By the same logic, a hallucinating LLM is also overkill versus just doing the simple task yourself and not needlessly adding risk to it. The point still remain

By the same logic, a hallucinating LLM is also overkill versus just doing the simple task yourself and not needlessly adding risk to it. The point still remains: let's see what the LLM delivered that the user actually used. Either it's legally binding and an appropriate use, or it's not fit-for-purpose. Equally, why not an interactive form using conditional logic? No hallucination possible. Much more simple and reliable.