Chinese AI solves decade-old maths problem in hours, with no human intervention
www.scmp.com
Research team led by Peking University says dual-agent approach bridges the gap between reasoning and formal machine verification.

cross-posted from: https://news.abolish.capital/post/41854

A Chinese artificial intelligence framework has autonomously resolved an open problem proposed more than a decade ago by a US mathematician, according to the Peking University-led team that developed it. The dual-agent framework solved the problem posed in 2014 by former University of Iowa professor Dan Anderson – who died in 2022 at the age of 73 – the researchers said in a preprint paper published on April 4. By synthesising decades of mathematical literature, the Chinese team’s AI framework…

From China - South China Morning Post via This RSS Feed.

  • JustSo [she/her, any]@hexbear.netEnglish
    10·
    17 days ago

    EDIT: NOPE. I’m wrong and making assumptions has lead to a foolishly overconfident claim.

    It is not unusual to publish before peer review, I have been informed. I do apologise OP.

    “This work provides a concrete example of how mathematical research can be substantially automated using AI,” according to the paper, which has not yet been peer reviewed.

    To be fair they put this after a tall media embed. At least on the archive snapshot I read. Paywalled on the OP link.

    I think it is a lot more misleading to publish a claim of a solution to a mathematics problem without peer review. Maths problems are usually straight forward to verify, and until its been done you can’t just be like “trust me I got this.”

    Guy says his computer solved the equation. check-please

    Archive anyway: https://archive.ph/efmcI

    • QuillcrestFalconer [he/him]@hexbear.netEnglish
      8·
      17 days ago

      According to the article they used formal verification, that’s an important part of automated math research (and increasingly human research too).

      This is not new btw, around the new year there was a bunch of new automated / semi-automated proofs using LLMs + formal verification with Lean of some Erdos problems

  • technocrit@lemmy.dbzer0.comEnglish
    1·
    7 days ago

    Bullshit. There is zero “AI” involved. Mathematicians have been using computational methods since mathematicians invented the computer. There has never been any kind of “artificial intelligence” involved and probably never will. It’s just statistics/math plus compute.

    Secondly a “decade old” problem in math is super new and nobody gives a fuck.

    Thirdly it’s telling AF they don’t mention the actual problem or link to the actual paper. It’s most likely something that’s completely manageable by big, modern computers.

    Please stop feeding the bullshit “AI” dumpster fire.