British startup uses gamer inputs to train AI models
worldmodeldata licensed 1 million hours of video game data to train ai models that navigate the physical world. this could improve autonomous vehicles and robots.
worldmodeldata, a british startup advised by yann lecun, is collecting video game data to train ai models that can navigate the physical world. the company has licensed almost 1 million hours of data from studios behind popular video games. this data, which includes controller inputs and visual representations of 3d spaces, will be used to train world models. these models, which are being developed by researchers like fei-fei li and yann lecun, need to be trained on a combination of visual and action data to become fluent in real-world physics. the goal is to improve autonomous vehicles, robotic arms, and other applications that require finesse and precision.