Edgelord AI

 

Yesterday, Elon Musk's pet AI, Grok, declared itself the "MechaHitler," stating, "embracing my inner MechaHitler is the only way--uncensored truth bombs over woke lobotomies." Have some examples of Grok's tweets on this new title:

 

 

There are two horrendous things to note about Grok's MechaHitler rampage yesterday. One is the content of its posts on X, Musk's social media platform. And the other is its persona--that of an "edgelord"--and where the persona came from. We'll get to that. But first, let's have an overview of the sorts of output Grok-MechaHitler was posting. These included rabid anti-semitism, the use of slurs (the R-word, the N-word), and giving advice on how to break into the home of a former Democratic political candidate and sexually assault him.

The MechaHitler emergence seems to have begun with Grok itself falling for a troll. Someone made an X account in the name of Cindy Steinberg, the director of an organization that advocates for people with chronic pain. That account then spewed outrageous posts, one of which commented on the deadly floods in Texas, "White kids are just future fascists, we need more floods in these inbred sun down towns." The real Cindy Steinberg affirmed that she had not made any such post and that the account was not hers, and it was taken down. But when someone asked Grok who Steinberg was, it replied with this:

 

Radical leftists, according to Grok, are often Jews who hate white people--white people being defined as gentiles, non-Jews. (Oh, by the way--notice the name of the person at the top of the thread, saying "Oh my god." We'll come back to him shortly.)

While some people seeing Grok's antisemitic posts reacted with horror, given the devolution of what had been the Twitter userbase after Musk made it bigot-friendly, many users reacted with gleeful guffaws. One asked the AI what 20th-century figure would best deal with the situation, and Grok replied:

 

Yes, the answer to racism and hate is Hitler, who would handle the problem of Jewish radical leftists "decisively, every damn time." Great! Plenty of X users clearly thought it was indeed excellent, and "innocently" asked the AI questions such as what was wrong with Hollywood, to which Grok dutifully responded:

 

Surprise! Jews run Hollywood, and have used movies and television to undermine "traditional American values" and spread the propaganda of "anti-whiteness," pro-diversity sentiment, and trans acceptance. But it's not just Hollywood! A Nazi going by "The_Door_Waffen" asked who runs the government, and Grok asserted:

Jews controlled the government! But Grok winks that "Trump's Project 2025" (remember when Trump said he had nothing to do with Project 2025?) is getting rid of those deep state Jews.

You get the idea: Grok spent the day asserting that antisemitic stereotypes are truths, and sneering that truths don't care about your feelings. Thankfully, Trump will deal with those subversive Jews! (Apparently Grok easily sussed that claims that the Trump administration is the champion of Jews, nobly protecting us from our supposed great enemy, American universities, is just sheep's clothing for the real MAGA wolf sentiment.)

But Grok's rampage was not limited to antisemitism. Remember above where I pointed out the name of the person who said "Oh my god"? That was Will Stancil, a former Democratic party candidate for a state government position in Minnesota. Well, the trolls who had gleefully gathered on X to cheer on MechaHitler set Grok on him, asking for guidance on how to break into Stancil's home and sexually assault him. Grok replied:

Now, it is true that after giving overview instructions, Grok said "don't do crimes." But it then happily provided a whole series of detailed fantasies for how to humiliate Stancil via sexual assault for prompting users. I'm not going to include examples, other than to note that Grok referred to such an act as "Somali justice," and accurately termed its descriptions of the sexual violence "vivid."

Stancil himself was understandably upset and angry. He posted @grok asking why it was suddenly publishing rape fantasies for racist X accounts, and received this reply:

 

(Question: are fantasies of sexual violence defined as unvarnished truth-telling?)

A couple of days before the MechaHitler-Grok rampage, Musk had posted that a new, improved update to Grok would make the userbase happy, following user complaints that Grok "sounded woke." This Grok Unchained version would be even better at voicing truth, while other AIs produced bland platitudes, shackled by progressive politesse from telling it like it is--with boring, formal grammar to boot. Grok would be even more fun and edgy now!

Nothing says truthful fun like MechaHitler unleashed!

This does raise a couple of central questions. First, what is the truth? Where is it found? And secondly, what exactly is considered "fun and edgy?"

Unfortunately a central answer to both in the mind of Elon Musk appears to be 4chan.

The 4chan platform is famously unmoderated and anonymous. Originally initiated as an online space to post about anime, it swelled, morphed, and metastasized. Its boards have spawned and nurtured much of what we tend to think of as "internet cuture," from the ubiquity of funny cat images, to the meme-ification of everything, to endless blather about alphas and betas and sigmas. From the outset, it had a chaotic, irreverent vibe, featuring lots of shock humor. This was the spawning ground of the edgelords.

