Elon Musk's xAI is stepping up its work on world models, positioning itself against competitors like Meta and Google in the effort to engineer AI that can comprehend and construct physical settings.

This summer, the San Francisco startup onboarded talent from Nvidia to help develop these sophisticated AI systems, which learn from video content and robotic inputs to interpret everyday realities.

Such models could surpass the limitations of text-focused large language models that drive services like ChatGPT and xAI's own Grok.

Sources close to the project revealed that xAI is prioritizing world models for use in video games, where they might enable dynamic 3D worlds. One insider suggested applications could extend to robotics as well.

The new recruits include Zeeshan Patel and Ethan He, both former Nvidia AI specialists versed in world models. Nvidia has been at the forefront with its Omniverse tool for building simulations.

Many in the tech sector anticipate broad impacts from world models, potentially integrating AI into tangible items like humanoid robots. Nvidia recently shared with the Financial Times that this technology's market could rival the current global economy in scale.

Elon Musk announced on X that xAI aims to launch a "great AI-generated game before the end of next year."

This week, on Tuesday, xAI unveiled an enhanced model for generating images and videos, highlighting significant improvements and offering it at no cost.

Tools like OpenAI's Sora currently create videos by anticipating sequences from learned data patterns. In contrast, world models promise deeper insights, incorporating live awareness of physical laws and how elements behave in various contexts.

xAI is actively hiring engineers specializing in image and video creation for its "omni team," focused on crafting immersive AI interactions that span text, visuals, video, and sound.