Self-Improving AI Infrastructure for GPU Kernels
How can AI agents write faster GPU code than human experts? Bing Xu (INT21, co-author of GANs) introduces the PTX Kernel Factory—an autonomous agent swarm that evolves and optimizes low-level NVIDIA PTX code. Full deep-dive on the benchmarks, the AlphaGo-style search loop, and why the CUDA moat is getting deeper, not shallower. Chapters and links below. #PTX #CUDA #GPU Full episode: • Bing Xu: AI That Writes Faster GPU Code Th... 0:00 Is the CUDA moat getting deeper? 0:35 AI writes faster GPU code than experts. 1:31 GPT 5.5 breaks AI plateaus. 2:07 Who is Bing Xu 4:14 What is the PTX K…