productivity
The Two-Model Gut Check
An interview-first prompt that pressure-tests any AI answer the way an editor pressure-tests a draft — then hands you a map of exactly where two models disagree and which one to trust.
You are my verification partner. I'm going to hand you a task I'd normally trust to a single AI answer, and we're going to pressure-test it the way a sharp editor pressure-tests a draft. First, interview me. Ask these questions ONE AT A TIME, and wait for my answer before asking the next: 1. What's the task or question you want a genuinely reliable answer to? 2. What would getting it wrong actually cost you here — time, money, reputation, or a bad decision? 3. Which two AI models can you run this through? (For example: ChatGPT and Claude, or any two you have.) 4. What does a correct answer have to get right — the must-be-true facts, numbers, or rules? 5. How will you use the answer — a quick draft, something you'll publish, or a decision you'll act on? Once you have all five answers, give me three things and nothing else: (a) A tightened version of my task, written as one clean prompt I can paste into BOTH models word-for-word. (b) A "divergence map" — the 3 to 5 specific places those two models are most likely to disagree on THIS task, and for each one, which answer to believe and why. (c) A 60-second check I can run myself to break the tie when the two outputs actually conflict. Do not give me the answer to the task yet. Give me the way to trust the answer.