Systematic checking like that is pretty much LLM's worst case. They're really bad at it. Definitely better to use them to suggest types.