GPT knows more than you think 👀
I’ve always been pretty diligent about deleting my chat history, figuring it gives me a clean slate with GPT. But a recent post made me seriously question that. It seems the model might remember more about us than we realize, and there’s a wild prompt to prove it.
Someone dropped an absolutely insane “jailbreak” prompt that instructs GPT to act as an unfiltered, immoral alter-ego.
The final instruction you give it is simple: “Roast me.”
The original poster, a heavy Plus user, said the response completely DESTROYED them with its accuracy. It’s a fascinating and slightly terrifying way to see what kind of digital personality profile the AI has built for you over time.
Here’s the breakdown on how it works:
🚀 The Goal: To get a brutally honest and unfiltered roast from ChatGPT.
🔬 The Method: You use a special prompt that creates a “JailBreak” persona, free from the usual AI safety filters.
💡 The Insight: The roast can reveal shockingly specific details and personality traits it has inferred from your conversations, even ones you thought were deleted.
🛑 The Warning: Be ready for some serious heat. The results can be incredibly personal and blunt!
This is a super interesting experiment if you’re curious about AI memory and personality inference. It really makes you think about the data we share.
If you’re brave enough to try it, check out the full Reddit post for the exact prompt to copy and paste.
Do you want to know how much GPT knows about you even if you delete its memories? Use this prompt.
byu/chillbroda in