How can we build AI systems that can think like a Math Olympiad Expert?
7 min read
Just now
--
Press enter or click to view image in full size
Introduction
Mathematical Olympiad problems are designed to humble the brightest of minds all around the world. International Mathematical Olympiad (IMO) tests a kind of mathematical intuition that takes years, or even sometimes decades, to develop. AI mathematical reasoning is closing that gap, and no one is exactly sure how far its limits are.
Yet, I am deeply humbled to have the chance to work on a problem at this frontier, as a student in the Data Science field. I attempt this not as a researcher or a mathematician, but as someone genuinely curious about how far AI can go.
My team and I were trying to build an AI system that could solve these problems. My teammate, Sohaib, is a mathematics lecturer. He is someone who tackles problem-solving, just like these competitions require. He has also been a mathematics lecturer at an overseas university for over 9 years.
Whereas I come from the applied side: systems, pipelines, and language models. This combination is complementary. The problems are not purely mathematical, and it was not purely technical either. It required…