Forgetting the villainization of overfitting, what happens when you just take a massive amount of training data and overfit the model on that. Is it possible to achieve meaningful compression with "perfect" accuracy?
