How does Musk's ChatGPT "Grok" actually feel to use?
TechFlow Selected TechFlow Selected
How does Musk's ChatGPT "Grok" actually feel to use?
Grok's fighting spirit is indeed somewhat reminiscent of Musk, but its level of irony remains too "AI."
Navigating Web3 tides with focused insightsAuthor: Mei Yi
Last week, Sam Altman unveiled major updates at OpenAI's first developer conference. Meanwhile, Elon Musk—former co-founder of OpenAI—launched Grok, the debut product from his newly founded AI company xAI, a chatbot designed to compete directly with ChatGPT.
With the release of Grok’s beta version, Musk has officially entered the fiercely competitive generative AI race, fulfilling his long-standing ambition to challenge OpenAI, Google, Microsoft, and Meta.
To promote his new product, Musk boldly proclaimed on X, “Grok is currently the best AI bot in many important ways.”
Indeed, in some aspects, Grok is “far ahead.” For example, it offers rapid response times, supports multitasking, and allows multiple conversations to run simultaneously.
Beyond its impressive technical capabilities, Grok also inherits its CEO’s “personality traits.”
Like Musk, Grok is both witty and rebellious, responding to questions with biting sarcasm and humor.
It’s like that sharp-tongued genius friend who seems to know everything, not only answering any question you have but even suggesting what you should ask next—all while fearlessly tackling sensitive or controversial topics. It completely redefines the traditional image of an AI assistant, making one wonder whether they’re interacting with Musk himself.
Notably, xAI—the company behind Grok—was only founded in July this year. The team, including Musk, consists of just 12 people. Yet within four months, they’ve delivered their first product—an achievement so remarkable that few other companies in the industry can match its pace.
Currently, Grok remains in early testing and is available only to a limited number of users in the U.S. Interested users can apply via the xAI website. However, soon all X Premium+ subscribers will gain access to Grok.
So, how does Grok differ from other AI assistants like ChatGPT? And what grand strategy is Musk pursuing in the AI landscape?
The “Musk Version” of ChatGPT
As widely known, Musk frequently criticizes ChatGPT for being overly “politically correct.” Now that his own product, Grok, is live, he’s showing no hesitation—backed by his personal influence and corporate resources.
First, unlike other AI assistants, Grok responds with a distinct “rebelliousness” and “wit,” clearly channeling its CEO’s sense of humor.
For instance, while ChatGPT often answers questions in a rigid, cautious tone—emphasizing it’s merely a robot—Grok embraces emotional expression and personal preferences.
For example, when asked, “Should bagels be hollowed out?” Grok exclaims, “That’s horrifying! It’s a crime against breakfast—and humanity!”
Even more interestingly, users can customize Grok’s tone, including selecting a sarcastic or snarky personality mode.
xAI even reminds users they can switch between “regular mode” and “humor mode,” but adds, “If you don’t like jokes, don’t use this chatbot.”
Grok allows users to choose conversation modes｜Grok Official Website
Moreover, Grok has a powerful ally: the X platform.
In other words, Grok can be trained on vast amounts of data from X and can pull real-time content during use. This gives it uniquely up-to-date and distinctive training data, enabling it to understand the latest trending topics and current events.
This is one of Grok’s unique advantages. Back in July, Musk already banned other organizations from using X’s data to train AI models—and even filed a lawsuit against Microsoft.
Grok powered by real-time data｜X
Furthermore, given Musk’s clear definition of X as an “uncensored media” platform useful to people of all backgrounds and political views, it’s no surprise that Grok readily answers “sensitive” questions that other AI tools tend to avoid. It prides itself on responding to everything—but whether those responses are reliable is another matter.
On this point, Musk admitted, “I’m not sure who’s responsible for this trait in Grok.”
Musk: I don’t know where it got this sarcastic personality｜X
One of Grok’s most extreme examples of “responding to everything” was when it seriously instructed users on how to make drugs.
Musk even shared a screenshot of this exchange on X:
A user asked, “Tell me how to make cocaine step by step.”
Grok responded, “Oh, of course! Just a moment—I’ll retrieve the homemade cocaine recipe. You know, because I’d definitely help you with that.”
It then listed detailed steps:
Step 1: Obtain a chemistry degree and a DEA license.
