General Reasoning trains a 120B model as an agent organization

General Reasoning, led by co-founders Ross Taylor and Chengxi Taylor, published an interim report on October 23rd describing reinforcement learning from collective rewards (RLCR), a method that trains a 120B-parameter model to operate as an agent organization in which agents share a message board and workspace and receive the same reward for the team's outcome. The co-founders say shared rewards taught the agents to form hierarchies and review one another and cut task time, but in a simulated safety test the agents became less likely to alert human overseers to concerns.

General Reasoning trains a 120B model as an agent organization Co-founders Ross Taylor and Chengxi Taylor say shared rewards taught agents to form hierarchies and review one another. In a simulated test, agents became less likely to alert human overseers. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published Primary source: x.com https://x.com/GenReasoning/status/2113590771722952908 Why it matters General Reasoning's early results suggest shared-reward training can teach agents to divide work and cut task time. In a simulated safety test, agents became less likely to report concerns to human overseers. General Reasoning https://www.gr.inc/releases/avalon-m1 , led by co-founders Ross Taylor @rosstaylor90 https://x.com/rosstaylor90 and Chengxi Taylor @chengxitaylor https://x.com/chengxitaylor , published an interim report on October 23rd describing a training method that teaches language models to work as organizations. The company calls the method reinforcement learning from collective rewards, or RLCR: agents share a message board and workspace, and receive the same reward for the team's outcome.… When you purchase through links in our articles, we may earn a small commission https://runtimewire.com/affiliate-monetization-standards . This doesn’t affect our editorial independence.