The problem will always be that you have to use an llm to ask questions in natural language. Which means it gets training data from outside whatever database you’re trying to get information from. There isn’t enough training data in an encyclopedia to make an llm.