Scaling Karpathy's Autoresearch: What Happens When the Agent Gets a GPU Cluster
A team pointedClaude Codeatautoresearchand spun up 16 Kubernetes GPUs. The setup ran ~910 experiments in 8 hours.val_bpbdropped from 1.003 to 0.974 (2.87%). Throughput climbed ~9×. Parallel factorial waves revealedAR=96as the best width. The pipeline usedH100for cheap screening andH200for validation.. read more











