Axiom Math, a burgeoning AI startup based in Palo Alto, California, is poised to revolutionize the landscape of mathematical research by democratizing access to a potent new AI tool. The company is offering Axplorer, a sophisticated AI-driven platform designed to aid mathematicians in the discovery of novel mathematical patterns, a critical step in potentially unraveling long-standing, complex problems that have eluded human intellect for decades. This initiative aligns with a broader national push, exemplified by the US Defense Advanced Research Projects Agency’s (DARPA) expMath program, which champions the development and adoption of AI tools within the mathematical community. Axiom Math positions itself as a key player in this ambitious endeavor, recognizing that breakthroughs in theoretical mathematics often have profound and far-reaching implications across the technological spectrum, from advancing artificial intelligence to fortifying cybersecurity.

Axplorer is not an entirely new creation but rather a significant evolution of an existing AI tool called PatternBoost. PatternBoost, originally co-developed by François Charton, now a research scientist at Axiom, during his tenure at Meta in 2024, demonstrated remarkable capabilities. It was instrumental in cracking the notoriously difficult mathematical puzzle known as the Turán four-cycles problem. However, PatternBoost’s computational demands were substantial, requiring access to supercomputing resources. Axiom Math’s innovation lies in its ability to condense this power into Axplorer, a tool that can now be installed and run on a standard Mac Pro, thereby making advanced pattern discovery accessible to a much wider audience of mathematicians, researchers, and students.

The core philosophy behind Axplorer, as articulated by Axiom Math founder and CEO Carina Hong, emphasizes the experimental and exploratory nature of mathematics. While recent successes have seen large language models (LLMs) like OpenAI’s GPT-5 assist in solving certain mathematical problems, particularly those left as puzzles by the prolific 20th-century mathematician Paul Erdős, Charton expresses a degree of skepticism regarding the true depth of these achievements. He argues that many of these problems were open simply due to a lack of dedicated investigation, making it easier to find solvable "gems." Charton’s ambition, and that of Axiom Math, is to tackle the truly formidable challenges – the "big problems" that have been the subject of intense scrutiny by eminent mathematicians for years. This ambition was partly realized last year when another of Axiom Math’s tools, AxiomProver, successfully found solutions to four such significant mathematical problems.

Charton distinguishes Axplorer’s approach from that of typical LLMs. He posits that LLMs excel at tasks that are derivative of existing knowledge, leveraging their vast pre-training on existing data. This makes them inherently "conservative" in their approach, tending to reuse established concepts. However, Charton highlights that true mathematical progress often hinges on novel ideas and unprecedented insights, frequently born from the identification of previously unrecognized patterns. These discoveries can, in turn, spawn entirely new fields of mathematical inquiry. PatternBoost was specifically engineered to facilitate this type of pattern discovery. The user provides an example, and the tool generates similar examples. The user then selects the most promising ones, feeding them back into the system for further refinement and generation. This iterative process mirrors the functionality of systems like Google DeepMind’s AlphaEvolve, which employs an LLM to generate and refine novel solutions to problems.

The primary hurdle preventing widespread adoption of powerful AI tools like PatternBoost and AlphaEvolve has been their prohibitive computational requirements. These tools typically necessitate access to extensive clusters of Graphics Processing Units (GPUs), a resource unavailable to the vast majority of individual mathematicians. Charton notes that while mathematicians are intrigued by AlphaEvolve, its proprietary nature and the need for direct access through DeepMind researchers limit its practical utility. Similarly, his own experience using PatternBoost to solve the Turán problem involved leveraging "thousands, sometimes tens of thousands" of machines for three weeks, a feat he describes as "embarrassing brute force."

Axplorer represents a paradigm shift in efficiency and accessibility. According to Axiom Math, it achieved the same result as PatternBoost on the Turán problem in a mere 2.5 hours, running on a single machine. This dramatic improvement in performance and reduced hardware dependency is central to Axiom Math’s mission to empower a broader segment of the mathematical community. Geordie Williamson, a mathematician at the University of Sydney and a collaborator on PatternBoost with Charton, expresses keen interest in Axplorer, although he has not yet had the opportunity to test it. Williamson acknowledges that Axiom Math has incorporated several enhancements into Axplorer that theoretically expand its applicability to a wider array of mathematical problems. He cautiously notes, "It remains to be seen how significant these improvements are."

Williamson observes a current "strange time" where mathematicians are inundated with AI tools vying for their attention. He suggests that the sheer volume of possibilities can be overwhelming, and the ultimate impact of yet another tool is yet to be determined. Carina Hong empathizes with this sentiment, recognizing that many AI tools require mathematicians to engage in complex tasks like training their own neural networks, a process that can be a deterrent. Axplorer, in contrast, is designed to offer a more guided, step-by-step experience, making it more approachable.

Demonstrating their commitment to open innovation, Axiom Math has made the source code for Axplorer open-source and readily available on GitHub. Hong expresses a strong hope that this accessibility will inspire students and researchers to utilize the tool for generating sample solutions and counterexamples, thereby accelerating the pace of mathematical discovery. While Williamson is a proponent of embracing new tools, including LLMs, he cautions against abandoning traditional methods. He affectionately refers to PatternBoost as a "lovely idea" but emphasizes that it is "certainly not a panacea," urging the community not to "forget more down-to-earth approaches." This sentiment underscores the ongoing dialogue within mathematics about the ideal integration of AI, balancing the pursuit of revolutionary computational power with the enduring value of foundational, human-driven mathematical exploration.