News for hackers. Stuff that matters.
Re:Towards Autonomous Mathematics Research by umairnadeem123 on Sunday February 15, 2026 @09:48PM (#47030324)
interesting that they call out success cases as rare, thats honestly the most useful part. are people here seeing better hit rate from tighter decomposition + verifier loops, or mostly just more compute?

also curious where failures cluster most: search, formalization, or proof checking?

Re:Towards Autonomous Mathematics Research by The_Gray on Monday February 16, 2026 @08:41PM (#47042652)
I think Terence Tao gave them feedback on initial drafts to emphasize this piece. Yes 4-13 ish meaningful Erdos problems (depending on what you count), but that's out of 700 problems run through the pipeline.