The real impact on the industry: - Data quality still matters, but it is no longer the only thing that matters. True scalability only becomes possible when we can effectively make use of low-precision, imperfect data at scale. - Human video can be transferred directly to the robot embodiment during pretraining, without any retargeting. Real-robot data is no longer the single dominant source of value. In that sense, UMI is essentially an intermediate form factor. - The dynamics in video can genuinely be learned and then transferred to robots through an IDM. The key to scaling is scaling up the learning of video dynamics—not scaling up the IDM itself. - The action model is at the core of System 1, and the ability to follow language is its most important capability.