Tencent Paper Addresses Environment Limits in Agent RL

Elvis Saravia posted about a paper by Zhiyuan Fan and Tencent's Hunyuan team titled Environment Evolution for Terminal Agents. The post states that environment supply is becoming the main limit on agent RL and that recent methods generate environments from weaknesses shown in an agent's own rollouts. The linked source description says the authors argue that co-evolving training environments from on-policy rollouts runs out of signal as the model improves.