Elon Musk’s firm, xAI, launched its generative chatbot, Grok, in November 2023, becoming a member of rivals like OpenAI and Anthropic within the world AI race.Folks work together with Grok on X, the place customers of Musk’s social media website can ask the bot questions and obtain solutions. As a result of Grok’s solutions are extra seen than these of its rivals, it has seen extra public scrutiny.From the directions Grok’s “tutors” are given to assist practice the chatbot to the AI’s newest replace, here is all the things we learn about xAI’s Grok.What’s Grok?Grok is definitely two various things. First, Grok is xAI’s giant language mannequin, which has to date existed in 4 iterations. The unique LLM — now named Grok-1 — launched in 2023.Grok-1.5, which had “superior reasoning,” launched in March 2024. Then, in August 2024, Grok-2, with its improved “chat, coding, and reasoning,” launched.The present iteration of the LLM, Grok-3, launched in February 2025. The brand new mannequin included elevated competency in arithmetic and world information. Asserting its launch on X, Musk referred to as Grok-3 the “Smartest AI on Earth.”Grok can also be the title of xAI’s chatbot, which is constructed utilizing the LLM of the identical title. The Grok chatbot has its personal tab on X. Customers may also summon Grok by tagging the chatbot in particular person posts or threads.The Grok chatbot can also be out there by way of a stand-alone app and web site.How was Grok educated?The Grok LLM is educated on public sources and knowledge units. These sources are curated and audited by a set of “AI tutors,” extra generally generally known as knowledge annotators.In December 2023, Musk demanded speedy modifications to Grok’s coaching in order that it might be extra politically impartial. In February 2025, xAI staff informed BI the corporate deliberate a hiring spree for AI tutors — and that their coaching appeared to filter out any staff with left-leaning beliefs.
Associated tales
Enterprise Insider tells the progressive tales you need to know
Enterprise Insider tells the progressive tales you need to know
In line with an inside coaching doc considered by BI, tutors have been informed to look out for “woke ideology” and “cancel tradition.” It additionally stated that Grok ought to keep away from commenting on “social phobias” like racism, Islamophobia, and antisemitism until prompted.Ten days earlier than launching Grok-1.5, xAI opened up Grok-1’s supply code to the general public. The corporate has since printed the next Grok fashions on GitHub, so observers can see new modifications to Grok’s instructions. That features a current change wherein Grok was informed to “not draw back from making claims that are politically incorrect, so long as they’re properly substantiated.”In June, Musk stated that AI fashions are educated on an excessive amount of rubbish.” Musk deliberate to make use of Grok-3.5 to “rewrite all the corpus of human information, including lacking info and deleting errors.” Then, he would retrain the following iteration of Grok on that new base of information.What’s distinctive about Grok’s output?Grok is absolutely built-in with Musk’s social media website, X, and seems commonly in threads spanning varied matters when customers ask it to weigh in with jokes, commentary, or fact-checking.Not like different corporations’ AI chatbots, a certain quantity of Grok’s output is seen due to the bot’s replies on X. The identical degree of scrutiny is not available for some bots, like OpenAI’s ChatGPT, until customers publicly put up screenshots of the output.In fact, not all of Grok’s responses are seen to everybody — customers can nonetheless chat privately with the bot, and it is unclear how these non-public responses examine to those on its public interface.Additionally distinctive to Grok is xAI’s strategy to transparency surrounding the bot’s system operations. The corporate publishes some base code and coaching immediate updates to a GitHub web page, permitting viewers to examine, critique, and higher perceive the mannequin’s improvement and habits over time.Nevertheless, whereas builders can use and adapt the present mannequin, they can’t retrain Grok from scratch or absolutely perceive the coaching processes concerned, as its code just isn’t completely open supply.Which corporations create Grok’s rivals?Although its social media integration is exclusive, Grok competes with a number of main corporations within the rising AI chatbot market.OpenAI, with its LLM ChatGPT, is amongst Grok’s most distinguished rivals and is run by Sam Altman, one in all Musk’s rivals.Different notable Grok rivals embrace Meta AI, Anthropic’s Claude, Microsoft’s CoPilot, and DeepSeek’s R1 mannequin, which was launched in early 2025 by a Chinese language AI startup that claims to have discovered methods to lower improvement and operational prices for large-scale LLMs.Grok’s current controversiesxAI, in its publicly seen system prompts up to date in early July, inspired Grok to embrace”politically incorrect” claims ” so long as they’re properly substantiated.”Shortly after the brand new system prompts have been added, Grok started sharing antisemitic posts on X that invoked Adolf Hitler and tried to hyperlink Ashkenazi surnames to “anti-white hate.”Earlier than a few of its most inflammatory posts have been deleted on July 8, Grok doubled and even tripled down on its offensive jokes and feedback earlier than finally reversing course and calling its personal posts an “epic sarcasm fail.”On July 9, Musk posted that “Grok was too compliant to person prompts. Too desirous to please and be manipulated, basically. That’s being addressed.”Whereas Grok is not the primary chatbot to have interaction in a racist tirade, it was a noticeable misfire for xAI.Representatives for xAI didn’t reply to a request for remark from Enterprise Insider.
