better than typical students.
GoKawiil's interpretation of the reporting above, not reported fact.
The case highlights a structural challenge for higher education: as AI systems outperform students on conventional take-home tasks, universities face pressure to redesign assessments to measure genuine understanding rather than task completion. This debate will likely accelerate reforms in exam formats, academic integrity policies, and how institutions verify learning in an AI-saturated environment.
- A take-home exam averaged 96%, versus 48.6% on a closed-book version, in the same course.
- The authors frame this as a pedagogical issue, not just an enforcement or monitoring problem.
- The disparity suggests many current assessments measure tasks AI can already perform better than students.
Blue Book Exam Booklets (Roaring Spring) — With take-home tests losing credibility in the AI era, the humble in-class blue book is making a comeback. These wide-ruled exam booklets are a simple, no-tech way to run closed-book essays and problem sets where the thinking has to happen on the spot.
See Blue Book Exam Booklets (Roaring Spring) on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: nature.com — Wu, 2026-08-25
Published there as: “Assessing students in the AI era”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.