Imagine you are in a conversation with someone you've just met.
What is the best question you could ask them?
Putting aside egoic interests such as status or influence, we will assume the goal of the question is to learn maximal information about your conversational partner. Following Claude Shannon's definition of information, you are seeking surprise. The lowest quality question would provide one bit: Yes/No.
The answer can even be less than a bit if it confirms what you already suspected, e.g. "Are you enjoying the sunny weather?". To avoid the drudgery of small talk, one naive pitfall I used to (and still do) fall for is to ask a broad and deep question right away. e.g. "What's the meaning of life?" It's a big question; but it often elicits a small answer.
Why?
Asking such a broad, deep question with no context is like pointing to a big vague cloud of information and asking "what's that big vague cloud about?", only to receive the response "well it looks like a big vague cloud to me".
The cloud of information analogy is useful to frame the conversational information space. Imagine a person's mind like an enormously intricate network of connected clouds of varying density and rigidity. They associate some of these clouds similar to you, and some vastly different.
Your question can spotlight a corresponding section of their association cloud, prompting them to describe its shape.
Consider a better question: "What is the biggest impact a book has had on you?"
You might learn:
Prompting a person to get a useful answer is a similar task as prompting an LLM. It's good communication.
An LLM prompted with nothing will produce a distribution of tokens that resembles a random sampling of its training data.
Good prompting leverages the information you posess to reveal connections in your conversational opposite (be it LLM or human) that surprise you.