Ellon Mask’s artificial intelligence company XAI is working to develop the next generation of AI, known as the World Models, to compete with technocrats like Meta and Google. This new AI model, based on video and robotic data training, will go beyond large language models that rely on text-based training, such as ChatGPT, Gemini and XAI-owned products, Grok.
This is a continuation of Mask’s earlier declaration that the XAI game studio will launch an excellent AI generation game by the end of next year. According to two sources quoted in the Financial Times, the “World Model” will be used to generate interactive 3D environments for video games, an advanced AI system that not only constructs virtual scenes, but also autonomous navigation and optimization of those scenes. The report reveals that XAI has secretly engaged two AI researchers, Zeeshan Patel and Ethan He, the chip manufacturer in Weaverda, who have gained extensive experience in the development of a world model for the Onniverse platform. Ethan He’s consular information indicates that he was involved in the development of the World Model COSMOS in Weaverda and that Zeeshan Patel was involved in the development of the Model prior to the in-depth imagination study.
XAI is continuously recruiting technical expertise for its “wide-area team”, which specializes in the development of AI technology beyond text (covering images, video and audio). However, the world’s model development still faces major challenges, as the amount of data it needs to learn is far greater than the large-language model, showing exponential growth. Although Internet users have generally expressed their rejection of fully AI-generated video games, some have pointed out the value of using them to simplify development processes and to unleash the creativity of developers. As one user said on platform X: “Ai’s involvement in game development can significantly increase efficiency and free the creative mind.”
It was reported that many practitioners in the technology industry have high expectations of world models, as they can allow artificial intelligence to go beyond software to physical products such as human robots. Last month, XAI launched its new image and video generation model Grok Imagine in social media and made it available to users free of charge.
While current video generation models (e.g., Sora in OpenAI) generate video frames through models in predictive training data, world models represent a major leap. The Financial Times states that they will have a real-time, causal understanding of physics and the interaction of objects in different environments.
