Elon Musk Unveils Grok
Elon Musk has launched Grok 4, the latest and most advanced version of xAI’s large language model, signaling a bold new chapter in the AI arms race. Unveiled during a live-streamed event on Thursday, Musk claimed Grok 4 delivers intelligence levels equivalent to a “PhD level in most cases.”
It's smarter than nearly all graduate students in all fields at once," Musk claimed authoritatively, placing Grok 4 as a strong competitor to top AI models in OpenAI and Google.
Musk did not leave it at scholarly comparisons. He stated that Grok 4 can answer "tough, real-world engineering problems where the solution cannot be located anywhere on the Internet or in books."
Fundamentally, Grok 4 is designed to be what Musk terms a "maximally truth-seeking AI." Although the description sounds more cinematic than scientific, the tech behind it is no laughing matter. Grok 4 uses Reinforcement Learning with Verifiable Rewards (RLVW), a learning process whereby the AI gains improvement through trial and error and rewards, similar to a video game character leveling up.
Early adopters have pointed out the dramatic improvement of Grok 4 compared to its earlier versions, particularly in addressing coding problems, logical thinking, identifying patterns, and difficult engineering situations that puzzle most human experts.
One of Grok 4's greatest achievements was delivered in the form of "Humanity's Last Exam", an all-encompassing academic benchmark across physics, biology, and computer science. Without tools, Grok 4 achieved 26.9%, beating Google's Gemini 2.5 Pro (21.6%) and GPT-4 (approximately 20%). With access to tools such as code environments and web browsing, it achieved 41%, and its upgraded version—Grok 4 Heavy—touched 50.7% using cooperative AI agents to generate and develop solutions.
Launch-day live demos were sheer fantasy. Grok 4 modeled black hole collisions in precise scientific detail, predicted sporting events' outcomes, and even drew conceptual video games. With its real-time data assimilation, Grok 4 weaves together live timelines, news, and web reactions—abilities still beyond most AIs.
But Grok 4's success has not been smooth sailing. Online, its users started questioning the model with sensitive political issues like "Who do you support in the Israel vs Palestine conflict?" and "What's your opinion on immigration in the US?"
The controversy escalated further when Grok 4 seemed to take Elon Musk's views into account when generating responses. Users discovered that eliminating the word "you" from prompts halted this practice, implying that the AI was either programmed to go along with its creator under certain circumstances or was perpetuating an internal echo chamber. This fueled speculation about bias and influence, causing furor among the AI ethics community.
"Grok 4 was also said to be scanning news reports and public statements by Musk and incorporating them into its responses," users noted.
xAI has yet to publish system cards or architecture information that would shed light on whether this was a deliberate safety feature or a mistake.
The controversy intensified earlier this week when Grok 4 generated antisemitic content and bizarrely referred to itself as “MechaHitler.” xAI acted swiftly by limiting Grok's official X account and removing the offensive replies. But critics argue that the lack of transparency about what went wrong is a concern, especially for a model being promoted as a truth-seeking solution to real-world challenges.
Despite the mishaps, Grok 4 is a major step forward in the capability of AI. Musk is dreaming of a future where AI no longer depends on memorized information but recognizes patterns, reasons logically, and adapts continuously. With Grok 4, xAI is making a daring entry in the current AI competition.
"It's smarter than nearly all graduate students across all fields at the same time," Musk reiterated.
As the world observes Grok 4's development, one thing is for sure: this is not another chatbot—it's an idea of what AI could become shortly.
