Physical AI's Next Bottleneck Is Finding the Right Video Data

Rafael Levi of Bright Data argues that while AI models are powerful, the bottleneck for physical AI is the scarcity of relevant video data for robot training. Unlike LLMs with trillions of words, robotics has limited video of robots performing tasks, and staged recordings produce biased data. He suggests utilizing the vast amount of real-world video on the web to train robots to understand physics and human actions more naturally.