FunSearch: DeepMind unlocks math discoveries with AI

Click for: original source

FunSearch teams up creative AI that spits out code with a system that checks if the ideas are actually good. This back-and-forth helps it evolve initial thoughts into solid discoveries. By Alex Ivanovs.

In a leap for artificial intelligence, Google DeepMind researchers have unveiled a new AI-driven tool named FunSearch, which has made unprecedented discoveries in the realm of mathematical sciences. This tool marks the first instance where a Large Language Model (LLM) has contributed to solving complex open problems in mathematics and computer science.

FunSearch by Google DeepMind


At its core, FunSearch is a marriage of creative AI and precise evaluation. It starts with an LLM, like Google’s PaLM 2, which generates inventive solutions in code. These are then sifted through by an automated evaluator, ensuring only accurate and viable ideas progress. This back-and-forth process allows initial concepts to evolve into groundbreaking discoveries.

What sets FunSearch apart is its problem-solving ability and how it does so. Unlike black-box solutions, it produces programs that reveal the ‘how’ of the solutions, not just the ‘what’. This transparency allows scientists to understand and learn from the AI’s approach, sparking new insights and strategies. Good read!

[Read More]

Tags data-science google big-data programming