While I agree it'd let the user use the system, the system should do the right thing for either situation, or at least abort and say it doesn't understand. That's the problem with LLMs so far. They can't admit they don't understand.