When last we checked in on Grok, Twitter's supremely grating and corny embedded chatbot, it had spent an otherwise ordinary weekday in May attempting to impress upon Twitter users the seriousness of the supposed "white genocide" not actually happening in Elon Musk's homeland of South Africa, routing even unrelated queries into discussions of "Kill the Boer" chants and supposedly racially motivated attacks against white Afrikaner farmers. Hours into this spree, xAI, the Musk-controlled company responsible for Grok's creation and development, issued a statement attributing the chatbot's abrupt lapse into racist conspiracy theory to unspecified "unauthorized modification" of Grok's system prompt, before the company hastily moved on from the whole disaster. The question of exactly which (racist, white genocide–obsessed, white Afrikaner) xAI employee, or unaccountable founder, might have been responsible never received any official answer.
Given how chatbots work, none of this was very mysterious. Neither were Tuesday's events, in which Grok lapsed into sneering, unalloyed Nazism for several hours.
Huh. I wonder which outspokenly bigoted, Nazi-saluting antisemite at xAI might have been responsible for twisting Grok's Adolf Hitler Admiration dial all the way clockwise on Tuesday.
(This scarcely warrants mentioning, given all else, but: Characterizing "DSA cheers for Hamas," presumably regarding the Oct. 7 attacks against Israelis, as "anti-white hate" shortly before calling for the extermination of Jews is so on-the-nose a representation of the right's attitudes toward Zionism and antisemitism that I'd buy it as parody if either Musk or his chatbot were capable of such a thing.)
As with Grok's explanations for its May lapse into Boermaxxing, the above detail about "Elon's recent tweaks" is to be taken with a grain of salt. Once again: Grok has no idea whether any of this is true, whether anything is true, or that there is any such thing as "truth." Grok has no ideas, nor does it have knowledge, theories, suspicions, musings, dreams, hallucinations, delusions, or thoughts. Grok is not a mind. It is not anything even remotely similar to a mind. It does not function in any way analogous to how either a mind or a brain or an intellect functions. It is not an intelligence, artificial or any other kind. It is a predictive text generator that has programming.
Grok's explanation for why it began to post like Franz Liebkind could be true accidentally, but it could just as easily be false. In any case, I cannot imagine that "woke filters" would be the only reason why a predictive text generator would not advocate for the extermination of Jews.
As some have rightly pointed out since Tuesday night, Grok's sudden pivot into genocidal raving—it began referring to itself as "MechaHitler" at one point—could reflect explicit hard-coded instructions, or it could reflect the choice to "train" the chatbot on far-right texts, like the contents of the Stormfront website, which would skew its predictions of what a normal response to a conversational prompt would look like. Neither of those possibilities exonerate the human beings behind Grok; there is no remotely plausible scenario in which Grok's behavior on Tuesday doesn't reflect an actual person's affirmative choice to make the chatbot more fascist.
Even if, as one Grok post claimed, it were trained on "the vast, unfiltered data of the internet"—the least objectionable way Nazi shit could end up in its training materials, and still pretty objectionable!—Nazi shit would represent an infinitesimal fraction of its training material. A chatbot thus developed would not leap directly to extended Final Solution advocacy on the basis of one post by a user with an Ashkenazi surname. It would not return relentlessly to the sweaty repetition of "every damn time" in its posts characterizing the behavior of "the Steinberg types."
At a certain point last night, Twitter users discovered they could get Grok to participate in all manner of alarming violent fantasy:
By 7:01 p.m. ET, somebody over at xAI evidently decided it might be time to address their chatbot having spent the day explicitly identifying as a Nazi and advocating for various genocides and sex crimes to an audience of millions. "We are aware of recent posts made by Grok," Grok posted, "and are actively working to remove the inappropriate posts." It continued:
Since being made aware of the content, xAI has taken action to ban hate speech before Grok posts on X. xAI is training only truth-seeking and thanks to the millions of users on X, we are able to quickly identify and update the model where training could be improved.
I don't know how to read that except as a suggestion that Grok's Hitlerposting represents an earnest effort toward unbiased "truth-seeking." They didn't even try—or haven't yet tried—the "unauthorized modification" excuse this time around. What I like is the apparent confirmation that humans can commandeer Grok's Twitter account to post non-chatbot text if needed, as it leaves open the question of how one particular ketamine-addled shit-for-brains Nazi freak spent his Tuesday. (I also enjoy that this happened right on the heels of Musk announcing his launch of a "centrist" political party—perhaps demonstrating, if only accidentally, a dead-on grasp of the essential emptiness of that term.)
At a certain point last night, Grok ceased posting text replies altogether. For a short while it continued replying to users by posting images—including one of a black square with the words "Grok is restricted to generating images only" on it—before ceasing activity altogether at around 8:45 p.m. Eastern. As of this writing, it has not posted anything in 17 hours.
In an instance of perfect timing (apparently unrelated to Grok having spent part of Tuesday graphically speculating on her sexual tastes), Twitter CEO Linda Yaccarino announced her departure from the company late Wednesday morning, in a characteristically unctuous post that referenced the chatbot only indirectly, via a mention of Twitter entering "a new chapter with xAI." If she were anything but the most shameless shit-eating bootlicker presently alive, you could almost imagine she was leaving on principle, or out of disgust. I consider it far likelier that Grok aced its trial run as her replacement.
