autoresearch

AI agents running research on single-GPU nanochat training automatically

One day, frontier AI research used to be done by meat computers in between eating, sleeping, having other fun, and synchronizing once in a while using sound wave interconnect in the ritual of "group meeting". That era is long gone. It has reached 83,247 GitHub stars, written primarily in Python.

Why now: Recent coverage — "Karpathy's Autoresearch: Autonomous AI Loop on a Single GPU" — alongside renewed developer interest is driving current visibility.

• 83,247 GitHub stars, 12,092 forks — verifiable at github.com/karpathy/autoresearch.
• Primary language Python.
• No license file detected — commercial users should confirm terms with the maintainers.
• Third-party coverage: "Karpathy's Autoresearch: Autonomous AI Loop on a Single GPU".

Considerations: At roughly 3 months old, it's early — the architecture and maintenance cadence haven't been battle-tested yet.

ESTABLISHED LEADER · Build on it: Proven at scale (Python) — safe to evaluate for production use or coverage; the open question is fit, not validity.

Sources: karpathy/autoresearch on GitHub · Karpathy's Autoresearch: Autonomous AI Loop on a Single GPU · Andrej Karpathy Open-Sources 'Autoresearch': A 630-Line Python ...