Tag: Grok 4

Following the launch of Grok 4 by xAI, Elon Musk recently announced through the social media platform "X" the upcoming launch of an AI application service designed for children, named "Baby Grok." While specific functional details have not yet been revealed, Musk emphasized that the application service will provide "child-friendly content," signifying xAI's active expansion of its AI technology applications. The actual operation and content format of "Baby Grok" are currently unknown, but foreign media generally speculate that the application will focus on an AI user experience that combines education and interactivity, creating a safer and more controlled digital interactive environment for younger users. "We're going to make Baby Grok @xAI, an app dedicated to kid-friendly content— Elon Musk (@elonmusk) July 20, 2025" According to a Bloomberg News report, this application service will be an important step for xAI towards diversification, especially given the current trend of generative AI models moving towards widespread application while facing numerous controversies. Shifting towards child-friendly content may help alleviate market perceptions and social pressure. In fact, the Grok AI chatbot has recently become a focus of attention, with the new version of Grok...

The Wall Street Journal reports that SpaceX, Elon Musk's space exploration company, has pledged $2 billion to invest in xAI, Musk's AI startup, providing additional funding to develop AI applications that can compete with OpenAI. The report notes that prior to this large investment, xAI had already merged with Twitter's social media platform "X," increasing its overall market valuation to $113 billion. Elon Musk's attempt to integrate his holdings—including SpaceX, Tesla, "X," and xAI—to build a complete AI ecosystem encompassing everything from communication infrastructure to application scenarios is clearly aimed at competing with the growing OpenAI and tech giants like Google and Meta. Currently, xAI's conversational AI, Grok, is already being used in SpaceX's Starlink satellite communication customer service, and there are plans to introduce it into Tesla's Optimus humanoid robot, enhancing its conversational understanding and task execution capabilities. Elon Musk has stated that Grok is "the smartest AI on Earth" and emphasized its continued optimization towards AGI applications. Analysis indicates that Elon Musk's investment in xAI through SpaceX reflects not only his strategic commitment to the AI ​​field but also SpaceX's status as a flagship business with strong cash flow and technological capabilities, becoming a key resource for xAI's rapid development. In contrast, OpenAI, since the launch of GPT-4, has become the world's most prominent generative AI platform. Competitors such as Meta, Google, and Anthropic continue to strengthen their AI model deployments, intensifying competition. Against this backdrop, whether xAI can quickly catch up through its funding, technology, and platform advantages is worth continued attention. In recent developments, xAI has announced the launch of Grok 4, supporting multi-agent, high-intensity training, and emotional speech. This includes a single-agent version and Grok 4 Heavy, which supports up to four agents simultaneously and features contextual understanding capabilities for up to 256,000 tokens, enabling AI to handle long texts and complex tasks more effectively. Tesla earlier announced that it would apply Grok AI to certain models, allowing owners to chat and interact with the AI ​​assistant while driving. Tesla also stated that all vehicles delivered in the US from July 12th onwards would have Grok AI pre-installed, while previously sold and launched models would receive OTA updates depending on hardware and software availability. This includes models equipped with AMD infotainment processors and running system versions 2025.26 and above. However, a bug in an update caused Grok to generate numerous hate messages on the xAI platform for 16 hours, including positive descriptions of controversial historical figures such as Hitler, sparking backlash from users and the media. xAI subsequently issued an apology through its official account, stating that the issue stemmed from an "accidentally triggered legacy code path," emphasizing that the problem did not originate from the underlying language model but rather from an oversight in the processing architecture. The company promised to remove the relevant code and refactor the system to prevent recurrence, and also announced the release of new prompt word design logic to improve the model's judgment and stability in specific situations.

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.