60-agent swarm tunes Stockfish NNUE weights, wins all comparisons in 96-game screen
imjustnewatai · x · 2026-11-06
A developer is running a swarm of 60 "GPT 6 Astra" agents (active for over a day and 16 hours) to iteratively edit Stockfish's NNUE weights.
Screenshots show the 96-game screening passed audit, with the edited NNUE ahead in all three comparisons: 11W-13D-8L vs original weights, 11W-11D-10L vs official development, and 10W-17D-5L vs rated reference. The system itself notes this is not yet a statistically verified Elo gain and has frozen the candidate for a longer confirmation test.
Related event: 60 Agents Tune Stockfish NNUE Weights, No Significant Elo Gain Yet(2 posts)→
More from coding & agent
- DoorDash blocks computer-use agents, so this team orders every meal via agents on Uber Eats — RishabJainK · 2026-11-06
- DoorDash blocks computer-use agents, so this team automates every meal via Uber Eats — RishabJainK · 2026-11-06
- Claude spots a real stack buffer overflow in OpenBLAS while casually explaining kernel code — Aizkmusic · 2026-11-06
- Google Antigravity rolls out remote agents: tasks run in cloud VMs, approvals via phone — gaganghotra_ · 2026-11-06
- Open-source sprite-gen generates 2D game sprites and animation atlases, smoother GIFs than GPT Image — songguoxiansen · 2026-11-06
- AI revives Facebook's abandoned Prepack, speeding up React renderToString 5.4x — Vjeux · 2026-11-06