UF team expands AI tool designed to help students learn from math proofs

A University of Florida research team is expanding an AI tool that gives students feedback on mathematical proofs without supplying complete solutions. The project is designed to help students work through challenging problems themselves, preserving the productive struggle educators say is central to learning.

The team, led by UF mathematics professor Vincent Vatter, will use an $814,000 grant to extend HaLLMos beyond introductory proof courses to advanced undergraduate linear algebra, real analysis and topology. The effort addresses a growing education question: how to use AI for support without letting it do the thinking for students.

How HaLLMos supports students

HaLLMos reviews students’ written proofs and points out weaknesses in their reasoning, so they can revise their work. Vatter has used it in his teaching, where students can submit a proof, receive feedback and try again before turning in an assignment. The tool is freely available to students and instructors.

Nearly 40,000 student proofs have been submitted to the system. Vatter said students need substantial feedback when learning to write proofs, but instructors often lack the capacity to provide it individually. Unlike AI tools that can produce full solutions, HaLLMos is intended to help students identify and correct problems in their own work.

What the grant will fund

The grant from the AI for Math Fund, supported by Renaissance Philanthropy and XTX Markets, will fund a fellowship program involving about 40 mathematicians, including faculty members, postdoctoral researchers and graduate students from institutions across the country. Fellows will assemble examples pairing proofs with expert feedback to help train the AI model.

The team says feedback needs to reflect the subject: advice that helps with linear algebra may not work for real analysis or topology. Co-investigators are Sarah Sword of the Education Development Center, Jay Pantone of Marquette University and Ryota Matsuura of St. Olaf College. The project builds on earlier work supported by the National Science Foundation.

This report draws on information from news.ufl.edu.