The limiting factor in physical AI isn't compute or architecture - it's data

The crux of the issue in advancing physical AI isn't the computing power or the architectural designs, but rather the scarcity of robot data. Kanishka Rao from Google DeepMind highlights that the vastness of internet data doesn't match the limited scale of interaction data robots need to learn from. This data gap is fundamentally impeding progress in robotics, regardless of the chosen architecture. As we delve deeper into the future of video world models, it's clear that more substantial datasets are crucial for making significant breakthroughs in physical AI.