A 10,000 GPU cluster in China and Nvidia's new self-improvement loop for robotics lead this week's research. The robotics system applies agentic learning to real-world physical tasks. These developments push the boundaries of hardware scale and autonomous refinement.
The Signal
Practitioners should monitor how these loops reduce the need for manual data labeling in robotics.