@troed @chainq One recurring pattern I see with LLM contributions is that code-level abstractions don't make sense. The feature works, all tests pass, but a considerable amount of functionality is at the wrong place. Simplest example is a C project where *technically* it doesn't matter in which file you implement a function, but I've seen the same with Java&classes too. This doesn't matter for machines because they just grep for the function name, but makes human contributions extremely taxing, like living in a mad mans house where the soap is in the fridge and you have to climb a ladder to the attic to find the salt.
I think this can be a valid argument against the LLM push.