FunSearch is a method that pairs large language models (LLMs) with evaluators to search for new solutions in mathematics and computer science. By iterating between the LLM and evaluator, initial solutions evolve into new knowledge. FunSearch has successfully discovered new solutions for challenging open problems in science and mathematics, such as the cap set problem and the bin-packing problem. Unlike traditional computer search techniques, FunSearch generates programs that describe how the solutions were arrived at, making them more interpretable and actionable for researchers. The approach shows promise for addressing difficult combinatorial problems and practical challenges in computing.
https://deepmind.google/discover/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models/