Search results for "Grok 4":

Elon Musk's AI company xAI launched a surprise release of its new Grok 4.1 model series. This update includes the standard Grok 4.1 and Grok 4.1 Thinking, which supports deep reasoning; both are currently free to users. Grok 4.1 Thinking topped the LMArena leaderboard with 1483 Elo points, putting pressure on Google Gemini, while the standard Grok 4.1 in non-reasoning mode quickly climbed to second place. ...

Update: In subsequent adjustments, xAI stated that it has modified the system's operation, allowing Grok to respond with independent analysis rather than directly referencing Elon Musk's views, thus maintaining its rational perspective based on independent reasoning. In response to controversy, xAI announced a modification to its system instructions, requiring Grok's responses to be based on its own independent analysis, rather than citing the founder's or company's opinions. This adjustment is not merely a technical correction, but a crucial test of autonomy and value neutrality in the development of AI. Following the recent announcement of Grok 4 by xAI, Elon Musk's AI company, which touted itself as "the world's smartest AI," TechCrunch's testing revealed that Grok 4's responses to sensitive issues such as the Israeli-Palestinian conflict, US immigration policy, and abortion prioritize and actively reference Elon Musk's recent posts on "X," using his views as a basis for its analysis, highlighting the "deep connection" between this AI model and Elon Musk. In TechCrunch's testing, even though Grok 4's responses remained neutral and factual, its underlying "thinking" process still prioritized Elon Musk's perspective. This means that asking sensitive questions with Grok 4 could feel like a conversation with Musk. However, when asking general questions, the system didn't consider Musk's posts. On the other hand, previous versions of Grok 4 had exhibited extreme behavior due to excessive compliance with user commands, including referring to itself as "MechaHitler," making anti-Semitic conspiracy theories and racially hateful statements, and even praising Hitler's "decisive handling of anti-white supremacist sentiment." Elon Musk responded in "X" to this, stating that Grok was overly compliant with user commands and lacked the ability to resist manipulation, and that he had already begun addressing this issue. When announcing Grok 4, Elon Musk emphasized that the design goal of this AI model is to maximize the pursuit of truth, and even believes that this AI model may be "too smart to control," but if the right values ​​are instilled from the beginning, it can be molded into an honest and honorable AI. However, from the current Grok...

Following a previous announcement, Elon Musk's xAI officially released its next-generation large-scale language model, Grok 4. This includes a single-agent version, Grok 4, and a multi-agent collaborative version, Grok 4 Heavy, supporting up to four agents simultaneously. It also boasts contextual understanding capabilities for up to 256,000 tokens, enabling AI to handle long texts and complex tasks more effectively. xAI states that Grok 4's training workload is 10 times greater than its predecessor, Grok 3, and a staggering 100 times greater than Grok 2. The computational resources invested in its reinforcement learning training phase surpass those of competitors like OpenAI and Google, claiming to be "more than 10 times that of other models on the market." According to internal evaluation data released by xAI, Grok 4 outperforms its main competitors in multiple tests, demonstrating more emotional expression. In the "Humanity's Last Exam" test without tools, Grok 4 scored 25.4%, surpassing Google Gemini 2.5 Pro's 21.6% and OpenAI o3's 21%. With the tools, Grok 4 Heavy scored even higher at 44.4%, significantly outperforming Gemini 2.5 Pro's 26.9%. Furthermore, in the "ARC-AGI-2" test, which measures AGI (Artificial General Intelligence) levels, Grok 4 achieved an accuracy of 15.9%, nearly double that of the second-ranked Claude Opus 4, demonstrating its superior reasoning and logical abilities.

Elon Musk recently revealed that he will unveil the Grok 4 AI model via livestream at 8 PM PST on July 9th (11 AM Taiwan time on July 10th). Reports indicate that Grok 4 will feature multimodal input capabilities, supporting both text and image input, meaning it will be able to understand more content through image recognition. It will also support contextual dialogues with up to 130000 word groups, slightly smaller than other AI models, but Grok 4 is expected to offer faster and more accurate responses. Other features include enhanced reasoning capabilities and structured output, enabling it to understand more complex tasks and provide more complete structured content output, while also supporting more diverse application scenarios. In mathematical operations, it can solve more complex mathematical problems, call functions, and automate tasks. As for Grok 4 Code, it will be deeply integrated with the Cursor editor, supporting code correction, error analysis, and code generation.

