Any moonshot ideas for large language model applications?

LLM applications are obviously booming now and there are many companies and startups working on realistic applications that can become true in the following 3 to 5 years. How about bold ideas that may take up to 10 years to come true?