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.