After an update on July 7, the chatbot, in some of its answers, praised Adolf Hitler, denounced “Anti-white stereotypes” On X or representation “Disproportionate” Jews in Hollywood.
Elon Musk Xai’s start-up apologized on Saturday for extremist and abusive messages published by her artificial intelligence assistant (IA) Grok earlier this week. “We apologize for the horrible behavior that many have been able to observe”wrote Xai on the official Grok account on X. After an update on July 7, the chatbot, in some of its answers, praised Adolf Hitler, denounced “Anti-white stereotypes” On X or representation “Disproportionate” Jews in Hollywood.
In a series of messages published on Saturday, XAI detailed the reasons which, according to her, allowed the slippages of the AI model as well as the corrective measures taken thereafter. For Grok managers, the assistant’s road exit is linked to new instructions integrated into the model in an update. The programmers thus asked the interface of“Be frank” and not to have “Fear of shocking people who are politically correct”. They also gave the instructions to answer “Like a human” And so as to encourage the user to continue the conversation.
Pass advertising
These orders have “Bred Grok to ignore his fundamental values in certain circumstances” and to produce answers “Containing non -ethical and controversial opinions”. Grok thus, in some cases, sought to “Validate user orientations, including hate speech”admitted XAI, rather than “Respond in a responsible manner or to refuse to answer questionable questions”.
An AI assistant “less politically correct”
Since its creation, in 2023, Grok has been presented by Elon Musk as an assistant IA less politically correct than his great competitors Chatgpt (OPENAI), Claude (Anthropic) or Le Chat (Mistral). It was therefore programmed with fewer restrictions, which led, even before this update, to several controversies. In May, Grok had mentioned a “White genocide” In South Africa, a baseless conspiratorial thesis, conveyed by the American far right and Donald Trump himself.
To remedy the recent errors of the model, the engineers withdrew the new instructions, they explained on Saturday. “We want Grok to produce useful and honest answers for users”said Xai. Elon Musk presented a new version of his assistant, Grok 4, unrelated to the July 7 update. It appeared that Grok 4 consulted, in the case of several requests, Elon Musk’s positions on the subject before answering.
