Axiom Math, a startup located in Palo Alto, California, has launched a groundbreaking AI tool named Axplorer, aimed at transforming the way mathematicians approach problem-solving. This innovative tool is designed to identify mathematical patterns that may lead to solutions for long-standing mathematical challenges. Axplorer is a refined version of its predecessor, PatternBoost, which was co-developed by François Charton during his tenure at Meta. While PatternBoost operated on a supercomputer, Axplorer is optimized to run on a standard Mac Pro, making it accessible to a wider audience of mathematicians who can install it on their personal devices.
The introduction of Axplorer aligns with the U.S. Defense Advanced Research Projects Agency’s (DARPA) initiative known as expMath, which promotes the integration of AI tools in mathematical research. Charton, now a research scientist at Axiom, emphasizes the significant implications that breakthroughs in mathematics can have across various technological fields, particularly in enhancing artificial intelligence and internet security. Unlike many existing AI solutions that primarily focus on resolving pre-defined problems, Axiom Math’s approach encourages exploration and experimentation in mathematics. As Carina Hong, Axiom Math’s founder and CEO, points out, the essence of mathematics lies in its exploratory nature, which Axplorer seeks to enhance.
In recent months, some mathematicians have turned to large language models (LLMs) like OpenAI’s GPT to tackle unsolved problems, particularly those posed by the late Paul Erdős. However, Charton remains skeptical about the efficacy of these models for tackling more complex and well-studied mathematical issues. He asserts that while LLMs excel at deriving solutions from existing knowledge, many mathematical challenges require original insights and innovative thinking that can arise from identifying new patterns. Axplorer is designed to facilitate this process by allowing mathematicians to input examples and generate related patterns iteratively. The tool is not only faster but also significantly more efficient, achieving results in mere hours compared to the weeks it took with PatternBoost. With its open-source code available on GitHub, Axiom Math hopes that Axplorer will empower both students and researchers to accelerate mathematical discovery by generating solutions and counterexamples to their ongoing problems.
Source: This startup wants to change how mathematicians do math via MIT Technology Review