Google DeepMind's FunSearch AI Outperforms Mathematicians in Solving Complex Problems
Google DeepMind unveils FunSearch, an AI system that leverages large language models to discover new mathematical solutions by writing and evolving computer code. The tool reportedly outperforms human mathematicians on previously unsolved problems.
Google DeepMind introduces FunSearch, an innovative AI system that uses large language models to solve unresolved problems in mathematical sciences. The name FunSearch stands for "searching in the function space" and operates by creating solutions in the form of computer programs rather than simply generating text.
The system pairs a large language model with an automated evaluator that ranks the generated programs based on their performance. This built-in fact-checking mechanism allows FunSearch to continuously evolve weaker programs into highly powerful solutions that discover genuinely new mathematical knowledge.
Google DeepMind CEO Demis Hassabis praises the breakthrough as an exciting milestone published in the journal Nature. FunSearch demonstrates that AI chatbots and language models possess the potential to outperform human mathematicians when tasked with complex, unsolved scientific challenges.