AI for Maths

How good are state-of-the art LLMs such as Gemini Deep Think at logic? That is a central question for the development of AI. How good are they at finding new connections between a priori distinct mathematical structures? It might take your breath away. I will discuss some down to earth examples related to current maths research where Deep Think delivered an impressive performance. I will also discuss current limitations. Michel van Garrel is an Assistant Professor in Mathematics and Statistics at the University of Birmingham, where he is a member of the group of Geometry and Mathematical Physics. Prior to Birmingham, he held a Marie Curie Fellowship at the University of Warwick, as well as postdoctoral positions at the University of Hamburg, the Korea Institute for Advanced Study, and the Fields Institute of the University of Toronto. He obtained his PhD from the California Institute of Technology in 2013. Michel started his research career in the field of Enumerative Geometry, which draws inspiration from String Theory. In recent years, his research has expanded to include works in Algebraic Geometry and Mirror Symmetry. Michel was part of a select group of mathematicians who were given early access to the Google Gemini Deep Think model that delivered a gold medal performance at the 2025 International Mathematical Olympiad. Read more: https://vangarrel.com/ Get a copy of the slides: https://www.slideshare.net/slideshow/gdg-ai-for-science-tech-talk-using-ai-for-mathematics-research/283829283 Join the GDG AI for Science community for talks, events, workshops, collaborations and more: https://gdg.community.dev/gdg-ai-for-science-australia/