#memory-efficient language agents

MIT and NUS researchers introduce MEM1, a reinforcement learning framework that enables language agents to efficiently manage memory during complex multi-turn tasks, outperforming larger models in speed and resource use.