Video 4 - How does taking powers of A help us get a better approximation?

When the question for explanation of LU decomposition loop comes up then Rachael mentions that taking powers of A help us get a better approximation. From where does that inference come from? Was that covered in anything earlier?