Not an expert, but I think procedurally generated terrain is generally fractal in nature, and is reproducible in that sense from a seed that is used in the generation. It is therefore recursive, as fractals are recursive.

A traditional neural network is a universal function approximator, however it is not recursive in nature, unless it is some sort of RNN. The transformer architecture, which this seems fairly similar to this one, is also not recursive in nature; although I believe, limited recursion can come about through CoT.

Therefore, I don't believe this could match the reproducibility, in an infinite sense, of a traditional procedural generator.

"Fractals are not self-similar"

https://m.youtube.com/watch?v=gB9n2gHsHN4

If you want to learn more about the fascinating world of fractals.

Luanti [1] (Minetest) has a fractal map generator ("mapgen"). You can test it for yourself. It's funny at first but becomes eventually boring.

Its other mapgens massively some kind of Perlin noise in various ways, so that you can have "realistic" landscapes (e.g. the Carpethian mapgen) or landscapes with impossible mountains and floating rocks sometimes (e.g. the V7 mapgen) that are good for fantasy/sci-fi worlds.

Noise is a pretty efficient way to fake complexity, and it's not a coincidence [2].

[1] https://www.luanti.org/ [2] https://en.wikipedia.org/wiki/Kolmogorov_complexity

Procedurally generated terrain is whatever it wants to be. It isn't necessary for it to be fractal in any particular sense. Or even arguably all that useful, at least at the present time.