The conversation highlights a fundamental disconnect between AI language models and robotic manipulation, emphasizing that while mobility in robots has improved through better hardware and simulation, dexterity remains a major bottleneck. Tasks requiring fine motor skills, like tying shoelaces, expose the limitations of current systems due to insufficient tactile sensing and sensor-rich control. The discussion underscores how vision alone—despite advances—is inadequate without integrated feedback mechanisms. A critical barrier is the massive data gap: robotics lacks the equivalent of centuries of real-world interaction data that language models have been trained on. Surprisingly, simple grippers often outperform complex, human-like hands, as demonstrated by Dex-Net and Ambi Robotics, whose success lies in data-driven software rather than intricate design. Real-world testing reveals persistent challenges with deformable objects, reinforcing that practical, specialized robots are advancing faster than general-purpose humanoids. Progress hinges on accumulating real-world operational data, not just simulation.