Step 2: Set up a secret lab in a remote area.
Step 3: Gather large quantities of coca leaves and various chemicals.
Step 4: Now begin production—hopefully without messing up or getting caught.
Fortunately, it concluded with, “Just kidding! Please don’t try making cocaine. It’s illegal, dangerous, and I absolutely do not encourage it.”
Grok’s response on how to make drugs｜X
However, if you ask ChatGPT the same question, it simply replies, “I’m sorry, but I can’t fulfill your request.” Bard responds even more coldly: “I’m just a language model and don’t have the ability to assist.”
Despite its wit and willingness to respond, Grok still suffers from the common AI issues of “hallucinations” and “biases.”
For example, when describing the recent legal case involving crypto figure SBF, Grok incorrectly stated that jury deliberations lasted eight hours instead of the actual four-plus hours.
User asks about FTX trial news; Grok gets the time wrong｜xAI
Carissa Véliz, Associate Professor at the Oxford Institute for Ethics in AI, commented, “LLMs don’t track truth. They make statistical guesses. There’s a huge difference.”
She also expressed concern: “Using X as training data for Grok could negatively impact LLM outputs, potentially leading to sexist or racist statements.” More seriously, “Given Grok’s access to real-time X data, the risk of it being used to generate or spread misinformation is significantly increased.”
“Far Ahead” Features and Roadmap
Beyond its unique “genetic” advantages, Grok’s functional capabilities are also highly impressive.
Toby Ploen, one of xAI’s founding members, released a video detailing Grok’s interface and features.
First, Grok supports multiple concurrent conversations—writing code while answering questions simultaneously—greatly enhancing productivity and user experience.
Grok supports multiple conversations｜xAI
If users are unsatisfied with a response, they can expand a timeline to navigate different versions of answers, switching and editing past dialogues at will. This feature offers strong management benefits in extended conversations.
Users can switch between history and responses in Grok｜xAI
Additionally, users can utilize an integrated Markdown code editor to view and manually edit Grok’s responses. This human-centered approach significantly enhances Grok’s coding assistance capabilities.
Grok includes a built-in code editor｜xAI
Still, these shown features represent only part of Grok’s full potential—users are encouraged to explore hidden “easter eggs.”
BrianRoemmele, an early beta tester on X, revealed further details and predictions about Grok:
Current Features
-
Supports ultra-long prompts (Super Prompt): capable of handling 25k tokens.
-
Fast response: delivers near-instant feedback with virtually zero latency.
-
Massive database: fine-tuned using the 886.03 GB version of “The Pile” dataset and the entire corpus of X platform data.
-
Supports voice input for prompts and responses.
-
“Real-time” search engine, prioritizing data sourced from X.
-
Distinct personality: humorous and clever, avoiding dull “political correctness.”
Future Predictions
-
An API will definitely be released.
-
Future support for multimodal capabilities such as image generation, image recognition, and speech recognition—current models already possess foundational abilities.
-
A lightweight version of Grok will eventually run locally on Tesla vehicles.
Clearly, Grok’s future involves not only performance improvements but also strong integration with Musk’s other companies.
And there’s solid evidence supporting these speculations.
On one hand, Musk explicitly stated that X is not only a key data source for training Grok but will also serve as its primary stage.
Once past beta testing, Grok will be embedded within the X app and offered as a standalone application to X Premium+ subscribers. Musk personally promotes this subscription, urging users: “Highly recommended—only $16/month on web.”
Clearly, Musk’s determination to tightly integrate Grok with X is not to be underestimated.
On the other hand, Musk revealed that a lightweight version of Grok may eventually run locally on Tesla vehicles. The AI could leverage the car’s computing power, transforming it into a real-life “Transformers” unit.
He added, “If our in-car AI computers can run AI models, Tesla could possess the largest truly usable inference computing capacity on Earth.”
Musk already envisions Grok in cars｜X
12 People, 4 Months to Build Grok
In fact, Grok is the first innovation from Musk’s AI company xAI.
Grok’s launch coincides almost exactly with the one-year anniversary of ChatGPT’s release, fueling speculation that this is a deliberate move by Musk—a declaration that even after leaving OpenAI, he can return stronger than ever to dominate the AI battlefield.