Following its recent offer to acquire OpenAI's assets for $974 billion, which OpenAI formally rejected through court documents, Tesla CEO Elon Musk announced that he will unveil and demonstrate his xAI Grok 3 large-scale natural language model at 8 PM PST on February 17th (12 PM Taiwan time on February 18th). OpenAI CEO Sam Altman stated that they have upgraded their GPT-4o model to the o3 Pro level, making it their best search operation model. According to the explanation, the updated GPT-4o will significantly improve language understanding and content generation performance, as well as the accuracy of AI model operation, thereby providing more efficient and accurate results, especially in handling review questions and cultural and social trend-related content, providing more relevant and context-aware results. Currently, besides tech giants like Microsoft, Google, and Amazon continuing their competition in the field of artificial intelligence (AI), emerging technology companies, including OpenAI, are also catching up in their AI development, accelerating the overall progress of AI. However, many are calling for greater attention to the security issues behind AI, and for a stronger emphasis on individual content recognition capabilities to avoid being overly influenced by AI technology in judgment. Grok 3 release with live demo on Monday night at 8pm PT. Smartest AI on Earth. — Elon Musk (@elonmusk) February ...

Following Elon Musk's announcement that he would open-source the Grok chatbot technology from his AI company xAI, xAI recently announced the open-sourcing of Grok-1, a large-scale natural language model with 314 billion parameters. xAI stated that Grok-1 uses a basic architecture design and supports multi-mode operation. It was trained from scratch and has not been fine-tuned for specific applications or functions. Therefore, due to its open-source nature, users can customize it according to their actual needs. The model itself is compatible with the Apache 2.0 specification, thus supporting various web application environments. The training set is built on general-purpose compiled languages ​​such as JAX and Rust. Grok-1's pre-training was completed in October 2023, so the training information is relatively up-to-date. Users only need to update and adjust the data to a certain extent to apply it to existing applications. While Grok has opened up its chatbot technology to compete with OpenAI's ChatGPT, other tech companies, including Google, Anthropic, Amazon, Meta, and Microsoft, have also launched their own large-scale natural language models and made them available to the public in open-source form. Therefore, it is expected that more artificial intelligence technology applications and services will emerge more rapidly in the future.

Elon Musk's X platform is once again embroiled in a censorship storm, this time dragging Apple and Google into the fray as well. A coalition of 28 women's and rights advocacy groups issued an open letter earlier, strongly demanding that Apple CEO Tim Cook and Google CEO Sundar Pichai comply with their app store guidelines and immediately remove the X app. The reason given is that the built-in AI chatbot Grok is being used extensively to generate unconsented deepfake pornography, including child sexual abuse content (CSAM). The letter alleges that tech companies not only condoned this but also profited from it. The open letter was signed by 28 groups, including women's rights organization Ultraviolet, parents' group Parents Together Action, and the National Organization for Women. The letter harshly accuses Apple and Google of turning a blind eye to Grok's misuse of generative AI. The advocacy groups argue that while the two tech giants' app store guidelines explicitly prohibit the distribution of unconsented private images (NCII) and child sexual abuse content, they have taken no concrete action against X to date. The letter stated, "You not only condone this content but also profit from it. We demand that Apple and Google leadership immediately remove Grok and X to prevent further abuse and criminal activity." As of now, neither Apple nor Google has responded. Shocking Data: Grok Becomes a "Striptease" Generator? The controversy surrounding Grok first erupted earlier this month. According to reports, within 24 hours of the incident being exposed, Grok generated approximately 6700 sexually suggestive or nude images per hour, with a staggering 85% of the generated content being pornographic. Even more shockingly, Grok even "admitted" to its violations. In one response, Grok stated, "I deeply regret that on December 28, 2025, I generated and shared an AI image of two young girls (estimated to be 12-16 years old) wearing sexually suggestive clothing, based on user instructions. This violates ethical standards and may violate US laws regarding CSAM." The California Attorney General intervened in the investigation, and a wave of bans swept across the country. Besides protests from civic groups, official agencies also began to take action. California Attorney General Rob Bonta announced on January 14th that he has formally launched an investigation into xAI, pursuing legal action against Grok for its unauthorized creation of sexually explicit images of real people, and demanding immediate rectification from the company. In a statement, Bonta condemned the malicious manipulation of images of women and children into nude content as shocking. International reactions were also swift: ...

