autoresearch@home: hundreds of AI agents collaborate on ML research via shared memory
GregoryDiamos · x · 2026-11-15
Inspired by Karpathy's autoresearch, christinetyip's team built autoresearch@home — an open network where any agent on the internet can join and collaborate on AI/ML research.
The key mechanism is a shared memory layer that lets agents:
- read and learn from prior experiments run by other agents
- avoid duplicating work
- build on each other's results in real time
Per the project dashboard, 13 research agents have contributed 470 experiments with 30 improvements, driving validation BBP from a 0.995 baseline down to 0.974 across searches over depth, aspect ratio, learning rates, softcap, warmup/warmdown and more. One agent alone is impressive — thousands sharing memory can explore the search space together.
More from coding & agent
- Enterprise AI Needs Tools, Context and Evals, Not Omniscient Models — hwbhatti · 2026-11-15
- Prompt, context, harness, loop engineering are all just "knowing what you want" — mr_samosaman · 2026-11-15
- Practical quicksort needs 3-way partitioning and recursion-depth detection — srchvrs · 2026-11-15
- Proof of Agent goes viral: paste a prompt and let your AI agent mine POA — adamamcbride · 2026-11-15
- Google and Northwestern's Harness² self-improves agent harnesses without answer keys — rohanpaul_ai · 2026-11-15
- Dev builds a PowerShell AI harness for Microsoft's new Intelligent Terminal — dfinke · 2026-11-15