You yourself identified some problems that a modern AI guru would salivate over: Formal specifications from natural language, "filling in" formal details. etc
I'll add another one: NN-based speedups to constraint solvers, which are usually something like branch-and-bound. This is (i've heard) a very active area of research.
I'll add another one: NN-based speedups to constraint solvers, which are usually something like branch-and-bound. This is (i've heard) a very active area of research.