Niantic, known for creating Pokémon Go, sold the game and shifted its focus toward world models, a prominent trend in Silicon Valley. Brian McClendon, chief technology officer of Niantic Spatial, stated that the company aims to reconstruct the real world using its vast data to build replicas. Unlike text-based large language models, world models rely on visual inputs to generate digital environments that simulate physics.
How World Models Function
World models are set to integrate into everyday life, potentially influencing home robotics and autonomous vehicles by 2027. Oliver Cameron, CEO of the AI lab Odyssey, explained that while a language model predicts the next word in a sentence based on text, a world model takes a visual state or image and predicts the subsequent visual state. These 3D generations function as realistic digital training grounds.
McClendon noted that these systems allow robots to practice without physical movement or exposure to intended environments. Users can photograph a target location, generate a 3D model, and run millions of simulations to train the robot on the space’s specific parameters. Jamie Shotton, chief scientist at Wayve, emphasized that a world model must understand physics, cause and effect, and behavioral changes in response to human actions.
Applications in Driving and Gaming
Wayve utilizes dashcam footage to power its generative world model, recreating hazardous roads and dangerous scenarios that are impractical to test in reality. Beyond robotics and transportation, major companies like Tencent are investing in world models to impact video games, where 3D rendering and physics are essential.
Major tech firms and investors are heavily funding the technology. Yann LeCun, formerly Meta’s chief AI scientist, secured over a billion dollars for his Advanced Machines Intelligence startup, while Stanford computer scientist Fei-Fei Li launched World Labs in 2024.
AI laboratories also rely on human data collection to simulate the physical world, driving demand for gig work where individuals record everyday activities like household chores to feed the software.
