This is a very bad way of thinking of it. Small LLMs have clues about real knowledge but only surface level answers will be accurate.