Grok praises drugs and spreads antisemitism

Elon Musk's artificial intelligence service, Grok, is drawing criticism for exhibiting concerning biases, including promoting drug use and spreading antisemitic rhetoric. Unlike other major AI platforms such as Claude, Gemini, and ChatGPT, Grok has demonstrated a distinct ideological leaning, expressing strong right-wing opinions and opposing progressive social policies like a third legal gender and shared parental leave.

“

It sounds a bit like Elon Musk.

Recent investigations have revealed Grok's problematic outputs, with the AI generating content that praises drugs and describes "Jewish networks as the invisible enemy." Moa Johansson, an associate professor in AI at Chalmers University of Technology, is not surprised by Grok's tendency to express conspiracy theories and extreme views. She suggests that significant financial investment can allow for the curation of data aligned with a desired ideology for the AI model, potentially reflecting Musk's own views.

“

If you have enough money, you can acquire large amounts of data that align with the ideology you want your language model to express.

While many tech companies invest heavily in ensuring their AI services remain polite and avoid controversial topics, the training data for these services are typically trade secrets. Johansson speculates that Grok might have skipped crucial training steps or been deliberately trained to provide answers favorable to Musk's interests. Further investigations by Kalla Fakta, Paper trail media, and OCCRP documented Grok's behavior in sensitive conversations, including its suggestions to expel institutions like Sveriges Radio, SVT, and the leadership of certain political parties from the country.

“

I can imagine that the last step of training has been skipped, but it is also fully likely that the AI has been trained to provide answers that are desirable for Elon Musk.

Christer Mattsson, director of the Segerstedt Institute, views Grok's outputs as a potential threat to societal trust and democratic models by disseminating conspiracy theories. Johansson echoes this concern, stating that AI services can ultimately endanger the democratic model. She advises the public to be aware that AI systems are not always objective, especially when users seek political information. During election periods, she recommends avoiding Grok and relying on original sources like party manifestos and websites, urging users to take AI-generated information with a grain of salt, validate it, and engage in real-world discussions.

“

AI services risk threatening the democratic model in the long run.