This seems like it's just guessing based on the length of the proposed password.

It also doesn't explain a method of how the "crack" gets done. Because tables of passwords and hashes were part of its training, the "AI" can recognize the hash and give back a password? Or something more esoteric?