I bet >$50m of inference compute in total has already been spent on open math problems, across the big labs, researchers, hobbyists, and startups
Incredible results, but I can't help but think that some kind of coordination layer could drastically decrease the cost per finding