1, 3, and 4 are likely to be resolved IMO. Others will get access to these models, they will get dramatically more efficient quickly, and the AIs will likely pass a point where they don't really need that much human collaboration to solve these problems.

2 is the concern. Even in a world where AIs become all-powerful oracles for math, we still need human mathematicians to ensure we as humans can understand the results, and direct progress toward areas that benefit humans. But this is a new role, very different from the current role of mathematicians. Current academics don't want this role and the upcoming generation of students probably don't either.