AI labs running out of hard math problems, progress may stall

tak3sh8 · x · 2026-09-30

tak3sh8 argues AI math capability is already superhuman in some aspects but faces diminishing returns: labs lack incentives to push further since solving Navier-Stokes-type problems doesn't transfer well. He adds that labs admit they're running out of hard math training problems with no clear fix, so progress will likely stall soon.

Original post →

More from AGI Musings

AGI Musings channel →