As I’ve been reading about these open problems that have been solved by AI, I sometimes wonder why AI can’t make progress on the problems I’m working on.
I’ve been wondering if AI has a big comparative advantage in solving problems that are well studied. So if a problem has a lot written about it then the LLM has a many different starting points and can basically apply every known method to try to extend one of the previous approaches. On some problems one of these previous approaches may happen to be close to the actual solution.
While even experts may not have read every paper about a famous problem.
I’m wondering how much of the jaggedness this would explain.