Theoretically, large language models have enough common sense to understand all variations of natural language commands, and to ask for clarification if they think some request is ambiguous. It's probably not yet feasible to do Siri via an LLM, or not via a properly large one (that has the necessary intelligence).