Grok: What We Know About Elon Musk’s AI Chatbot

Grok is an AI chatbot developed by Elon Musk’s company xAI. Grok is able to generate text and engage in conversations with users, similar to ChatGPT and other tools. Unlike other chatbots, though, it can access information in real-time through X (formerly Twitter) and is programmed to respond to edgy and provocative questions with witty and “rebellious” answers.

Grok is essentially Musk’s answer to ChatGPT, whose maker (OpenAI) he co-founded in 2015 but left in 2018 after a reported power struggle with now-CEO Sam Altman. Musk has since condemned ChatGPT for being too left-leaning and dangerous. According to Musk, xAI is intended to be a direct competitor to OpenAI, with its Grok chatbot not only serving as ChatGPT’s “anti-woke” counterpart, but also showcasing new possibilities in the larger generative AI space.

 

What Is Grok?

Grok is an artificial intelligence chatbot developed by xAI. Released in November of 2023, it is now available to users with a Premium+ subscription to the social media platform X. 

Grok-1, the large language model that powers Grok, was trained using a custom tech stack based on software management system Kubernetes, machine learning framework JAX and coding language Rust, all of which helped xAI to develop Grok faster and more efficiently than other chatbots. 

Like all LLMs, Grok-1 was trained on massive amounts of text data scraped from the internet, which includes everything from Wikipedia articles to scientific papers. But what makes Grok different is its direct access to posts made on X. This enables Grok to have “real-time knowledge of the world,” according to the company, which gives it a “massive advantage over other models,” as Musk put it.

Grok offers two interaction styles: “Fun Mode” and “Regular Mode.” By default, Grok operates in “Fun Mode,” which causes the chatbot to take on a more edgy or humorous personality and, at times, produce factually incorrect responses. “Regular Mode” typically delivers more accurate answers, but as with all AI chatbots, xAI says it can still generate false or contradictory information.

Grok’s sense of humor and “personality” was modeled after The Hitchhiker’s Guide to the Galaxy by Douglas Adams, one of Musk’s favorite books. 

“It’s a book on philosophy, disguised as a book on humor,” Musk said in an interview with computer scientist and podcaster Lex Fridman. With Grok, xAI is working to maintain that ethos while also adhering to “the truth of the universe,” Musk explained, and eventually discovering new truths — approaching something closer to artificial general intelligence, where a machine can learn and think like (or even better than) a human.

More on AI and the FutureWhat Is Sentient AI?

 

Why Is It Called Grok?

Grok’s name is believed to have originated from Robert A. Heinlein’s 1961 science fiction novel Stranger in a Strange Land, in which the story’s main character, a Martian, uses the term “grok” to convey a profound and intuitive understanding of something.

While Musk has never officially confirmed the meaning of Grok’s name, he posted “Stranger in a Strange Land” on X the day after Grok was announced, likely referencing Heinlein’s book. By adopting this word, xAI appears to envision Grok as more than just another chatbot, but a tool to “assist humanity in its quest for understanding and knowledge,” according to its website.

 

What Can Grok Do?

Grok can draft emails, debug code, generate ideas and more — and all in fluent, human-like language. It simply receives an input (like a command or question), applies knowledge from its training data, and uses sophisticated neural networks to generate a relevant text output.

While it is used in the same ways as other AI chatbots, “Grok will probably say ‘yes’ to a lot more jobs that you give it,” said Sharon Gai, an author and speaker who focuses on the AI industry.

Indeed, xAI says Grok is willing to answer questions that most other chatbots would refuse, no matter how taboo or potentially harmful they may be. For example, Musk shared a screenshot of Grok offering a step-by-step guide to making cocaine for “educational purposes,” which included instructions like “start cooking and hope you don’t blow yourself up or get arrested.” He shared another screenshot of Grok offering advice for what to do if you get an STD in increasingly “vulgar” ways. 

“It’s touted as a little bit like Musk himself, in that it’s supposed to be kind of tongue in cheek,” Lance Whitney, freelance tech journalist who has covered Grok and other AI chatbots extensively, told Built In. It is designed to be more of a “playful and fun chatbot,” he added, “where you can go to have a more offbeat or snarky conversation.”

From a user interface standpoint, Grok can also handle multiple queries simultaneously and users can toggle between those answers, as shown in a video demonstration by xAI co-founder Toby Pohlen. Code generations can be opened directly in a Visual Studio Code editor, while text responses can be saved in a markdown editor for later use.

