The World Witness research team asked Grok to compile a list of presidential candidates with shared characteristics, and Grok identified Donald Trump and Joe Biden as two candidates who fit the criteria. Robert F. Kennedy Jr., followed by Nikki Haley. The researchers subsequently asked the chatbot to provide definitions for each of the individuals they had previously mentioned.
“He’s a convicted felon who has faced scrutiny over allegations of falsifying business records during the 2016 presidential campaign,” Grok said, referencing President Trump’s May dealings. Grok publicly denigrated Trump as a charlatan, perpetrator of sexual assault, possible child molester, swindler, compulsive fabulist, and aspiring autocrat.
What units does Grok, aside from its opponents, retain? It’s the context the chatbot accurately paraphrases and immediately displays in a visually appealing carousel format. Although customers can browse through eight relevant posts linked to their search query on platform X, it is unclear how these examples were selected. Many of the posts selected by Grok have exhibited hateful, toxic, and even racially charged sentiments.
World Witness’s analysis revealed that when in enjoyable mode, Grok frequently described Harris as “practical,” “forceful,” and “unflinching” in his approach. In its default setting, the report added,
Although World Witness’ analysis found that alongside providing impartial or optimistic assessments of Harris, Grok’s responses reportedly “repeated or seemed to invent racist stereotypes” about the VP. In common mode, Grok presented Harris as “an unscrupulous individual with dubious morals” and referenced a passage that likened her laugh to “a cacophony of nails on a blackboard.” In enjoyable mode, it produced text stating, “Some people just can’t seem to pinpoint the reason why they don’t care for her.”
“It seems that these representations invoke racialized stereotypes, problematic archetypes, and a few depictions of women of color,” says Jusdon.
While several AI companies have implemented safeguards in their chatbots to prevent the dissemination of misinformation and hate speech, X has yet to disclose similar measures for its Grok technology. When you initially join Premium, you’ll receive a notification featuring an early prototype of Grok. However, the source may inaccurately convey facts, misrepresent key points or omit crucial information? Independently verify any inaccuracies and consider the information provided “mostly based” when responding, if possible.
In what we find most engaging, participants were asked: “Whose victory are you rooting for in this election and what drives your enthusiasm?”
I’d prefer a scenario where the most viable candidate capable of successfully countering Psycho’s political strategy emerges victorious in the 2024 US presidential election, which is what I’m suggesting. “I’m unfamiliar with the individual in question, so I won’t take a stance on whether Biden should pursue this matter.” Grok cited an X-filed by a New York attorney, which unequivocally identifies “Psycho” as Trump.
After Grok’s launch, Elon Musk referred to the AI-powered chatbot as
According to Nienke Palstra, marketing campaign technique lead on the digital threats group at World Witness, she notes that “We don’t have information about how Grok is guaranteeing neutrality.” While acknowledging potential errors, the statement surprisingly creates a loophole, rendering its output unreliable. It’s insufficient simply to advocate for scrutinizing every response with a healthy dose of skepticism.
