This startup wants to change how mathematicians do math
View original at technologyreview.comMIT Technology Review - Ai Research Title: This startup wants to change how mathematicians do math Date: 2026-03-25 13:59 Source: https://www.technologyreview.com/2026/03/25/1134642/this-startup-wants-to-change-how-mathematicians-do-math/ <p>Axiom Math, a startup based in Palo Alto, California, has released a <a href="…
What we drew from this source
The claims Via News extracted from this document. We point to the source; we don't replace it.
Mathematicians are excited about AlphaEvolve but it's closed and you need special access to use it
60% confidenceMathematicians are somewhat overwhelmed by the possibilities of AI tools and it is unclear what impact having another such tool will be
60% confidencePatternBoost is a lovely idea but it is certainly not a panacea, and mathematicians should not forget more down-to-earth approaches
60% confidenceThere are tons of problems that are open because nobody looked at them, and it's easy to find a few gems you can solve
60% confidenceBreakthroughs in math have enormous knock-on effects across technology, with new math being crucial for advances in computer science, from building next-generation AI to improving internet security
60% confidenceSome AI tools require mathematicians to train their own neural networks, which is a turnoff, but Axplorer walks users through what they want to do step by step
60% confidenceThe Turán four-cycles problem that PatternBoost cracked is a big problem in mathematics
60% confidenceWhen solving the Turán problem with PatternBoost at Meta, he had access to thousands or tens of thousands of machines and it ran for three weeks using embarrassing brute force
60% confidenceAxplorer took just 2.5 hours to match PatternBoost's Turán result and runs on a single machine
60% confidenceThere are lots of problems in math that require new ideas and insights that nobody has ever had, sometimes coming from spotting patterns that hadn't been spotted before
60% confidenceLLMs are extremely good if what you want to do is derivative of something that has already been done, but they are conservative and try to reuse things that exist
60% confidenceMath is exploratory and experimental, not just about finding solutions to existing problems
60% confidenceAxiom Math has made several improvements to PatternBoost that in theory make Axplorer applicable to a wider range of mathematical problems, but it remains to be seen how significant these improvements are
60% confidenceShe hopes that students and researchers will use Axplorer to generate sample solutions and counterexamples to problems they're working on, speeding up mathematical discovery
60% confidence