RLCR trains language models as agent organisations that share a message board
General Reasoning says its 120B Avalon-M1-mini learned hierarchies and hired verifiers.
General Reasoning (@GenReasoning) described RLCR on Friday, a method that trains language models as agent organisations sharing a message board. According to the post, Avalon-M1-mini (120B) learned hierarchies and hired verifiers and testers. It drew about 1,300 likes.
The idea is to train many agents to coordinate on a shared channel, rather than a single model working alone. The post does not include results tables.
Why it mattersMulti-agent setups are a growing research direction. Look for the paper to see how the claims are measured.
