FunSearch: Making new discoveries in mathematical sciences using Large Language Models

The paper “Enhancing human performance in combinatorial competitive programming” by Petar Veli?ki?, Alex Vitvi?, Emilien Dupont, et al. Has been published in the journal Nature and was done by a team with contributions from Bernardino Romera Paredes, Amin Barekatai, Pengming Wang, and Alhussein Fawzi. The work aims to find new solutions for classical problems in combinatorial competition programming that require complex and highly specific algorithms. They used FunSearch, a tool developed by Google’s AI division, which can search large collections of code models or functions, to help them identify better solutions. This research also has practical implications for artificial intelligence-driven software development, as it demonstrates how powerful tools like AlphaTeensor can be applied in complex problems, such as those related to combinatorial competitive programming.