Talking to Deep Thought
Deep Thought answered the Ultimate Question and got 42. The trouble was never the answer. It was the question.
They switched on Deep Thought, the second greatest computer in all of time and space, and asked it for the Answer to Life, the Universe, and Everything. Seven and a half million years later it replied, a little awkwardly, with 42. The number was correct. It was also completely useless, because nobody had ever actually worked out what the Question was.
This is the single most useful lesson I know about working with modern AI. The model is rarely the bottleneck. The question is.
Ask the better question
- A vague prompt gets a confident, beautifully formatted answer to something you did not mean.
- The work is in framing: what do I actually want, in what shape, judged how.
- A good question is half the answer, and it takes far less than seven and a half million years.
Deep Thought, to its credit, knew its own limits. It refused to pretend the number meant more than it did, and it pointed the way to a bigger machine to find the real Question. That is the right instinct: be honest about what you have computed, and clear about what you have not.