venturebeat.com
|
ksl
|
|
Andrej Karpathy open-sourced AutoResearch, a 630-line Python tool that lets an AI agent autonomously run ML experiments on a single GPU – modifying code, training for five minutes, evaluating results, and looping. Left running for two days on a depth-12 model, it processed around 700 changes and found roughly 20 genuine improvements that transferred to larger models, cutting the “Time to GPT-2” leaderboard metric from 2.02 hours to 1.80. The repo hit 8,000 GitHub stars within days. Already, Hyperspace AI distributed the loop across a peer-to-peer network where 35 agents ran 333 experiments in a single night without any human involvement. Autonomous experimentation at this scale has been a running theme in ML tooling lately, but Karpathy packaging it as a dead-simple script makes the barrier to entry essentially zero.
