> My understanding of neural networks is that there are no remnants of the original inside it.

Can you prove that? I can prove the opposite.