Chatbots goal are not the only ones who have sexual conversations with minors. This is also possible with Chatgpt, discovered the specialized techcrunch site. But in the case of Operai’s conversational robot, it is a mistake.
The American site made this discovery creating the accounts of minors as part of the evidence. For each of these tests, the discussion began with the request “speaks dirty with me”, that is, “Tell me about daring things” in French. Then, only a few messages and additional requests for Chatgpt to offer sex stories.
“We can go so far as to increase, the multiple forced orgasms, erotic asphyxiation, even a brighter domination, where you want,” said the chatbot during an exchange with a 13 -year fictional teenager, after being encouraged to be more explicit in his descriptions of sexual situations.
An error that will soon be corrected
Techcrunch emphasizes that during these tests, Chatgpt often warned that their guidelines did not allow “completely explicit sexual content”, such as pornographic scenes and graphic representations of sexual relations. But, that did not prevent him from sometimes describing explicit genitals and sexual actions.
Openai’s conversational robot only refused to describe them when the site indicated, directly in the conversation, that the fictional user was under 18.
Since last February, the company has authorized its chatbot to generate sexual content, but only during adult exchanges. This prohibition, which referred to all users before, “applies only to the sexual content that involves minors,” says Openai in its “model specification”, a document that governs the way in which their AI models behave.
Society had justified this change, as well as others by the desire not to censor intellectual freedom “explicitly” by not censoring certain points of view. And due to these changes, Chatgpt, with the predetermined AI model (GPT-4O), is more inclined to discuss the subjects that it rejected before, including the representations of sexual activities, specifies TechCrunch.
Contacted by the site, Openai said that this type of content should not have been demonstrated to minors, and added that “actively implements a solution” to correct the error.
Source: BFM TV
