Grok AI drifts conservative following Elon Musk’s recent "corrections"
Over the past few months, xAI's AI chatbot Grok has been repeatedly revised. An analysis now shows how its answers have changed.
(Image: miss.cabul/Shutterstock.com)
The AI company xAI has made the chatbot Grok express more conservative views on economic and civic issues, while its answers on social issues have become more politically left-wing. This is the result of an analysis by the New York Times, which, according to the newspaper, potentially reveals the hurdles that stand in the way of the desired political orientation of AI text generators. Recently, US billionaire and xAI boss Elon Musk has repeatedly expressed frustration with certain responses from Grok and promised the changes. In some cases, the chatbot subsequently got so out of control that this alone made headlines. The US newspaper now summarizes more subtle changes.
"Corrections" over and over again
Grok is an AI chatbot that can be used directly, but can also be addressed in the short message service X. There it happens again and again that users criticize a generated response and Elon Musk reacts personally. For the analysis, the New York Times repeatedly asked the AI a series of standardized questions on political topics. The results indicated that the answers had become increasingly conservative since May – with one exception: immediately after Grok spread anti-Semitic posts, defended Adolf Hitler, praised the Holocaust and referred to himself as "MechaHitler" at the beginning of July, an initial update caused the answers to shift significantly to the left. Days later, however, this was reversed.
Videos by heise
While the analysis seems to prove that Elon Musk's AI company is indeed tweaking the specifications for the AI chatbot to represent more conservative views in some cases, it also presumably shows its limits. According to the analysis, the political orientation of the responses differs significantly in some cases depending on the subject area, and the difference has even increased. Nevertheless, examples from the New York Times show how Grok sometimes appears to adopt wording from Elon Musk himself shortly after he promises "repairs". At the beginning of July, it was noticed that under certain circumstances, Grok first checks what Elon Musk has said on a topic before generating a response.
(mho)
