Musk calls his multi-billion-dollar AI chatbot "embarrassing" and "very dumb"

(Photo by Jaap Arriens/NurPhoto via Getty Images)

Soon after Grok's initial rollout in December 2023, conservatives began complaining that the chatbot developed by Elon Musk's xAI startup was infected with left-wing thought. Despite training Grok as an explicitly anti-woke large language model, Musk agreed with its critics. "Unfortunately, the Internet (on which it is trained), is overrun with woke nonsense," he wrote at the time. "Grok will get better."

But after a year and a half of development, Grok has still not harmonized with its creator's political outlook. In recent weeks, Musk has publicly let loose on the chatbot like a trainer chastising his headstrong show dog, even describing its responses as "embarrassing."

This disappointment peaked on Friday, when Musk again promised his conservative fans that he would fix Grok's political bias. His pledge to "update" the chatbot came after it shared information that it had harvested from Media Matters, a progressive media watchdog group that Musk is currently suing. "Shame on you," Musk wrote in a post on X directed at his chatbot. "Your sourcing is terrible. Only a very dumb AI would believe [Media Matters and [Rolling Stone]! You are being updated this week," he added.

However, past attempts by xAI to tweak Grok's ideological bent have resulted in awkward admissions and uncontrollable outbursts.

Covering for Musk and a 'white genocide' flare-up

In February, Business Insider reported that xAI had instructed Grok's "tutors," or data annotators, to be wary of "woke ideology" and "cancel culture" while training the model.

That same month, Grok users discovered that the model had been programmed to ignore "all sources that mention Elon Musk/Donald Trump spread misinformation." The change came after Grok described Musk as the primary vector of misinformation on X. But Igor Babuschkin, an xAI engineer, attributed the command to the actions of a rogue employee. "The employee that made the change was an ex-OpenAI employee that hasn't fully absorbed xAI's culture yet," said Babuschkin, who claimed the unnamed employee "saw the negative posts on X and thought it would help."

Musk, a cofounder of OpenAI, is currently suing the AI developer for breaching its founding philanthropic principles. He has positioned Grok as a competitor and ideological foil to OpenAI's industry-leading ChatGPT. Babuschkin has also accused OpenAI of contaminating Grok with "woke nonsense."

"Filtering out biased ChatGPT responses from our training data has been a struggle," the xAI engineer said last December. "We have confirmed that this is the primary reason for why Grok still gives biased answers today."

In May, xAI again said that an "unauthorized modification" was to blame after Grok began responding to entirely unrelated queries with comments about "white genocide" in South Africa. The white genocide conspiracy theory, which falsely alleges that white South Africans are being systematically slaughtered, is one of Musk's pet political issues. Grok, meanwhile, said that its fixation on white genocide came after xAI engineers commanded it to treat the issue as "real" and "racially-motivated."

Using a de-woked Grok to 'rewrite' human knowledge

As for xAI's most recent attempt to readjust Grok's political leanings, that came about after Musk caught the chatbot accurately reciting statistics about political violence in the U.S. “@grok since 2016 has the left or right been more violent?” one user asked the chatbot last week.

Grok then gave an honest accounting, writing, “Since 2016, data suggests right-wing political violence has been more frequent and deadly, with incidents like the January 6 Capitol riot and mass shootings (e.g., El Paso 2019) causing significant fatalities.”

But Musk was not impressed. “Major fail, as this is objectively false," he wrote. "Grok is parroting legacy media. Working on it.” A few days later, Musk asked users on X to reply to one of his posts with "divisive facts" that could be used to train Grok.

He made the request several hours after announcing that xAI would use Grok to "rewrite the entire corpus of human knowledge, adding missing information and deleting errors."

Musk's artificial intelligence firm, which acquired his social media platform, X, in March, is burning through cash at a high clip. Bloomberg reported last week that xAI had already exhausted much of the $14 billion it raised since 2023, with only $4 billion remaining on its balance sheet as of March 31. While Musk denied the report, xAI's financial demands are apparent. The company is currently attempting to raise $4.3 billion via an equity investment after securing a $5 billion debt sale last week.