1 paper · 1 filter
Yu Cheng, Yongkang Hu, Shuaijie Ma +12
LLM agents deployed in real-world environments continually encounter new tasks and safety risks, while execution feedback typically becomes available only after each task is comple…