Scaling Karpathy's Autoresearch: What Happens When the Agent Gets a GPU Cluster
Summary
A SkyPilot blog post details scaling Karpathy's autoresearch by giving a coding agent access to a GPU cluster to run hundreds of experiments in parallel. It shows how parallelism transforms search strategies, highlights using heterogeneous hardware for speedups, and documents the cost and setup to scale autoresearch on your own GPU infrastructure.