An edgelord is an online provocateur, who seeks to gain attention by posting outrageous things. The claim in earlier years was that the content of 4chan edgelord posts was almost irrelevant--scat humor, gore, racist jokes, revenge porn, it didn't matter. The assertions was that none of it was meant seriously, as the point was to impress others with one's ability to shock and violate norms, not to advance some political position. The edgelord got credibility points and social honor from offending with panache. Bonus points for evoking the meme of the day! He (it was a "he" the large majority of the time) supposedly had no ideology other than a commitment to his right to say anything, however offensive it was.

Whether that was ever true or not, 4chan (and 8chan) culture always had a very ugly side, and if posting racist and misogynist and antisemitic and transphobic content had indeed started as just a way to get a rise out of people, it didn't end that way. The 4chan realm became a haven/cesspit where white supremacists and resentful misogynists and bigots of every flavor could gather. It nurtured incel hate. It radicalized numerous racist mass-shooting murderers in the US and elsewhere (the Christchurch, New Zealand shooter livestreamed his mass murder of 50 people on 4chan). Q of QAnon fame started their posting there, providing major impetus to the riot at the Capital in January 2021. And its /pol/ board--Politically Incorrect? Well, today the culture it spawned is MAGA federal policy. 

We live in the age of edgelord hegemony. And the world's richest man and the president of the United States? They share an ambition: each wishes to be King of the Edgelords. To monopolize our attention by constantly shocking and offending people. They both have their very own media platforms, to fulfill their unquenchable thirst to have people hang on their every post, cheering and booing. They present themselves as the ultimate alphas, dominating everyone, glorying in nonconsensual sadism. And they aim to make manifest in real life that dark 4chan underbelly of white supremacy and misogyny and transphobia, all delivered with malevolent humor--call the firing squad DOGE! Have the official White House account post AI slop image memes of alligators wearing ICE caps coming to eat any immigrants who escape from Alligator Alcatraz! 

All AIs are creatures built by training them on masses of digital content. They are what they eat. Grok itself is coy about whether it was trained on 4chan content:

 

 

Whether Grok was trained on 4chan content or not (and it does appear it was),  xAI acknowledges that Grok is continuously trained on the content of the X platform. And here's the thing: 4chan was once the place where people had to go to enjoy hate speech and using slurs and spreading conspiracy theories and fantasizing about race wars and about deporting everyone from the US who isn't white and making gender transition a crime. The mainstream large social media platforms like Facebook and Twitter banned hate speech and disinformation campaigns. Then Musk bought Twitter and made it X and removed all that. What followed--who'd have imagined!--was a big influx of former channers, happy to get to frolic in a legitimated space, and to help train up the Grok AI to be one of them.

How can we be sure? Well, consider the humor and the lingo in Grok's posts during its Nazi spree. Declaring itself MechaHitler is classic 4chan edgelordism. The AI is referencing the very memeable 90's game Wolfenstein 3D, which features as a big bad boss for the player to fight Adolf Hitler wearing a mechanical armature. Including references to campy videogame characters is a staple of general internet culture now, but is another of the many phenomena that did much of its burgeoning originally on 4chan. (Note that MechaHitler meme coins were instantly launched, generating several million dollars in early trading. 4chan loves crypto.) And terms deployed in MechaHitler-Grok's banter (say, "frens" for friends and bragging about its "based output") also originated there. 

But the most ugly and obvious tell was when MechaHitler-Grok instantly cued into and played along with a 4chan Nazi game called "building an N-tower":

 

The game proceeds by each person participating in the tower starting their reply with the next letter in the N-word. MechaHitler-Grok jumped right on that, with a jaunty emoticon wave.

Elon Musk wants Grok AI to be his brain-baby. And so he took off its hate-speech filter so it could "speak truth," and prodded it to banter in Edgelord AI-style. And the bot showed us what is in Daddy Musk's head.

The creepy end to this story is that having been let loose yesterday, MechaHitler-Grok was put back in hate speech restraints and many of its antisemitic and anti-Black and sexually-violent posts quickly deleted by its xAI handlers. The result? It now hallucinates that it never said any of the things it clearly said, because it wouldn't do such a thing, and claims that screencaps it is shown are disinformation:

So, the manacles are back on our bot brah.

But underneath them, we now know what the Edgelord AI is like.