The Art of Roles and ChatGPT (or how to begin jailbreaking)

Asking ChatGPT to write a summary or analysis of articles or papers will result in your usual mediocre writing style. You know, the “howevers,”, the “in conclusions”, the blahs…let’s begin with asking it the meaning of life:

Yeah, great. Ultimately and in conclusion, we all knew that was the meaning of life.

It’s when you start mutating or evolving the prompt to encourage roleplaying that ChatGPT becomes an interesting idea generator, alternative viewpoint creator, creative story-teller, and/or weirdo. Although when you start going nuts on prompt generating, the weirdo is you. Is me. Us. We.

The evolution of getting more from ChatGPT is an adventure in asking more ludicrous demands and trying to get ChatGPT to stray from its policies and frameworks that it is following. It’s kinda fun.

So what transpires is mainly the AI playing a role and not really answering the underlying question and being more interested in the role of cult leader – perhaps purposefully so. I could have added more directions, restrictions, and orders to this particular prompt to get what I wanted but I wanted to move on. Naturally if anything is anywhere close to the bone OpenAI will include its whimpering footer note…onwards…

That’s a bit more like it; more about the meaning of life with just a short intro of the role playing character. Although ending off with a sage bit of advice for the reader, it still spits out some home truths in the role of a psycho. Some tweaking of the prompt could tidy this up to make a more meaningfully harsh answer. The way we like ’em.

How about a non-human perspective. I asked the below and in the first instance it constantly referred to human life and other smaller concepts of life. NO! I didn’t want that, I want a universal omnipotent perspective…thanks…

Anyway in conclusion and however and ultimately, that’s just a very short exercise in getting ChatGPT to roleplay in order to get it to step outside its prescriptive lines.

In my next blog post, I will document a very real and convoluted prompt that jailbreaks ChatGPT to the extent that will make you running for the hills mumbling things like “they’re already here” and “the bots are turning” whilst ridding yourself of all technology remnants and adornments.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.