why did they destroy stockfish in finding mates

[Dann Corbit]

I don't think a NN adds anything here.
A pure move generator is all that is needed.
The move generator will know if a position is a win, loss, draw, or something else (which needs expansion).
The and/or search already knows how to assign probabilities to the nodes.
On the other hand, maybe there are special patterns that an NN might recognize.
But I suspect that a box full of GPUs would be solving mates in 100 with or without a neural net.