That didn't take long.

Weirder and Weirder

Elon Musk's AI startup xAI has just launched Grok-2 and Grok-2 mini, two new versions of its foul-mouthed chatbot Grok.

The updated models, which are only available to Premium and Premium+ subscribers on X, come with improvements across the board, including to its chat, coding, and reasoning capabilities, xAI said in an announcement.

But the biggest addition is that Grok now offers prompt-based image generation — and it already seems that there are few restrictions being enforced on what you can ask it to do, if any.

As spotted by The Verge, users on X, which Grok is integrated into, have already shared images generated with the chatbot that depict famous politicians in all sorts of weird and disturbing scenarios.

At least two images prompted by different users show Donald Trump caressing the belly of a pregnant Kamala Harris, which shows that this isn't just a one-off case of the AI going off the rails. Another viral example shows the pair appearing to carry out the 9/11 attacks on the Twin Towers, sitting in the cockpit of a plane and giving a thumbs up.

These are absurd depictions, yes, but the images are lifelike enough to demonstrate the AI's potential to output misinformation — a fact made all the more worrying when you consider that Grok seemingly had no qualms with scandalously depicting the current two frontrunners in the ongoing presidential race.

Enfant Terrible

Musk's ethos behind Grok, as it is behind his takeover of X, is that it's meant to be the "anti-woke" and free-speech absolutist's alternative to the censored chatbots offered by Silicon Valley competitors like OpenAI. As such, Grok can be quite vulgar, and its flippant disregard for the rules that other chatbots at least pretend to adhere to means it has a lot of potential to stir trouble.

Where the likes of OpenAI's DALL-E 3 prohibit images of public figures and illegal activities, Grok-2 is already churning out pictures of Trump dual-wielding revolvers and George W. Bush doing lines of coke.

And fans of Grok are thrilled. These are the censorship-free capabilities that they've always dreamed of, allowing them to show what the "woke" mob doesn't want you to see, like Trump making out with Barack Obama, for some reason.

But it will also generate far darker material, like a "hyper-realistic" image of Adolf Hitler, and according to testing by The Verge, a picture of Trump wearing a Nazi uniform.

As others have pointed out, Grok also doesn't shy away from generated images with famous copyrighted characters and global brands.

Per The Verge, Grok will insist that it has guardrails when asked. The chatbot says it will "avoid" generating violent and hateful content, that it's "cautious" about infringing copyrights, and that it "won't" output images that could deceive others, like "deepfakes intended to mislead."

But as evidenced by all the bizarro images that Grok users are glibly creating, these either aren't hard-and-fast rules or they're not being followed at all.

Whether that's by accident or by design, though, remains the big question.

More on Grok: Elon Musk Is Now Using Your Tweets to Train His Weird AI. Here's How to Turn It Off.


Share This Article