Welcome to the AI crisis in math
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI published "10 Advances in Mathematics and Theoretical Computer Science" alongside several hundred pages of documentation, claiming internal model Astra solved longstanding problems in disciplines including quantum game theory and sphere packing in higher dimensions.
- OpenAI initially framed the 10 problems as having had "no progress in the last 10 years," but the underlying papers acknowledge building on prior work — reigniting familiar complaints about labs failing to credit the human research underneath.
- OpenAI's internal model (almost certainly Astra) disproved the 80-year-old unit distance conjecture in May, the first proof point in what The Verge's Robert Hart calls an explosive capability shift over the past six months.
- Researchers told Hart that if a human had solved all 10 problems simultaneously, "we probably wouldn't believe it," though they were "quite impressed" because the problems are ones mathematicians genuinely care about — not fringe areas.
- Despite the abstract-math gains, AI models remain "truly terrible" at elementary tasks — counting R's in "strawberry," telling time, or naming the correct day of the week — exposing a sharp capability gap inside a single discipline.
- Mathematicians flagged topology as an abstract area where AI still performs poorly, and Hart reports the community is now debating whether the surge in attention to math is a genuine research shift or "just a big marketing exercise" for frontier AI labs.
Why it matters: OpenAI staked frontier-reasoning credibility on a domain where AI still can't count or tell time. If those abstract-math skills fail to transfer to other fields, the bombshell announcement stays a marketing win rather than a generalizable breakthrough. For academic math itself, the deeper question is whether research grants and PhD pipelines survive when frontier models start answering the field's open problems.
Ask SkimNews
