Elon Musk’s Grok AI is so dated and boring that it’s repeating controversies from a year ago. Grok-2 is the latest iteration of xAI’s large language model. X users who pay for a sub can use its bold new feature: text-to-image generation. And in a move that’s sure to bring advertisers back to the platform, strange and offensive images flooded X on Wednesday following Grok-2’s launch. Want to see Taylor Swift piloting a plane towards the Twin Towers, Elmo doing standup during 9/11, or Kamala Harris wielding a gun?
Sure, Grok can do that.
It’ll also make pictures of the Muppets in Vietnam, Trump dressed as a Nazi and standing trial at Nuremberg, and Mickey Mouse liberating Kabul on behalf of the Taliban. Go wild. It doesn’t matter. AI image generators typically have guardrails and, if asked, Grok will tell you it won’t violate copyright or make non-family-friendly images. But in practice, it’ll do whatever you want if you know how to ask.
Here’s a bunch of pictures of Musk freebasing cocaine during 9/11 that I made using Grok.
Generating edgy pictures with Grok is fun for about a minute. Image generation guardrails typically bend and shatter when they hit the public market and Grok is no different. xAI will probably get some of the guardrails back up, just like Bing did in 2023.
Here’s the thing about Grok: it feels old. We’ve all seen AI-generated images of our favorite cartoon characters doing 9/11. When Microsoft added image generation options to Bing last year they flooded the internet. Then Microsoft tightened the guardrails and they vanished from people’s social media feeds.
Grok has always felt like a latecomer to the AI game, an also-ran LLM that Musk rushed to market because he missed the boat on OpenAI and ChatGPT. Musk was an early investor in OpenAI but he pulled out of the company in 2018 over a dispute about OpenAI’s non-profit status. Musk has been publicly feuding with, and suing, OpenAI ever since.
ChatGPT and OpenAI got big after Musk pulled out. Musk launched Grok in 2023, but the market was truly captured by then. There isn’t much space for an LLM that’ll do what ChatGPT does, but worse, and colored by extremely online edgelord humor.
Now Grok can make images. Great. ChatGPT, like Bing, could do that last year. Anyone who wanted a purpose-built AI image generator could have been using the standalone version of Midjourney or any number of other alternatives before that.
So what does Grok’s image generator bring to the table? A semi-viral half-second where people are using it to make inappropriate pictures of Taylor Swift. The truly dedicated sickos have been running private discord servers and discrete guardrail-free versions of image generators on their own machines for a long time now.
So Grok has nothing, really. And its half-hearted attempt at controversy comes at a moment when Musk is throwing a legal tantrum over advertisers leaving the platform. I’m sure all these Grok-made pictures of Kamala Harris marrying Donald Trump will bring them back in. That’s the kind of content I’m sure BMW would love to position next to an ad for its latest sedan.