United StatesEducation2023 – ongoing

Harvard University

Course-specific GPT tutor for introductory physics

Harvard reported about double the learning gains with a custom physics tutor in a classroom experiment

WorkingCompany's own claimOfficial ↗

Problem

Students entered introductory physics with different levels of preparation. A shared classroom pace could leave some students struggling while others lost interest.

Approach

Gregory Kestin built a course-specific tutor using the GPT API and lesson prompts refined by instructors. In fall 2023, students alternated between an AI-supported lesson at home and instructor-led active learning. Researchers compared pretests and posttests across the 2 conditions.

Results

about doublelearning gains with AI tutoring versus in-class active learning, in Harvard's preliminary analysisCompany's own claim[1]

194students whose learning outcomes were analyzedCompany's own claim[1]

Takeaway

The tutor used instructor-designed lessons and feedback. The results concerned a short experiment; final publication was still pending when Harvard reported them.

Sources

Professor tailored AI tutor to physics course. Engagement doubled.
Harvard Gazette5 Sep 2024Company publication

Source summaries

[1]Harvard Gazette5 Sep 2024

Harvard described its instructors' physics-tutoring experiment with 194 students. Preliminary analysis found about double the learning gains under AI-supported instruction.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026