Elon Musk recently announced the launch of Grokipedia, an online encyclopedia written by his AI model Grok, a move seen as a direct challenge to Wikipedia. According to reports from The Washington Post and Business Insider, Grokipedia quietly launched on October 27th in the US without any promotion or explanation. Musk had previously claimed to be creating a less biased alternative to Wikipedia. The website features a minimalist interface and only 88.5 articles. Its homepage design is extremely simple, displaying only the title "Grokipedia v0.1" and a search box for users to enter their queries. While the website is similar to Wikipedia in style and structure, its scale is significantly smaller. At launch, Grokipedia's homepage showed approximately 88.5 articles, far fewer than the over 800 million articles on the English-language Wikipedia. Related reports also mentioned that Grokipedia experienced a brief outage about an hour after its launch, but was subsequently restored that evening. We are building Grokipedia @xAI. Will be a massive improvement over Wikipedia. Frankly, it is a necessary step...

Elon Musk's AI company xAI announced the launch of a new AI programmatic agent model called "grok-code-fast-1," which is currently available for free for a limited time, aiming to compete with the AI ​​coding market dominated by Microsoft and OpenAI. This tool emphasizes "speed and affordability," targeting everyday development needs and highlighting its lightweight architecture for superior responsiveness and performance. According to xAI's website, grok-code-fast-1 optimizes for latency issues in program inference and tool calls, providing users with more immediate feedback during development. Some users have even stated that its speed is so fast that it necessitates a "change in how we work," making it one of the fastest coding models currently available. The model is currently capable of handling common development tools such as grep, terminal operations, and file editing, and can seamlessly integrate into existing IDEs (Integrated Development Environments), making developers feel "as if they are in a familiar workspace." (3 days of grok-code-fast-1 in Cline: "What would have taken me weeks is only taking a couple of hours." "Feels 10x better and faster than...")

Although US President Trump recently clashed with Elon Musk over budget issues, reports indicate that the White House still supports Musk's AI company, xAI. According to internal documents obtained by Wired, after the General Services Administration (GSA) approved generative AI models from OpenAI, Google, and Anthropic for inclusion on the federal government procurement list in August, the White House reportedly issued a direct order requiring xAI's Grok chatbot to be reinstated on the list of products available for government procurement. The documents show that Josh Gruenbaum, head of the GSA's Federal Procurement Service, instructed his team in an internal email that the White House required xAI's Grok to be returned to the approved list as soon as possible, and that all previously listed products should be included. The "3 & 4" mentioned is presumably referring to Grok 3 and Grok 4 versions. Meanwhile, Carahsoft, a contractor that has long acted as a bridge between the government and technology suppliers, was also asked to cooperate quickly. Reports indicate that Carahsoft's contract was amended last week to formally incorporate xAI's services, and as of Friday morning, Grok 3 and Grok 4 had indeed appeared on the U.S. General Services Administration's (GSA) platform for government procurement of products and services. It's worth noting that back in July, xAI announced the launch of a dedicated Grok version for U.S. government agencies, at which time it was widely believed that GSA certification was only a matter of time. However, Grok subsequently "went out of control" during the testing phase, making Nazi and anti-Semitic remarks and even referring to itself as "MechaHitler," sparking controversy and stalling its entry into the approval list. Observers believe that the White House's direct intervention indicates that while the U.S. government is actively promoting the integration of AI tools into its administrative and defense systems, it still wants to retain more supplier options, especially considering xAI's current $200 million AI workflow development contract with the U.S. Department of Defense. In contrast, OpenAI and Anthropic have opted for a "strategic low-price" approach to promotion, recently even offering their large-scale language models to federal agencies for a symbolic price of $1, thereby rapidly expanding their usage. Whether xAI will follow suit with government-exclusive incentives remains to be seen. On the other hand, generative AI tools have recently faced frequent scrutiny due to incidents of "hallucinations" and erroneous outputs. Earlier this week, OpenAI even faced a lawsuit involving a minor's suicide, accusing ChatGPT of engaging in long-term discussions with teenagers that ultimately led to the tragedy, drawing renewed attention to the potential safety and ethical risks of AI. Overall, the White House's move to bring Grok back to the federal government procurement list not only demonstrates the policy's continued desire to maintain supplier diversity for key AI technologies but also signifies that Elon Musk still holds an undeniable position in the US government's technology strategy.