Yet Another Musk ProjectWhat Is Neuralink? What We Know So Far.

 

Grok vs. ChatGPT: How Are They Different?

While both Grok and ChatGPT share the goal of facilitating human-like interactions through artificial intelligence, they approach it in fairly different ways, offering their own unique strengths and weaknesses. Here are a few:

 

1. Grok has Access to Real-Time Information, ChatGPT Does Not

Grok has direct, real-time access to posts on X, whereas ChatGPT’s free version only knows information up to January of 2022, and its paid version only knows information up to April of 2023. This means Grok can engage in conversations about more recent events, such as the Israel-Hamas war or the 2024 Super Bowl. In fact, depending on the question asked, Grok will actually display real posts on X that it is referencing in order to show where its point of view is coming from.

However, a Vice investigation found that Grok tends to spout inaccuracies about current events and lend credence to unproven conspiracy theories — due largely to X’s propensity for disinformation and even hate speech since Musk’s purchase of the site in 2022. 

 

2. Grok Is Less Politically Correct Than ChatGPT

In Musk’s words, Grok is “maximum truth-seeking” and “based,” meaning it is unapologetic and communicates without regard for political correctness.  

“[Musk] believes that free speech should be allowed to every and any degree,” Gai said. “That is what Grok is modeled after as well.”

XAI’s creation of a less politically correct chatbot comes at a time when most other big AI companies are working to make their own chatbots even more PC. OpenAI claims its new GPT-4 LLM, which powers ChatGPT’s paid version, is 82 percent less likely to respond to requests for “disallowed content,” which includes material that is “hateful, harassing” and “violent.” And Anthropic’s Claude chatbot was trained using constitutional AI, which helps to reduce the likelihood of it generating toxic, dangerous or unethical responses.

“They have entirely different purposes.”

Because ChatGPT and Grok approach truth and safety in such different ways, “they have entirely different purposes,” Whitney said. Grok is explicitly designed to answer questions in a non-PC way, so “it’s not necessarily a chatbot I would go to for research,” he continued. “I would sooner go to ChatGPT.”

At the same time, though, ChatGPT is much more limited in the subjects it is willing to discuss with users, so it’s not always as useful as Grok. For example, Gai said she tried to use ChatGPT to help summarize a text involving suicide, and it outright refused to do the job. “It’s not like I was investigating into suicide for myself or anybody else, it was just refusing to even touch that job,” she said. “But that is something Grok would not refuse.”

Grok could also be useful in understanding the “zeitgeist,” Whitney said, because it has direct access to social media posts. “If I wanted to get a sense of what people are thinking about a certain topic, what they’re discussing and how they feel about it, I would go to Grok.”

 

3. Grok Scores Better Than ChatGPT on Exams

Grok-1 scored better on benchmark exams in middle school math, high school math, code completion and language understanding than GPT-3.5, the LLM that powers ChatGPT’s free version. It also earned a C-grade on the Hungarian national high school finals in mathematics — a test the model was not explicitly tuned for  — outperforming GPT-3.5 once again. 

It’s worth noting, however, that Grok-1 still didn’t score as well as GPT-4 on any of these exams. But xAI says that is because GPT-4 is trained on a “significantly larger amount of training data and compute resources.” 

“Grok-1 displayed strong results, surpassing all other models in its compute class,” the company said. “This showcases the rapid progress we are making at xAI in training LLMs with exceptional efficiency.”

 

4. ChatGPT Can Process Images, Grok Cannot

GPT-4 is multimodal, meaning it can handle both text and image inputs. For example, it can suggest recipes from a photo of an open refrigerator, or make predictions based on what is happening in a corresponding picture.

For now, Grok can only accept text as inputs. But xAI says it will be equipped with visual and audio capabilities in the future.

 

5. ChatGPT Is Free, Grok Costs $16 a Month

The basic version of ChatGPT is completely free to use. A premium version is also available for $20 a month. There is no free version of Grok at this time; it is only available to people who pay $16 a month for a premium subscription to X.

 

6. Grok Has an Open Source Version, ChatGPT Does Not

In March 2024, xAI released the network architecture and base model weights of its large language model Grok-1 under the Apache 2.0 open source license, which allows other developers to use and build on the model — including for commercial purposes. The open source version is from the pre-training stage of development, meaning users will likely have to fine-tune the model on their own before putting it to work. 

 

Frequently Asked Questions