I think my largest annoyance with LLM slop PRs is the authors complete lack of taste in good abstraction and looking at what they are doing in context of the existing codebase.
The LLM is going to go "parse, don't validate" and create a struct over a type and just completely ignore the code duplication. Abstractions for the type? No thank you!
I'll duplicate this important for-loop in two different places.