The next flagship AI model from Elon Musk’s AI company, xAI, might be nearing release.
Over the weekend, several users on X, including reverse engineer Alexey Shabanov, reported gaining access to Grok 3, the successor to xAI’s current model, Grok 2, via X’s Grok chatbot app. Before their access was revoked, the users say that they managed to get the model to respond to a range of queries, including logical reasoning and coding-related questions.
According to the users, Grok 3 was able to successfully answer riddles and field requests like “Generate HTML and JavaScript code for a roulette wheel casino.” The model wasn’t perfect, however. In the code for the roulette wheel, Grok 3 missed a few details — and made a programming error.
Breaking 🚨 Grok 3 understands nuanced riddle that ALL of open ai models have failed to get right. The only other model to get this question right was Deepseek R1 🤯
Grok 3 is beating o1? Has @xai caught up this fast ? pic.twitter.com/WLSpM00qHK
— Chris (@chatgpt21) January 26, 2025
Shabanov said that he got Grok 3 to reveal its system prompt, which is the set of instructions that define how the model should behave. One interesting detail is that the prompt appears to explicitly state Donald Trump is the 47th president of the U.S., which could be a hardcoded fix for Grok’s political hallucinations.
Grok 3, which has been in development for several months, was optimistically slated for release last year, according to Musk — but missed that deadline. Early this month, Musk said that Grok 3 had completed pre-training, a crucial stage in a model’s development cycle, and that he expected Grok 3 to arrive in January or early February.
xAI has been using its enormous data center in Memphis — a data center containing around 100,000 GPUs — to train Grok 3. In a post on X, Musk claimed that Grok 3 was trained with “10x” more compute than Grok 2.
Grok, xAI’s answer to models like OpenAI’s GPT-4o and Google’s Gemini, can analyze images and respond to questions, and powers a number of features on X. Grok recently gained standalone apps, and it may soon get a “voice mode” that reads the model’s responses aloud.
When Musk announced Grok roughly two years ago, he pitched the chatbot as edgy, unfiltered, and anti-“woke” — in general, willing to answer controversial questions other AI systems won’t. He delivered on some of that promise. Told to be vulgar, for example, Grok will happily oblige, spewing profanities and colorful language you wouldn’t hear from ChatGPT.
But Grok as it exists today hedges on political subjects and won’t cross certain boundaries. In fact, one study found that Grok leans to the political left on topics like transgender rights, diversity programs, and inequality.
Musk has blamed the behavior on Grok’s training data — public webpages — and pledged to “shift Grok closer to politically neutral.” Evidence of an “Unhinged Mode” for Grok recently emerged, which will provide responses “intended to be objectionable, inappropriate, and offensive,” according to an FAQ page on xAI’s website.
Musk has also said that Grok 3’s training data incorporates filings from court cases, improving its ability to understand legal subjects.
