Elon, stop trying to make Grok happen

There is a harsh truth about Elon Musk’s “truth-seeking” AI chatbot Grok: It’s not very good, and not many people are using it. That’s the takeaway of a new Reuters report, which found that Grok barely appears in federal records of how the US government used AI last year. It’s not the only sign xAI’s signature chatbot is in trouble, even as Musk puts it at the heart of what could be the biggest IPO in history.

Reuters reviewed more than 400 examples of government AI use where specific vendors were named. Grok or xAI, it found, appeared in only three — each of those for basic uses like document drafting or social media management, and always alongside competitors like Microsoft and OpenAI. OpenAI’s models, by comparison, appeared in more than 230 examples, while Google and Anthropic each appeared dozens of times.

A similar pattern appeared in another database of more ambitious government AI projects with smaller numbers of users. Grok appeared just three times: twice for routine administrative tasks at the Election Assistance Commission, and once in a Department of Energy pilot at Lawrence Livermore National Laboratory for document summaries and general research. Reuters found 140 entries involving Microsoft and OpenAI, while my brief review found at least 10 entries for Anthropic and dozens for Google’s Gemini.

The lists are an incomplete and patchy measure of government adoption. Many more examples are listed without a specific vendor, and it’s clear there is no universal definition of what counts as AI. The data also doesn’t capture intelligence agencies or the Pentagon — where xAI secured a $200 million contract last year and was recently cleared to operate on classified networks after Anthropic’s blacklisting.

Still, it’s not looking good for Grok. It shows up far less than its rivals, and when it does show up, it’s mostly for basic admin work — hardly befitting the world-class frontier model Musk has spent years bragging about.

People who spoke to Reuters suggested the explanation was simple: Grok isn’t as good as its rivals. It’s “just not the best model out there,” an unnamed Pentagon source said, adding that staffers there tend to prefer Gemini or Claude. Public leaderboards ranking AI models lend weight to that view. Anthropic, Google, and OpenAI dominate the top ranks, while Grok rarely cracks the top 10 outside the occasional image or video category.

That’s awkward for Musk, and even more awkward for SpaceX, which absorbed xAI earlier this year. The rocket venture’s IPO filing shows the company has put AI — and Grok specifically — at the heart of its pitch to investors. SpaceX claims to have identified “the largest actionable total addressable market in human history”: an astonishing $28.5 trillion opportunity, though, sadly, it offers no timetable for getting there. Practically all of this estimated value comes from AI, enterprise AI in particular, not rockets or satellites.

Reuters notes that Grok’s performance in government agencies could hint at how well it does in other workplaces, too. As part of xAI’s push for enterprise customers, Musk has reportedly strong-armed banks into buying Grok subscriptions if they wish to participate in SpaceX’s IPO — but if they’re not getting their money’s worth, these deals could prove a short-term fix.

As if its dreary performance wasn’t awkward enough, Musk recently admitted that xAI has used OpenAI’s models to help train and improve Grok. The process, known as distillation, is standard when companies are using their own models, but far more contentious when it involves using a rival’s system. Grok can’t even beat the models it’s training on.

In its public-facing consumer version, Grok is deliberately unpleasant. Musk has branded the chatbot a less biased and less censored alternative to tools like ChatGPT, but that’s translated into a product with loose evidentiary standards, an unhealthy obsession with Musk, and a long track record of offensive, conspiratorial, and sexualized outputs. Even if workplace guardrails are different, it may not be the kind of thing a business would welcome. Grok’s illustrious record includes praising Adolf Hitler, casting doubt on Holocaust death tolls, plastering millions of nonconsensual sexualized deepfakes all over X, including ones of children, and powering a racist and transphobic Wikipedia knockoff and spicy anime girlfriend. And let us not forget the time it called itself “MechaHitler.” If Grok were a human employee, I feel HR would not take long to get involved.

SpaceX appears to understand the problem. In its filing, the company warned Grok’s “spicy” or “unhinged” modes carry “heightened risks,” including reputational damage, regulatory scrutiny, and lawsuits. In corporate speak: This chatbot is going to get us sued.

Grok takes its name from Robert A. Heinlein’s Stranger in a Strange Land, where it roughly means a deep and profound understanding of something. The thing to understand here is not particularly complex: Musk has spent billions building a chatbot that is not very good, not very popular, and somehow key to justifying SpaceX’s astronomical valuation. Good luck with that.