Elon Musk’s artificial intelligence company xAI is accelerating the development of so-called “world models,” joining competitors like Meta and Google in creating AI systems that can understand, build, and manipulate physical environments.
This summer, xAI recruited several experts from NVIDIA to build the next generation of AI models. These models are trained on video and robotics data to understand the mechanics of the real world.
The industry believes that world models have the potential to break through the limitations of current large language models, which rely on text training and power popular AI tools like ChatGPT and Grok. This technology is ultimately expected to help create “Artificial General Intelligence” (AGI) that surpasses human capabilities.
World models are central to achieving spatial intelligence. By integrating multimodal data, they provide an internal environmental representation for spatial intelligence technologies to reason and operate in three-dimensional space. They are also considered the foundation of physical reasoning, enabling systems to simulate future states and move closer to human-like intelligence, thereby bridging the cognitive alignment gap between humans and machines.
xAI has hired two AI researchers from NVIDIA: Zeeshan Patel and Ethan He. NVIDIA is a leader in world model technology with its Omniverse platform, which is used to create and run virtual simulation environments.
Some tech companies are pinning high hopes on world models, believing they could enable AI to move from software and virtual spaces into the physical world, driving products like humanoid robots. Last month, NVIDIA suggested that the potential market size for world models could be comparable to today’s global economy.
On social media platform X, Musk posted that xAI plans to launch an “excellent AI-generated game” by the end of next year to fulfill a goal he set last year.