No, this would be precisely the sort of optimization that they are seeking. But... It's hard. To do this over large numbers of high-entropy bits with a small

No, this would be precisely the sort of optimization that they are seeking. But... It's hard. To do this over large numbers of high-entropy bits with a small seed (varying) and some simple algorithm (varying) rapidly requires a search space of a scale implying more computation than might plausibly be made to exist in our observable universe. There has to be a minimum-message-length compression solution for any arbitrary message and decoder environment (whether their code or the uncompressed text), but good luck provably finding it for messages even as long as this post. "You would have to be a god" is a reasonable response if it's talking about using computation capability outside of our observable universe and the physical rules we understand it by.