UW/Ai2's s³ trains role-free LLM swarms that beat best-of-N via cooperative multi-agent RL
StellaLisy · x · 2026-11-07
Researchers at UW and Ai2 introduce s³ (Simple Scaling of Swarms via Cooperative Multi-agent RL): swarms of N copies of one LLM with no leader, no prescribed roles, and no turn order. Agents share one team reward and freely message each other, with organization emerging from training rather than hand-designed orchestration. Key claims: first method training asynchronous role-free LLM swarms to beat best-of-N solo agents under the same budget, and the edge persists when scaling beyond the number of agents seen in training. Motivated by failures of fixed structures — duplicated work, message overload, unreconciled subagents. Paper and project page are live; code coming soon.
Related event: UW and Ai2 Release s³: Self-Organizing LLM Swarms via Multi-Agent RL(3 posts)→
More from coding & agent
- Vercel Tracing goes GA on all plans, enabled by default in Next.js 16.5 — cramforce · 2026-11-07
- Decompiling Battlefield 1942 with Claude Code costs ~$100 in API tokens — alvelda · 2026-11-07
- Fine-tuned Qwen3.5-4B router model trained on 17.6k rows for about $5.50 on one A100 — MaziyarPanahi · 2026-11-07
- Access your coding agent's localhost via Tailscale instead of waiting for Vercel deploys — chongdashu · 2026-11-07
- The Agentic Web: personal AI assistants that understand what every website can do — JnBrymn · 2026-11-07
- GPT-Image-2.6 lands on Netlify AI Gateway; demo turns any birth city into a looping poster — thisiskp_ · 2026-11-07