07 · 31 · 2026 Another Typical Autoresearch Post I built a (mostly) autonomous system that took 3rd in GPUMode's eigh kernel competition. Two weeks, 1,293 experiments, 779 million output tokens, and ultimately an 8.56x speedup over torch. And three small fixes later, a real Shampoo training run 2.2x faster to quality. AutoresearchSystems