Caltech mathematicians build open-source AI to own math solutions
Caltech researchers are building open-source, math-specific AI after OpenAI's October solution dump, testing whether academics can retain control over solutions, data and model development.
By Amara Osei
3 min read

What's News
- OpenAI solved hundreds of thorny math questions in October, according to Business Insider.
- Gukov said his Caltech lab's AI models solved a few of those questions first.
- Most of Gukov's models solve only one specific math problem; a generalized model is not ready.
- Terence Tao demonstrated the Open Math OS platform at Caltech on Friday.
- Two mathematicians accused OpenAI of using their chat logs to help solve the Navier-Stokes problem.
By Amara Osei
OpenAI solved hundreds of thorny math questions in October, according to Business Insider, after the problems had stumped the field's best researchers for years. Sergei Gukov was not surprised: his Caltech lab's AI models had solved a few of them first, he said.
Why are mathematicians building their own AI?
Gukov, a Caltech professor, runs a lab where he and a handful of students design and build open-source AI models from scratch. He said the systems are engineered specifically for mathematical problems.
Gukov said: “The more we work with the systems, the more we understand them. As a result, the better we can control them.”
Business Insider reported widespread antipathy toward major AI companies as those companies rapidly published solutions without consulting academic luminaries. Mathematicians had criticized that approach in Substack essays and lecture halls across the country.
OpenAI was the main target, according to Business Insider. The outlet reported that mathematicians accused major AI companies of brazenly exploiting their field to build hype for an IPO.
Business Insider reported that many mathematicians called OpenAI's solutions “slop.” They said the proposals were meaningless unless a human researcher spent weeks deciphering how they were reached.
Business Insider described OpenAI as dominating the current balance of power in mathematics. Mathematician-built systems could offset that imbalance by putting ownership of AI-generated solutions back into the field's hands.
Tasmin Chu, a Caltech Ph.D. math student who runs a popular anti-AI Substack, called mathematician-created AI systems an effort by mathematicians to “own the means of production.” Chu added: “It's complex. I guess it depends on how it really plays out.”
How far do the specialist models reach?
Gukov likens his models to Ferraris: luxury cars that are fast, carefully made and largely inaccessible to the general public. He said most models solve only one specific math problem and sputter when asked to branch into another area.
By comparison, Anthropic models can handle math prompts across multiple disciplines, Gukov said. A more generalized model from Gukov's lab is in development but is not ready yet.
Business Insider described a field divided by AI. One group opposes its use in mathematics altogether, and Chu leans toward that camp.
“I'm personally invested in it remaining viable to do mathematics without AI at all,” Chu said. Business Insider reported that students on the other side were going all in on AI, either to save the field or because they saw little worth saving.
Terence Tao, a mathematician at UCLA, has repeatedly advocated that mathematicians create their own AI systems. On Friday at Caltech, he demonstrated a platform created by a nonprofit he founded.
The nonprofit calls the platform Open Math OS. It provides access to open-source frontier AI models, cloud computing and collaborative tools.
Business Insider reported that the platform lets mathematicians control how their data is used. That control has become material since two mathematicians accused OpenAI of using their chat logs to help solve the Navier-Stokes problem.
Gukov was born in Moscow and immigrated to the United States to study mathematics, according to Business Insider. He treats AI as an opportunity for growth rather than an existential crisis in the field.
“This AI research aspect is super exciting,” Gukov said. “You feel like Columbus exploring new territory.”
The next test for mathematician-built AI is whether Gukov's lab can broaden its models beyond one problem while retaining the control he values. Anthropic still reaches across more mathematical disciplines, and Gukov says his generalized model is not yet ready to close that gap.
Source: feeds.businessinsider.com
More from Amara Osei
Show full bio
Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.
2,358 articles