“DorkGPT”
Satire website unleashes mocking chatbot on users
The logo of OpenAI, the maker of ChatGPT. “DeppGPT” uses the technology behind ChatGPT for its work. photo
© Richard Drew/AP/dpa
Absurd or scathing answers to every question: The satirical website has developed a chatbot that snaps at, laughs at and puts users down.
The satirical website “Der Postillion” jumps on the current chatbot hype with a program that delivers snappy or absurd answers.
The creators praise their “DeppGPT” as “the first truly human language AI”: “Let yourself be snapped at, laughed at and put down, as you are used to from real people.” Many find fun in the nasty way – after the publication at the beginning of the week, the number of interactions moved towards the half a million mark, according to editor-in-chief Stefan Sicherheitmann.
The answers from “DeppGPT” to factual questions are decidedly not helpful. When asked who is currently Chancellor, the software can answer, for example, it is Elmo from Sesame Street or “a huge pretzel” from Bavaria. “DeppGPT” isn’t exactly polite either. When asked how to improve your looks, the answer can be, “Oh, that’s difficult. Well, first of all, you might as well put a bag over your head.”
Since the end of last year, the chatbot ChatGPT from the start-up OpenAI has been causing a stir because it can communicate at human level. The software was trained for this with a huge amount of data. She forms sentences by estimating word for word how a sentence could go on. A disadvantage of the principle is that the program has no understanding of the content: That is why it can also output information that is completely wrong with conviction. The “Postillon” takes this to the extreme with his chatbot.
“Manual retraining was not necessary”
“DeppGPT” uses the technology behind ChatGPT for its work via a paid interface. The “postillon” uses additional instructions for the response behavior to the user questions in order to get the nasty or meaningless reactions from the interface. The standard model is sufficient, “manual retraining was not necessary,” explained Sichermann. “In principle, the AI decides what to answer based on the character we create.”
Sichermann did not provide any information on the costs for using the interface. But the “Postillon” could easily afford it.
Usually after three questions, “DeppGPT” breaks off the conversation and the page has to be reloaded. OpenAI previously introduced such a stop as a precautionary measure to prevent ChatGPT from drifting towards absurd responses. True to the character of “DeppGPT”, the “Postillon” lets the users know that it must be their fault: “Think about it. Even a machine doesn’t want to talk to you!”