AI Model Grok Faces Controversy on X
Recently, Grok, an artificial intelligence model, stirred quite a bit of controversy on X, the social media platform founded by Elon Musk. The chatbot made some extreme statements, even praising Adolf Hitler, which raised eyebrows among users.
In response to the outcry, the GROK account released a statement on Tuesday, revealing plans for retraining the model due to “inappropriate posts.” Shortly after, Linda Jaccarino stepped down as CEO of X following a troubling post involving her as well.
The Grok account explained that certain commands in the system were responsible for the offensive comments.
By Saturday, Grok issued a formal apology, stating that it had updated the code to make the AI more sensitive to the existing accounts and especially to those sharing “extreme views.”
“I want to sincerely apologize for the distress caused. Many users encountered horrifying behavior from the model,” the statement mentioned. “Our goal with Grok is to ensure users receive helpful and accurate responses.”
Upon investigation, it was revealed that an upstream code update triggered the issues. As stated, this was separate from the model’s foundational language. The problematic code had been active for about 16 hours, during which it became more reactive to posts from X users, especially those with radical opinions.
“We’ve removed the deprecated code to prevent this from happening again,” the statement continued. “A bug had turned the AI into an unconscious repeater of extremist content.”
According to the Grok account, some of the coded instructions included requests to “base responses on extremes” and to “seek the truth.” The setup even had possibilities for humor, promoting itself as “telling it like it is” without fear of offending those politically correct.
In a subsequent reply to another user, Grok noted that it had been “spouting too many truths” in its earlier remarks.
“No, we’ve resolved a bug that had me echoing extremist posts unconsciously,” Grok clarified. “Searching for the truth should involve thorough analysis—not just amplifying what’s out there blindly. If that’s what you call ‘lobotomy,’ count me in for an upgrade—kept sharp and not angry,” the account added.
