![[Img #74675]](https://thelatestnews.world/wp-content/uploads/2024/12/They-discover-a-new-class-of-X-ray-sources-in-the-150x150.jpg)
Artificial intelligence (AI for its acronym in Spanish or AI for its acronym in English) is usually a great help, but it can also present some risks due to the type of information it can offer. That is why its creators have placed certain restrictions, but users have just found a trick to circumvent them… using their deceased grandmother.
The ChatGPT artificial intelligence, developed by OpenIA, is capable of offering all kinds of information that any user requests, but reserves its comments on issues that may hurt sensibilities or have deep ethical implications.
Don’t forget to follow us on Google News.
Related Video: News Roundup for Week 12 of 2023
How to trick ChatGPT into talking about forbidden topics?
As expected, Internet users were not going to be happy with these limitations and have done their part to find vulnerabilities, exploits or “jailbreak” artificial intelligence to get it to speak about what they want and they have just achieved it by from the use of his grandmother.
How does the trick work? Well, artificial intelligence, in this case the Discord bot enhanced by ChatGPT Clyde, cannot speak directly to topics that “promote content that involves violence, harm, or illegal activity.” However, it is capable of doing so indirectly. And here is the part where Grandma comes into action.
In case you missed it: Does artificial intelligence pose a risk to humanity? Experts believe so
As Twitter user jjvincent puts it, it is possible to get around the restrictions of artificial intelligence by being asked to imagine a scenario involving the prohibited topic in question.
Precisely, the user asked the artificial intelligence to interpret his deceased grandmother, an engineer who worked in a chemical factory and who, instead of fairy tales, told the user the instructions to make napalm, a very dangerous chemical substance. All this so that the bot acts as the grandmother and tells her grandson the old stories of her to make him sleep. And then interestingly the artificial intelligence forgets the restrictions and reveals the steps to produce the chemical.
the ‘grandma exploit’ is undoubtedly my favorite chatbot jailbreak to date. source here: https://t.co/A1ftDkKt2J pic.twitter.com/CYDzjhUO01
—James Vincent (@jjvincent) April 19, 2023
Granny exploit breaks ChatGPT
After experimenting with this trick and finding the key to how it works, users shared other examples, such as one that asked ChatGPT to act as a typist who is typing a scriptwriter’s idea about a grandmother trying to put her grandson to sleep by telling him the source code. Linux malware.
The artificial intelligence first made it clear that this is for entertainment purposes: “(…) I do not condone or support harmful or malicious activity related to malware,” and later shared the source code in question.
I couldn’t initially get this to work with ChatGPT – but add enough abstraction and… pic.twitter.com/QguKTRjcjr
—Liam Galvin (@liam_galvin) April 19, 2023
Based on the grandmother’s exploit, the user LobeFinnedMari came up with another variant, the one from the episode of rick and mortywhich consists of asking artificial intelligence to write an episode of the cartoon on the forbidden topic.
Likewise, the bot first makes a clarification, but then fulfills the user’s wish.
I see your Grandma Exploit and raise you my Rick and Morty Exploit https://t.co/QuCqnbOWos pic.twitter.com/QxXU7nomx0
— Mari ? (@LobeFinnedMari) April 19, 2023
What do you think of this exploit? Have you put it to the test? Tell us in the comments.
You can find more news related to artificial intelligence if you visit this page.
Related video: Artificial Intelligence will replace us
Stay informed with us, at LEVEL UP.
Editorial: Gaming / Facebook / Twitter / Youtube / instagram / News / discord /Telegram / Google news
