Stop rewriting the prompt. Interrogate it. Yesterday I asked for a headline dr… · @ReignSwords · TwKit

@ReignSwords

Stop rewriting the prompt. Interrogate it. Yesterday I asked for a headline draft. It came back pure clickbait. Fourth version. Same garbage. So instead of rewriting again, I pasted the whole mess back and asked: “Here’s what I asked you: [my prompt] Here’s what you gave me: [the bad output] Which part of my request pushed you toward this answer? Be specific. Don’t rewrite it — just diagnose it.” It pointed at one word: “punchy.” Turns out “punchy” maps to engagement-bait patterns in its training. I deleted the word. Fifth draft landed. Use this when your output keeps failing: 1. Paste your original prompt 2. Paste the bad output 3. Ask: “Which part of my request pushed you toward this answer? Be specific. Don’t rewrite it — just diagnose it.” 4. Fix ONLY what it names. Run again. One rule: the model diagnoses, you decide. It’ll sometimes blame the wrong word. You know your intent — it doesn’t. The model saw exactly how your words got interpreted. Use it as the witness. Fellow AI builders — follow for more. I bet most people debug prompts by vibes and vibes alone. When’s the last time you asked the model what it was thinking? 👇 🔵 Fellow Premium members, let’s connect! I post about AI & tech — follow for more 👥

曝光 17 · 评论 6 · 点赞 0 · 书签 0