“The Seikan paper was the first to demonstrate how common sense knowledge from language models could be applied to robotics, significantly reducing the need for robot-specific data collection.”