OpenAI pulls math papers but traders see a Millennium solution as unlikely this year

OpenAI withdrew three manuscripts from its new collection of mathematical work after finding a sign error, and revised 14 others, Retraction Watch reported. The correction adds a concrete example to the scrutiny of the release, but it was not an announcement of a solution to a Millennium Prize problem. Traders see another qualifying OpenAI announcement this year as very unlikely. Expectations fell sharply in the latest move, though the reason for that decline is not established by the withdrawal alone.

The drop came after a burst of optimism around OpenAI’s October 6 release of hundreds of mathematical results. Expectations had eased from near even at the start of October and then held steady for several days. They rose on October 6 before falling back the next day. The Guardian and Nature reported mathematicians’ concerns about the results’ vetting and whether researchers could check work produced with models unavailable to them. Those reports plausibly contributed to the reversal, but The Guardian’s story appeared after the decline had begun. The correction reported by Retraction Watch is further context, not proof of what drove the move.

OpenAI’s September announcement about a proposed Navier-Stokes solution had already raised expectations of further advances, but that problem is excluded from this question. The October collection did not announce a complete solution to any of the five eligible problems. OpenAI’s repository describes work on the Riemann Hypothesis as proving zero-free regions, and its Birch–Swinnerton-Dyer results as applying to restricted cases. Neither amounts to solving the full problem.

A longer-term question about OpenAI announcing another Millennium Prize solution before 2028 is now closer to a toss-up, though it had looked likely when this story began. OpenAI says it plans to consult mathematicians and improve how it presents future work, but has given no timetable for another release. A complete solution announcement from OpenAI or an official representative would change the picture; partial results and further revisions would not qualify.