xAI’s Grok arrived with a clear pitch: answer questions with information, humor and a rebellious streak. Its early examples leaned into sarcasm, including replies to questions about cocaine and software concepts. The launch put personality at the center of the conversation, alongside questions about what Grok knows and how reliably it can answer.
A chatbot that makes its tone part of the pitch
Elon Musk announced an early beta of Grok on Saturday. The xAI model was presented as an alternative to assistants such as ChatGPT, with a willingness to take on some questions that other systems might refuse because of their content filters.
In an introductory X post, Musk described Grok as having “a little humor in its responses.” One screenshot showed a user asking for step-by-step instructions to make cocaine. Grok responded with a sarcastic list that mentioned a “chemistry degree,” a “DEA license” and coca leaves, then warned the user not to try it.
The joke showed how Grok’s tone could make an answer feel different from a conventional chatbot response. But humor alone does not establish whether a model is accurate, safe or helpful. The launch examples made the voice easy to notice; they did not settle those broader questions.
Building a different kind of assistant
Musk founded xAI in July and staffed it with people who had worked at DeepMind, Google, Microsoft and Tesla. The project had reportedly been taking shape earlier: in April, Musk began purchasing GPUs for a new AI venture.
At the time, Musk criticized conventional assistants such as ChatGPT as too “woke” and said he wanted an alternative that was “based,” a slang term the article describes as roughly meaning authentic to itself. Grok’s public identity reflects that ambition: less buttoned-up, more willing to use sarcasm and informal language.
xAI said Grok-1 was created after two months of training. The model has 33 billion parameters, and the company says its inspiration includes The Hitchhiker’s Guide to the Galaxy and the humor of its author, Douglas Adams. The name “grok” comes from Robert A. Heinlein’s 1961 science fiction novel Stranger in a Strange Land; the word has come to suggest an intuitive understanding of something.
The model’s early persona, however, was more apparent than any claim of deep understanding. xAI described Grok as witty and rebellious, and offered a choice between “regular mode” and “fun mode.” Product features outlined by xAI employee Toby Pohlen included branching conversations, multiple concurrent chats, and ways to open generated code or responses in editors.
Fresh information brings both opportunity and risk
xAI said Grok could draw on information from X, the social platform also owned by Musk. Musk framed that connection as an advantage over language models whose underlying knowledge could lag behind current events. A demonstration showed Grok responding to news about Sam Bankman-Fried, who was convicted of fraud on Thursday.
Access to current posts could help a chatbot discuss events beyond the period covered by its training data. But freshness is not the same as accuracy. The article noted concerns that disinformation on X could affect Grok’s answers, particularly as key moderation team members had left the company since Musk took over last year.
That tension matters for any assistant that uses social platform information. It may be able to respond to newer developments, while still needing to distinguish reliable material from misleading claims. One X user warned that Grok could become a “conspiracy-fueled misinformation hate machine” if it could not filter out nonsense.
Early benchmarks leave room for doubt
xAI reported that Grok performed better than Meta’s Llama 2 and OpenAI’s GPT-3.5 on some exam benchmarks. The same tests placed it below Google’s Palm 2, Anthropic’s Claude 2 and OpenAI’s GPT-4. The comparisons offered an early snapshot, not a complete measure of how useful the chatbot would be in everyday conversations.
The article also cautioned that the value of these tests for measuring large language models has not been proven. Grok was described as a relatively small model trained in a short period, and xAI presented it as an early beta expected to improve with use.
Musk planned to include the assistant in the “X Premium+” service for $16 a month, with access through the X app and a standalone app also planned. For now, Grok’s defining feature was its willingness to answer with a sharp edge. Whether that approach could sit alongside dependable answers remained an open question.