In a groundbreaking achievement, artificial intelligence (AI) has successfully solved a highly intricate mathematical problem that surpassed the capabilities of human mathematicians. Google DeepMind researchers announced this historic breakthrough, marking a significant step toward a potential revolution. The results have been published in the journal Nature.
The mathematical problem in question is the "cap set problem," which entails determining the maximum number of points that can be drawn without forming a straight line under specific conditions. The AI tool responsible for this accomplishment, named FunSearch, is built upon the AlphaZero AI framework, which approaches mathematical problem-solving akin to playing a game. This AI incorporates a large language model (LLM) called Codey, trained on computer code to differentiate between correct and incorrect answers.
The remarkable aspect of this achievement lies in the fact that the AI not only found a solution beyond its training data but also one that was previously unknown to exist. The researchers, including Pushmeet Kohli, Vice President of Research at DeepMind, suggest that the AI may have reached a level where it can "think for itself."
While the resulting solution to the cap set problem requires extensive validation through testing, the researchers acknowledge the significance of AI creating something entirely new, even when it took several days to find the solution. DeepMind research scientist Alhussein Fawzi admitted that they had hypotheses but were uncertain about the project's success initially.
This breakthrough represents a unique instance of a large language model (LLM) contributing to a genuine scientific discovery, marking a notable advancement in AI capabilities. The validation of AI-generated data will play a crucial role, given previous instances where LLMs produced incorrect or nonsensical outputs, even with correct answers in the training data.
