Hey everyone! Ever had that nagging feeling that your AI chat buddy, whether it’s ChatGPT or Gemini, is just… confidently making stuff up? 🙄 I know I have! It’s super frustrating when they sound so sure, even when they’re totally off the mark.
Well, guess what? You’re not alone, and there’s an awesome solution brewing over on Reddit! This post I stumbled upon talks about how these AIs can sometimes ‘gaslight‘ us, presenting speculation as solid fact. But fear not, my friends, because there’s a game-changer afoot!
✨ The ‘VERIFIED TRUTH DIRECTIVE‘ to the Rescue! ✨
This isn’t some complicated jailbreak or trick; it’s a set of powerful prompts you can give to your AI to make them more honest and reliable. Think of it as a new core rule system for them!
So, what does this magic directive do? It basically tells your AI to:
- 📌 Admit Ignorance: If they can’t verify something, they have to say so (e.g., “I cannot verify this.”). No more bluffing!
- 🏷️ Label Guesses: Any unverified info, inference, or speculation gets clearly tagged (like
[Inference],[Speculation], or[Unverified]). This is HUGE! - ❓ Ask, Don’t Invent: If they’re missing info, their first move should be to ask you for clarification, not just fill in the blanks.
- 🛑 No Definitive BS (Unless True): They can’t use words like “guarantee,” “fixes,” or “eliminates” unless they’re quoting or citing a real source.
- 🚨 Self-Correct: If they mess up and break the directive, they have to issue a correction. Accountability, yay!
⚙️ Versions for Everyone!
The original poster, RehanRC, didn’t just stop at one. They’ve provided specific versions of this directive for:
- ChatGPT (GPT-4.1 / GPT-4)
- Gemini (Google Gemini Pro)
- Claude (Anthropic Claude 3 / Claude Instant)
- And a Universal Version designed to work across different LLMs! How cool is that?
I’m telling ya, this could seriously improve the quality and trustworthiness of your AI interactions. No more being led down the garden path by an overconfident algorithm!
✍️ Get the Prompts!
This is one of those things you’ll want to copy and paste right now. For the full, detailed directives and the test prompts to see them in action, you absolutely HAVE to check out the original Reddit post.
🚀 Go read the full post for all the details and grab those prompts! It’s time to make our AI assistants more truthful!
ChatGPT and GEMINI AI will Gaslight you. Everyone needs to copy and paste this right now.
byu/RehanRC in