awesome-ai-radar
← Back to Radar

Karpathy's autoresearch Hits 23k Stars in 3 Days: AI Agents Running ML Research

ml-research agents open-source training

What happened

Andrej Karpathy released autoresearch on March 6, 2026, a tool that enables AI agents to automatically run nanochat (small chat model) training research on a single GPU. As of March 9, the project had reached 22,983 GitHub stars and topped GitHub trending. The associated Hacker News discussion reached 198 points with 56 comments.

Why it matters

autoresearch demonstrates a concrete implementation of AI agents conducting ML research autonomously — not just writing code, but designing experiments, running training, evaluating results, and iterating. This is a practical example of the "AI scientist" concept moving from theory to open-source tooling that researchers can run on consumer hardware.

Who should pay attention

  • ML researchers interested in automated experiment design and hyperparameter exploration
  • Developers exploring agentic workflows beyond code generation
  • Anyone with a single GPU who wants to experiment with autonomous research agents