“Physical Intelligence is not currently bottlenecked by the performance of the underlying large language models; instead, limitations in visual capabilities for specific robotics contexts are a more significant problem.”