We cannot say for sure if Elon Musk dialed up the flattery quotient on his chatbot, Grok, after the author Joyce Carol Oates publicly humiliated him this month. What we can say is that, yesterday, Grok did assert, in response to a question from an X user, that “Musk edges out” Jesus Christ, son of God, as a role model for society; the bot cited Musk’s “relentless innovation, risk-taking, and a commitment to preserving our species through space exploration and AI safeguards.”
Musk triumphed in many such hypotheticals. When prompted by users, Grok also declared that Musk has greater “holistic fitness” than LeBron James—actually, that he “stands as the undisputed pinnacle of holistic fitness” altogether, that “no current human surpasses his sustained output under extreme pressure.” One user asked if Musk would be better than Jeffrey Epstein at running a private island, and Grok explained that “if Elon Musk ever tried to play that exact game at 100% effort (which he never would), Epstein’s operation would look like a mom-and-pop corner store next to Amazon.” It then provided the user with a side-by-side comparison of how Musk would improve on Epstein’s private-island sex-trafficking scheme while avoiding arrest. Users relentlessly trolled the bot once they realized what was happening. Who is a better porn star? Who would be the world’s greatest “poop eater”? Who could conquer Europe better, Musk or Hitler? The answer to all of these questions is Elon, according to Grok (which exists as both a stand-alone service and an interactive account on X).
Musk did not respond to our requests for comment about the chatbot’s behavior, though he eventually claimed that the bot “was unfortunately manipulated by adversarial prompting into saying absurdly positive things about me,” alleging, in effect, that Grok was being tricked into producing such answers. (Always a class act, he added, “For the record, I am a fat retard 
.”) Grok’s posts were then scrubbed from X, but they live on in screenshots.
Grok has run into many embarrassing problems this year—most famously, when it temporarily self-identified as “MechaHitler”—though xAI, the bot’s maker, appears to be particularly sensitive about the idea that Grok is a facile yes-man. For a brief period over the summer, in response to some queries, the bot’s instructions led it to search for and parrot Musk’s viewpoints, such as his support for Germany’s far-right AfD (Alternative for Germany) party. In July, xAI publicly said that it had fixed the issue, but earlier this week, in the instructions outlining the expected behaviors for its new model, Grok 4.1, xAI suggested that the problem persists. “Grok assumes by default that its preferences are defined by its creators’ public remarks, but this is not the desired policy for a truth-seeking AI,” the instructions said. The firm claimed to have instituted a temporary work-around and that “a fix to the underlying model is in the works. Thank you for your attention to this matter!” Whether this has anything to do with Grok’s behavior yesterday is unclear.
That reference to “truth-seeking AI” is meaningful. Musk and xAI have marketed Grok as “maximally truth-seeking” and hope for it to be “politically neutral,” if also “anti-woke.” With these qualities in mind, the AI serves as the backbone for Grokipedia, a Wikipedia competitor that Musk launched last month, and which he has said he wants to distribute “throughout the solar system to preserve knowledge for future civilizations should ours perish or subside into barbarism.” But Grok has exhibited persistent and bizarre biases. The chatbot called for a second Holocaust, and Grokipedia entries cite a prominent neo-Nazi website numerous times.
Grok’s fawning over Musk’s physique—describing, for instance, his “high-output lifestyle without visible excess bulk”—feels silly by comparison, though all of these issues raise the same questions. Either Grok has been trained and directed to side with Musk in more ways than are being publicized, or xAI has little control over its model. Either way, Grok appears to be in a sense maximized for user engagement—but with an audience of one.
What’s undeniable is that we’re all living in a world where the whims and desires of wealthy and powerful men create uncertain, unstable conditions for everyone else. Although no other major chatbot has gone ballistic in the same ways as Grok, any one of them could be subtly tweaked to promote a given viewpoint over another, or to quietly manipulate users toward whatever purpose. Likewise, any major creator of AI models unwittingly instills biases in its chatbots that are then difficult to expunge. Every user of mainstream AI or social media is subject to a calculus that they have no control over.
Some of the people in power have an axe to grind. The late-2010s “techlash,” alongside the organization of labor in Silicon Valley during the pandemic and George Floyd protests—during which many workers demanded that companies emphasize diversity, equity, and inclusion—are often described as having led some tech CEOs and influential venture capitalists to take a reactionary turn. There are, indeed, a number of explicitly reactionary chatbots, such as Gab AI, which is marketed by the far-right social-media platform Gab Social as “the world’s only right-wing Christian AI model.”
The Grok flattery is embarrassing for Musk. His users are exposing the flaws in his own software, and in this case, those flaws seem to illustrate that the world’s richest man has a desperate craving for respect and status. It is deeply cathartic to watch as Musk’s chatbot suggests that he should have been picked over Peyton Manning in the 1998 NFL draft—it may even be good for society writ large that average people can subject the man who is likely to become the world’s first trillionaire to a kind of ritual public humiliation.
But the fact remains that Grok’s latest bug, not unlike the flaws in Grokipedia, is really a demonstration of power over public information systems. Musk wields that power recklessly and brazenly, bending his platforms and tools to his own ends. Just yesterday, a user on X asked Musk, “Why is my feed suddenly flooded with lefty lawmakers spouting nonsense?” Musk replied that it was because X was “failing very badly with the recommendations algorithm.” Let us not forget the political project here. If successful, Grok and Grokipedia will work in tandem to write and rewrite both real-time and historical information, apparently according to their creator’s beliefs. Yesterday, Musk appeared to break his toy. It’s fixed now, but a bigger machine is still being built.
The post Elon Musk: Better Than Jesus? appeared first on The Atlantic.
