> something that few years ago was decades away.

There's no way of knowing this - I see articles fairly often on HN of mathematicians (sometimes grad students or younger) solving problems where progress previously had stalled.

What I meant was not that these problems wouldn't get solved for decades, but that few years ago (before advent of LLMs) if you've asked average researcher how far away are we from AI solving unsolved math problems, the median answer would be that we are far, far away from that.