Why is Elon Musk's xAI joining the 'world models' race after Meta, Google?

Elon Musk's artificial intelligence startup xAI is working on developing its own "world models" to rival the likes of Meta and Google.

The next-generation AI models, which are trained on videos and data from robots to understand the real world, will surpass traditional large language models (LLMs) which are trained on text, such as ChatGPT, Gemini, and xAI's own Grok.

This follows Musk's earlier claim about xAI's game studio releasing a "great AI-generated game before the end of next year".

The 'world models' will reportedly be used to generate interactive 3D environments for video games, a Financial Times report said, citing two unnamed people in the know. The advanced AI systems will also be able to navigate through these systems and refine them.

The report also claimed that xAI had quietly hired Zeeshan Patel and Ethan He, two AI researchers from chipmaker NVIDIA with experience developing world models for its Omniverse platform.

NVIDIA's Omniverse Enterprise is a collaborative 3D platform that lets developers build generative AI-enabled tools.

Ethan He's LinkedIn profile shows he previously worked on NVIDIA's COSMOS world model as a Staff Engineer. Patel's profile shows he worked on the COSMOS model too, before doing additional Deep Imagination Research (DIR).

The San Francisco-based startup continues to recruit technical staff for its "omni team"—a team handling AI development beyond text (across image, video, and audio).

However, world models continue to remain a significant challenge, owing to the colossal amounts of data they learn from—exponentially more than LLMs.

While netizens have often expressed disgust at video games that are completely AI-generated, a few have pointed out that it would be great if used to simplify game development to allow developers to be more creative.

"AI in game dev makes a ton of sense for efficiency and letting these creative juices flow freely," a user wrote on X. 

Sci/Tech