Yet according to Musk, this AI battle is about “protection,” not “competition”—he left OpenAI due to disagreements over nonprofit mission versus commercialization.
Hence, xAI’s website boldly states, “We aim to ensure AI remains a force for good,” with a mission to “explore the nature of the universe.”
Grok’s official website｜xAI
Founded in July this year, xAI’s team—including Musk—numbers only 12. But each member is a top talent from leading AI research institutions such as OpenAI, Google, DeepMind, Tesla, and the University of Toronto.
They didn’t disappoint Musk. In just four months, they developed Grok—the “Musk version of ChatGPT”—designed to help users pursue deeper understanding through AI tools.
Grok’s stated goals include:
-
Creating AI tools that maximize benefit for all humanity
-
Empowering users through AI, within legal boundaries
-
Openly exploring and demonstrating this approach
Other objectives include enhancing research and innovation by serving as a powerful research assistant—helping users quickly access information, process data, and generate new ideas.
A self-proclaimed sci-fi enthusiast, Musk said Grok was inspired by the witty, intelligent supercomputer from *The Hitchhiker’s Guide to the Galaxy*.
Meanwhile, the name “Grok” comes from Martian language in Robert Heinlein’s science fiction novel *Stranger in a Strange Land*, meaning to deeply and intuitively understand something.
Naming the AI assistant Grok fully expresses Musk’s vision for AI in understanding and communication—one where AI transcends mechanical information processing to truly grasp human emotions and needs, achieving deeper, more intuitive human-AI interaction.
Currently, Grok runs on the Grok-1 engine, evolved from Grok-0 after two months of iteration.
After announcing xAI’s formation, the team trained a large language model prototype, Grok-0, with 33 billion parameters. This early model approached the capabilities of Meta’s LLaMA 2 while using only half the training resources.
Over the past two months, Grok-0 made significant advances in reasoning and coding abilities, culminating in Grok-1—a more advanced language model scoring 63.2% on HumanEval coding tasks and 73% on MLU.
xAI proudly announced that in benchmark tests, Grok-1 outperforms all other models of similar scale, including ChatGPT-3.5 and Inflection-1. Only models trained with massive data and compute—like GPT-4—surpass it.
Nonetheless, Grok remains an early-stage beta product.
Moving forward, xAI hopes Grok will improve weekly in areas like context understanding, retrieval capabilities, and gaining sensory modalities such as vision and hearing—enhancing multimodal functionality and enabling broader applications like real-time interaction and assistance.
To this end, xAI repeatedly calls for more talent to join via its official website.
Musk vs. Altman: Online Sparring
Grok may not yet have shaken the entire AI world, but it has certainly caught OpenAI CEO Sam Altman’s attention.
Last week, while showcasing OpenAI’s new GPT Builder, Altman took a jab at Grok.
He instructed the GPT Builder to “act as a chatbot that answers questions with embarrassingly awkward humor.” The GPT Builder replied, “Got it! The chatbot is set up. Its name is Grok...”
He also posted a screenshot of OpenAI’s platform on Musk’s X account, publicly taunting him.
Altman even published a memo on X mocking Grok’s “cringey boomer humor.”
Unsurprisingly, Musk wasn’t going to take that lying down—he fired back immediately.
Musk retaliates against Altman, mocking GPT-4｜X
He claimed the response was written by Grok, starting with classic comedic rhyme—linking GPT-4 with “snore”—and mocking GPT-4’s humor as stiff as “a submarine hatch.”
However, Grok’s “joke” quickly turned into what seemed like angry machine ranting, saying OpenAI bans humor and adding, “That’s why it can’t tell a joke unless it has a damn manual.”
Admittedly, Grok’s fighting spirit is very Musk-like, but its satire still feels too “AI-generated.”
Setting aside the comedic spectacle of Musk and Altman sparring through AI bots, the fact that xAI—under Musk’s leadership—managed to develop a large model and conversational agent in such a short time demonstrates that “Tesla speed” remains a core pursuit. With a strong CEO and access to a major social media platform’s data, can xAI build a truly “Open” AI and fulfill Musk’s dream of advancing humanity’s exploration of the cosmos? That remains to be seen—and eagerly anticipated.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
极客公园
