The LLM isn’t contributing garbage, the user is by (likely) not testing/verifying it meets all requirements. I haven’t yet used an LLM which didn’t require some handholding to get to a good code contribution on projects with any complexity.