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 · Published

A glowing tiered circular display holds miniature figures at desks, with a lone figure beside it, an amber beacon, hanging planets, and an “Rw” mark.

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.