‘World models’ are AI’s latest sensation: what are they and what can they do?
Researchers are developing 'world models', AI systems that can produce interactive, virtual environments, trained on real-world data, to improve AI-powered robots and self-driving vehicles. Companies like Google, Nvidia, and start-ups are investing heavily in this technology, with potential applications in training AI systems to control robots.
An ongoing trend in artificial intelligence (AI) is the development of 'world models', systems trained on real-world data that can embody virtual, interactive, and 3D environments. These models can produce consistent and explorable worlds, reminiscent of first-person video games, and have the potential to improve AI-powered robots and self-driving vehicles. AMI Labs, founded by Yann LeCun, has raised over $1 billion for its radical approach to world models. Google and Nvidia are also developing world models, as are several start-up companies. World models are trained on thousands of hours of real-world videos and simulations of physical environments. They can provide a safe setting to train AI systems that control robots, much faster than letting robots learn by interacting with physical objects. Genie 3, a world model released by Google DeepMind in August 2025, uses simple text descriptions to generate photorealistic environments that can be explored in real time. The development of world models is attracting huge investment and business interest, with potential applications in robotics and self-driving vehicles.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.