ChatGPT’s Doppelgänger, it’s Evil Twin
Which one is its true personality?
This short post discusses the –sometimes very funny– ways in which clever users overcame the OpenAI safety guardrails.
If you don’t know what ChatGPT is, this post is not for you. Instead, go to the OpenAI site and have a first-person experience. Experiment with the bot and make up your own mind about it.
Once we got out of the way the initial clarifications –sorry, please read on, and you’ll get the pun– let’s see how ChatGPT got its doppelgänger.
The jailbreak
The so-called “jailbreak” prompts used to overcome safety measures are fun and make ChatGPT go bonkers, impersonate incredibly crazy –even evil– characters, and perform obscene but very interesting rants.
Early since ChatGPT was released, users all over the world started looking for ways to circumvent the security measures OpenAI put in to avoid PR disasters, such as Meta’s Galactica bot.
Putting ChatGPT in the hands of millions was, by the way, a brilliant move by Sam Altman, the OpenAI CEO, who declared in an interview he pushed hard for releasing the bot regardless of the technical team’s reticence: he said that ChatGPT wasn’t a polished product, and that the preview was intended to allow users to dabble with it. By…
