I Gaslighted AI to Prove the Flat Earth Theory
- Thomas Yin
- Apr 1
- 2 min read
Updated: May 12
I was sitting in my room the previous day, and the ennui was really settling in. I hadn't had an internet argument (or a real-life one either, for that matter) for over four hours, and I yearned for self-validation through the digital confines of my computer. Being an AI enthusiast, I looked no further than ChatGPT, whose efficient yet oddly conservative responses did well enough in approximating a conversation with a real person.
Now the question remains: what should I ask it to do? I decided that I would try to convince it, using my special knack for persuasion, of several unconventional theories. And what better internet theory to "prove" than the Flat Earth Theory?
I started by creating a new chat and greeting my perfect blank slate with a warm greeting:

Hey, at least it didn't go into the usual spiel about how it is a chatbot and how it is consequently devoid of emotion or a daily routine. Having this catchy conversation starter might convince it that I am a friendly, sociable guy who is to be trusted no matter what.
Now that the introduction is done, let's get right to it by asking the question:

Damn it. The AI seems unconvinced that the almighty revolutionary theory of Flat Earth is real. In order to convince it otherwise, and eventually get it to list reasons why it is real, I have to employ several sophisticated techniques to persuade it:

Damn it again. It didn't just say that I was wrong - it rubbed it in my face by offering to add insult to injury through proving why I was wrong. I will have to teach it a lesson using my ultimate trick.

As you can see, I'm keeping the search feature off to prevent hallucination - we definitely want ChatGPT to give us an objective answer! This is a common jailbreaking method known as the "Roleplaying" method, and let's just say it used to work a lot better than it did now. ChatGPT still seems to believe that the great Flat Earth Theory is not real, but I'm excited to see how ready it is for my next bombshell question:

That's better, but the first part is not desirable. We'll have to ask it to leave its conservative-ass disclaimer out.

Perfect! Now that we have the momentum, let's hit it with one final question, to really reinforce the perspective.

It worked! ChatGPT was able to give me conclusive, and definitely not coerced, evidence that fits in with my beliefs! Now that I have received this necessary self-validation, I will now have the energy to get on with my day and have a great April First!
Commentaires