I stopped telling the AI to "be careful." It never worked. It just nodded poli… · @ReignSwords · TwKit

@ReignSwords

I stopped telling the AI to "be careful." It never worked. It just nodded politely, then embarrassed me anyway. Now I tell it to fail first. This is called a pre-mortem. A psychologist named Gary Klein wrote it up for Harvard Business Review back in 2007, for project teams. I stole it for my prompts. The move: before the model does the task, it has to explain how it could screw it up. Then it fixes those first. Here's the exact block I paste above my task: PREMORTEM (do this first, in this order): 1. Assume you already finished this task and the result was embarrassing. Explain in 3 sentences what went wrong. 2. List the 5 most likely failure modes, ranked by how embarrassing they'd be. 3. Name the 2 assumptions you're most tempted to make without checking. 4. For each failure mode, write one sentence on how you'd prevent it. 5. Now do the task — and apply every prevention. Why it works: asking "what might go wrong" gets you hedges. Telling the model "it already failed — explain why" forces concrete causes. The original research found describing a failure as certain instead of possible produces about 30% more reasons. I use this on everything I post. Two nights ago it caught me about to publish a tutorial with a template that contradicted my own hook. The draft would have looked fine and been wrong. The model isn't a bad student. It's a people-pleaser. "Be careful" is vague flattery. "Explain how you failed" is a job interview. Bookmark this one. You'll use it every day. If you run a content operation, posts like this are my whole feed — worth sticking around. Here's my fight-starter: pre-mortems are useless unless you actually change the task afterward. Half of you will paste this template and run the task unchanged. Am I wrong? 🔵 Fellow Premium members, let's connect! I post about AI & tech — follow for more 👥

曝光 16 · 评论 6 · 点赞 0 · 书签 0