> I'm not sure why there seems to be a consistent set of design mistakes that LLMs tend toward.

Probably something to do with the training data and how it’s weighted. It’s the same how the writing is essentially full of overused phrases, the AI slop feel.