New AI Math Tutor Checks Your Proofs and Catches Mistakes Instantly

Ask an AI chatbot to help with a math proof and it will happily oblige, sounding confident the whole time. The trouble is that it can invent a step that looks right but isn't — the same "hallucination" that makes people nervous about AI everywhere. Mathematicians already have software, called Lean, that checks proofs line by line with zero tolerance for hand-waving. But Lean is brutally picky and takes months to learn, so most students never touch it.

LeanSide, built by a team of university researchers, sits in the middle. Students type their proof the way they would write it on an exam. The system automatically translates that into Lean's strict language, has it verified, then translates the computer's verdict back into plain feedback: what holds up, what doesn't, and why. Nothing gets a pass just because it sounds reasonable.

The team tested it in real undergraduate math classes. They watched which features helped students push forward and which left them stuck — for instance, when the verifier demanded detail a student considered obvious, or when the translation from messy human phrasing into rigid code went wrong. Those findings turned into concrete advice for anyone building AI tools where accuracy is non-negotiable.

Why should non-mathematicians care? Because the recipe — let AI handle the friendly conversation, but put a strict, independent checker behind it — is exactly what's missing in medicine, law, accounting and finance, where a confident wrong answer costs real money. LeanSide is a small, controlled test showing the pattern works, and that people will tolerate a pickier tool if it earns their trust.