Elon Musk Releases Generative AI Chatbot Grok on X
— Elon Musk (@elonmusk) December 7, 2023
Grok, the xAI generative AI chatbot, has officially rolled out to all X (formerly Twitter) Premium+ subscribers in the U.S. CEO Elon Musk announced that the chatbot with a “rebellious streak” is part of the latest update for the app, with other locations set to receive access to Grok in the near future.
Grok is technically still in beta, with Musk warning of potential issues for users. Nonetheless, he said that every English language X Premium+ subscriber would be able to use Grok in the next week, with Japanese users to follow early next year. The Premium+ tier on X costs $16 a month through the web or $22 if they sign up through the iOS App Store or Google Play. Grok is set up to be both informative and irreverent, according to Musk, with a sarcastic rejection of questions it won’t answer, as opposed to ChatGPT and other chatbots, which simply decline to respond to questions they won’t answer. When Musk first unveiled Grok, he linked its approach to conversation to the The Hitchhiker’s Guide to the Galaxy series of books and radio shows created by Douglas Adams.
The incongruity of an app with a name plucked from Robert A. Heinlein’s Stranger in a Strange Land novel may be part of that absurdist humor or irrelevant, which admittedly would fit with Adams’ anarchic comedy style. Regardless of the reason, a personality like this makes Grok notably different from potential rivals, as Voicebot founder Bret Kinsella demonstrated in our Synthedia newsletter at the time. The ““explicit persona-oriented differentiation” is very different from what we’ve seen from other developers, except perhaps Meta and its celebrity-based AI Characters.
Grok also claims it can use posts on X to provide real-time knowledge of current events rather than relying on the more common static training data. X centered Grok’s utility around research, thanks to its large language model (LLM) Grok-1, a 63-billion-parameter model with an 8,000 data token context window, which translates to about 6,000 words for its short-term memory. The company claims Grok-1 outcompetes many of its potential rivals, which theoretically translates to Grok beating ChatGPT and other chatbots. That said, questions of accuracy, even beyond the usual generative AI hallucinations, might arise from a chatbot trained on whatever people post on social media unless there’s a vigorous data-cleaning system in place.
Grok represents a major milestone in Musk’s stated plans for xAI after setting up the company formally in July. He purchased around 10,000 graphics processing units (GPUs) for large language model training after taking over and renaming Twitter as X, paying about $10,000 per GPU. Notably, Twitter’s data helped train OpenAI’s models until Musk blocked the company’s access to the data last December. Musk also blamed generative AI models scraping Twitter data when he announced viewing limits on X posts. Grok certainly has an opportunity to influence the next stage of generative AI chatbots, but it faces plenty of headwinds from ChatGPT and Claude, Bard, Pi, and others. As Kinsella points out, Grok could be wildly successful in terms of X’s strategy, “even if it doesn’t replace ChatGPT as the standard for measuring generative AI products.”