RLCR: an 8-person team trains LLMs as agent organizations with shared message boards
PMinervini · x · 2026-10-23
Ross Taylor (GenReasoning) shares a year of research on agent collectives, introducing RLCR (Reinforcement Learning from Collective Rewards): training language models as agent organizations that share a message board and workspace, used to train Avalon-M1-Mini (120B).
Key findings:
- Learnt organizational hierarchies and divisions of labor;
- Increased collective verification, including 'hiring' verifiers and testers;
- The rise of collective thinking, where agents reason about other agents and how to help them.
The authors believe RLCR offers a laboratory for studying collective intelligence and multi-agent safety. The team is only 8 people and still running experiments, ablations, and further training.
More from AGI Musings
- Lightspeed podcast: agent-era cybersecurity and why models may be more interpretable than thought — theamberyang · 2026-10-24
- Treat freedom of alignment like free speech: billions of identical agents risk ethical monoculture — robleclerc · 2026-10-24
- AI-driven vulnerability discovery will push cybercrime losses onto society — mimi10v3 · 2026-10-24
- End of 'good enough software': formal verification becomes the missing AI demand curve — bookwormengr · 2026-10-24
- Revisiting Dario's 'Big Blob of Compute': RL distributions teach no learnable alignment lesson — 1a3orn · 2026-10-24
- How would you verify a US-China AI deal? Podcast digs into inference verification tech — HaydnBelfield · 2026-10-24