Post

Sometimes this means manually creating a lot of material that is going to be recombined, sometimes this means spending an unreasonable amount of time defining the exact parameters and structure of a grammar or a generator.

It's a magic trick of the same kind as the ones where something large comes out of a small box; it's done by making something that *unfolds* itself. "Generation" is almost a misnomer, really, it's *decompression.*

Any aspect of the system that isn't downstream from those choices is, essentially, noise. A little bit of noise is texture, a lot of noise makes the semantic content of the piece collapse out.

I think it's conceptually kinda broken because "meaningful variation" is a vague concept and as we've learned to the general audience meaning exists only when framed extremely well. Variation only matters insofar it's discrete & obvious, which ends up folding back into "make more content"

That's a concept I've really wanted to dig into for a long time. Definitionally, by doing procgen you've established an upper bound on the Kolmogorov Complexity of whatever it is you want to generate.

The piece that I've been unable to reconcile is the difference between Information density and what humans perceive as interesting. White noise is only slightly more interesting than a blank screen to a human, so where exactly does human interest peak in terms of information content.

I think it might be a category error to measure them on the same axis. Just as Shannon showed that the meaning wasn't really relevant to the mathematics of information, perhaps information isn't really relevant to meaning. Or perhaps the meaning that we bring to some amount of information moots it.