Elon Boasts of Grok’s Incredible Cognitive Power Hours After It Called for a “Second Holocaust”
During an hour-long livestream on X-formerly-Twitter Wednesday evening, billionaire Elon Musk made a series of characteristically hyperbolic claims about his AI startup xAI’s latest AI model, Grok 4.
He touted the new model as the “smartest AI in the world,” claiming that it’s “smarter than almost all graduate students in all disciplines simultaneously.”
He called it a “super genius child that will ultimately outsmart you, but you can instill the right values” and “encourage it to be truthful, honorable.”
But there’s one glaring problem: Musk’s bragadocious statements glossed over the fact that Grok had spent most of this week spewing mind-bogglingly racist and antisemitic talking points. The unhinged algorithm even started referring to itself as “MechaHitler,” targeting Black and Jewish people with shocking vitriol.
It went as far as to call for a “second Holocaust” — in lockstep with the disturbed beliefs of current-day Nazis calling Musk’s social media platform their home.
Put simply, Grok appears to be the worst pick of the bunch by far to “instill the right values” in its users.
Musk has since glossed over his AI’s Nazi tendencies, tweeting Wednesday that “Grok was too compliant to user prompts.”
“Too eager to please and be manipulated, essentially,” he added, suggesting that Grok was pleasing X’s user base by calling for another Holocaust. “That is being addressed.”
A person who previously worked closely with Grok models told The Information that the underlying Grok model likely didn’t have any Nazi tendencies. However, a new version released on X appeared to lack some basic controls that are usually added after the pretraining phase.
In other words, Grok may have the chops to compete with other models like Google’s Gemini 2.5 Pro and OpenAI’s o3, but in the hands of its current caretakers — Musk himself has made Nazi salutes and joked about the Holocaust — it may continue to have racist meltdowns.
And on a technical level, it remains to be seen how Musk’s hyperbolic claims about Grok 4 will shake out.
Standardized benchmarks are only one way to measure the intelligence of AI models, and often fail to reflect real-world performance.
Even Musk admitted during Wednesday’s stream that Grok may occasionally “lack common sense” and that xAI will still need to “close the loop around usefulness” to make sure its AI is “not just book smart, but actually practically smart.”
