Since you last thought about math, the people who think about math every day have been freaking out.
A month ago, when OpenAI announced that one of its powerful internal models had cracked a notoriously difficult Millennium Prize Problem, it looked like the most spectacular demonstration yet of AI’s progress in mathematics.
Then mathematicians revolted.
The backlash from the math world coincided with a global reckoning about technology that Silicon Valley labs are racing to build. On the same day that OpenAI published its landmark proof, a researcher quit Anthropic over safety concerns. Within hours, the AI breakthrough was overshadowed by fears that AI could destroy humanity.
Meanwhile, mathematicians were dealing with existential questions of their own.
But as they publicly debated the future of their ancient field, OpenAI’s model kept churning out proofs behind closed doors.
This left the company with a problem that a bunch of humans would have to solve: what to do with all that AI-generated math.
On Tuesday night, OpenAI dumped those proofs online for mathematicians to wrap their minds around. The deluge included more than 700 papers covering hundreds of problems across number theory, algebraic geometry and other fields with names that could give people haunted by high-school math a migraine. Among the major findings reported by OpenAI were advances related to three of the five remaining Millennium Prize Problems, including the Riemann hypothesis.
“If a human did this, it would be an instant Fields Medal, no questions asked,” Alex Kontorovich, a Rutgers University distinguished math professor, wrote on X about the Riemann result.
As stunned math experts began to absorb the results, even a researcher inside Anthropic called OpenAI’s release “obviously the most significant moment in mathematical history.”
It was also the latest head-spinning twist in a historic year for AI and math itself.
In recent months, both OpenAI and Anthropic have turned complex mathematics into an unusual form of marketing for their smartest models, trumpeting solutions to problems that had stumped humans for decades.
But what once seemed like a safe way to flaunt AI’s capabilities started to backfire as Americans have become increasingly skeptical of AI.
The labs have also lost the trust of some people who have devoted their lives to math and once embraced AI tools as outrageously powerful calculators but now feel like collateral damage in a battle of trillion-dollar companies.
Their rivalry is so fierce that OpenAI says it unleashed its frontier model on the Millennium Prize Problems—and poured millions of dollars in computing power into the effort—only after hearing that Anthropic had already solved at least one. After OpenAI published its solution to the Navier-Stokes problem, an academic working with an Anthropic researcher suggested that the model may have drawn on his unpublished work, which the company denies.
The criticism intensified when more than two dozen winners of the prestigious Fields Medal published a furious open letter called “A Severe Misalignment of AI in Mathematics,” warning that the industry’s rush to solve math problems threatened to undermine the very purpose of math.
Under pressure from the scientific community that OpenAI says it’s trying to empower, the company recently helped assemble an independent group of leading mathematicians for advice.
“There are difficult questions ahead about how AI can support mathematical understanding,” OpenAI said last month. “We want mathematicians to be at the center of shaping the answers.”
Last week, that advisory committee laid out guidelines for how AI labs should publish their discoveries based on feedback from hundreds of mathematicians.
OpenAI says it drew on those recommendations as it released this avalanche of math.
The release included 722 papers organized into 372 groups of mathematical results, and the company said it had fed the model about 4,000 problems in total. It also provided summaries of the model’s “chain of thought” for 10 results and estimated that the average result required only a fraction of the computing power that went into the Millennium Prize triumph.
Earlier this week, as rumors swirled that OpenAI was preparing to drop an unprecedented barrage of proofs, mathematicians scrambled to publish their own AI-assisted work before they could be scooped.
In releasing the results, OpenAI says it took the math council’s advice and concerns to heart—with one exception.
“We ask them to stop testing advanced mathematical problems on proprietary models,” the top mathematicians wrote last week.
OpenAI responded: “It is important to continue to evaluate our internal frontier models on mathematics and other sciences, so we can accelerate developing the tools to advance those fields.”
