I wonder if we can use the compression ratio that an LLM-driven compressor could generate to figure out how much entropy is actually in the system and how much is just boilerplate.

Of course then someone is just going to pregenerate a random number lookup table and get a few gigs of 'value' from pure garbage...