Elon Musk announced on social media platform X on August 2 that Grok, the model developed by his artificial intelligence company xAI, now has the capability to analyze any video. To showcase this breakthrough, Musk posted a lifelike AI-generated video of late basketball legend Kobe Bryant, claiming that Grok can not only rapidly “digest” long-form videos but also accurately identify deepfake content and trace its technical origins.<\/p>\n
In the 63-second video, Bryant, dressed in a black suit, sits on a sofa against a nighttime Los Angeles backdrop, pitching Grok 4.5 to viewers and signing off with “Mamba Out.” Given that Bryant passed away in 2020, the video’s realism drew widespread astonishment. Musk then displayed Grok’s analysis report of the video, in which Grok clearly stated: “This is an AI-generated deepfake video. Kobe Bryant passed away in 2020, so the entire video is completely synthetic.” Additionally, Grok inferred that the video was most likely produced by xAI’s own Grok Imagine Video 1.5, citing that the 63-second runtime far exceeds the typical limits of text-to-video generation, and likely employed a hybrid workflow combining multi-segment stitching with “image reference + text prompt + lip-sync” techniques.<\/p>\n
Despite the impressive demonstration, skepticism remains over whether Grok truly “understands” video frames. Some technical experts have pointed out that Grok’s processing logic may not be based on frame-by-frame visual rendering, but rather relies on video subtitles or audio transcription. One user tested the model with an 11-minute video and found that some information in the summary contained transcription errors or fabricated details. To verify the claims, our publication conducted hands-on testing, selecting a video about the famous mathematical problem known as the Collatz conjecture (also called the 3n+1 problem). The conjecture has remained unproven for nearly a century. In 2019, mathematician Terence Tao made a breakthrough advance on the problem.<\/p>\n
Test results showed that Grok not only accurately interpreted the video’s core mathematical content but also analyzed the logic behind the animation’s construction. The video depicted a Collatz reverse tree built with the number 1 as the root node, visually presenting the conjecture’s structure that “all positive integers eventually enter the 4\u21922\u21921 loop.” Grok was able to clearly articulate the mapping relationships between even and odd nodes and highlight the animation’s mathematical significance. Additionally, Grok was asked to analyze a 50-plus-minute interview video featuring AI expert Geoffrey Hinton discussing topics such as AI consciousness, control, and alignment. By combining post annotations with subtitle extraction, Grok successfully distilled the interview’s core viewpoints and timeline, and even analyzed the connection between the video and the post’s context.<\/p>\n
These test results appear to demonstrate that Grok does not rely solely on subtitles, but possesses deeper video comprehension capabilities. One user marveled that Grok took only about 36 seconds to generate a detailed summary of a 30-minute video’s worth of information\u2014a true “overwhelming advantage.” For contemporary users suffering from information overload, this feature could significantly boost information-gathering efficiency.<\/p>\n
While Grok was showcasing its video understanding capabilities, Musk’s attitude toward competitors also drew attention. Netizens noticed that Musk quietly followed the official account of Chinese AI startup DeepSeek on X on August 1. The move coincided with DeepSeek’s July 31 announcement that the official API for its next-generation model, DeepSeek-V4-Flash, had entered public beta testing. When asked directly about “the boss following a competitor,” Grok candidly responded that DeepSeek has achieved\u6781\u81f4 efficiency optimization, using lower compute and costs to deliver model performance approaching\u2014and in some areas surpassing\u2014top Western standards. For xAI and Tesla (TSLA-US), Grok noted, this is a competitor that demands close attention, and following the account is the most direct form of “market monitoring.”<\/p>\n
Looking back at Musk’s attitude toward DeepSeek, it has been anything but consistent. In early 2025, when DeepSeek R1 was released, Musk dismissed claims of its low-cost development and echoed Scale AI’s founder’s assertion that the company possessed 50,000 NVIDIA H100 chips, earning him the nickname “DeepSeek’s number one hater” among netizens. However, starting in February 2026, Musk’s stance began to shift. In May, he publicly praised DeepSeek for building a large-scale model rivaling top global standards with fewer than 2,000 NVIDIA chips and a budget of approximately $5.5 million. In July, he stated in an interview that China possesses formidable AI capabilities and could potentially take the lead in the future.<\/p>\n
The newly launched DeepSeek-V4-Flash official version maintains the same model architecture and parameter count, employing a Mixture-of-Experts (MoE) architecture with 284 billion total parameters and 13 billion activated parameters. Through post-training recalibration alone, the model’s agentic capabilities were significantly enhanced. Market reports also suggest that Tesla’s in-car systems in China will adopt a dual-model approach combining ByteDance’s Doubao and DeepSeek in the future.<\/p>\n
Analysts note that Musk’s high-profile demonstration of Grok’s video understanding capabilities comes at a time of intensifying AI competition. Earlier, OpenAI announced that its next-generation model Astra had solved 10 significant open mathematical problems, sending shockwaves through the industry. Musk’s move may be aimed at showcasing xAI’s strength in the multimodal domain to counter the increasingly fierce competitive landscape. Meanwhile, his shifting attitude toward DeepSeek reflects the rapid rise of China’s AI power, which can no longer be ignored.<\/p>\n","protected":false},"excerpt":{"rendered":"Elon Musk announced on social media platform X on August 2 that Grok, the model developed by his…\n","protected":false},"author":2,"featured_media":129058,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10],"tags":[64537,64536,140,6364,47109,64538,63613,17122,2616,2899],"class_list":["post-129057","post","type-post","status-publish","format-standard","has-post-thumbnail","category-xai","tag-collatz-conjecture","tag-deepseek-v4-flash","tag-elon-musk","tag-grok","tag-grok-4-5","tag-kobe-bryant","tag-openai-astra","tag-terence-tao","tag-tesla","tag-xai"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/129057","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=129057"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/129057\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/129058"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=129057"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=129057"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=129057"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}