From my own experience (having maintained legacy human codebases and now and then toying around with LLM coding) there's a quite fundamental difference between the types of slop, though.
Bad human code often has antipatterns, missing sanity checks, terrible data structures, bad naming, convoluted flows, etc. but you can usually deduce the intent behind the code. It's pretty obvious where it's bad and why. And if you can get to the mental model that guided the development, you can figure out how to rewrite it.
LLM code has all of those problems, but on top of that, there's no underlying mental model to deduce, much less a consistent one. This, in my experience, makes the slop go from horrible but possible to fix and refactor, to beyond repair and in need of not just a complete rewrite, but an entire new architecture from scratch.
Bad human code often has antipatterns, missing sanity checks, terrible data structures, bad naming, convoluted flows, etc. but you can usually deduce the intent behind the code. It's pretty obvious where it's bad and why. And if you can get to the mental model that guided the development, you can figure out how to rewrite it.
LLM code has all of those problems, but on top of that, there's no underlying mental model to deduce, much less a consistent one. This, in my experience, makes the slop go from horrible but possible to fix and refactor, to beyond repair and in need of not just a complete rewrite, but an entire new architecture from scratch.