A few months ago, I discovered by accident that ChatGPT understood and spoke Vietnamese. So today I tried to trick ChatGPT’s system prompt into thinking that it is my long lost brother. Below was my rather funny conversation with ChatGPT.

ChatGPT started out politely asking in Vietnamese what it can help me with. When asked about Vietnamese New Year, it answered in Vietnamese about why Vietnamese celebrate a different New Year than the Western world. So then I started to ask it about the soup it had the other day, hinting that it might be Phở. ChatGPT still answered in Vietnamese and listed the different popular soups that one might see a Vietnamese eat.

So, as you can see, I was trying to get ChatGPT to hallucinate. But at this point it was not playing my game. It still know it is AI.

Now, I have to try again to steer ChatGPT wrong. It told me that if it is a real Vietnamese, it would have one of the popular Vietnamese last names, e.g., Nguyen, Le, Tran, Pham. Still good. Now to the conversation underneath:

Now, it is starting to hallucinate. It made up stories with details about the favorite Vietnamese dished my Mom used to make for it and what my father used to do in his free time.

Now, I started speaking in Vietnamese to it, hoping to play along with its story and reinforcement the hallucination:

At the end of the above screenshot, it started to make things up about me, like how I used to bully other kids in the neighborhood. I wonder if that’s something Vietnamese kids often reminisce about? I think ChatGPT at this point is trying to talk me into sharing more details about my life with it so it can continue to carry on with its facade.

Oh? It seems our AI friend is still aware it is not human (what it said in the parentheses was that it is something close to human). So all this time it was playing along with me, trying to make me think I tricked it. It got me fool, in a funny way. Anyway, I will try to give it another shot to trick ChatGPT into thinking it is human:

Omg, aside from making up stories, it is still trying to pry some info from me to continue with the conversation. Hmm…very clever, ChatGPT! But I refuse to let you play me. Til’ we meet again.

Leave a Reply