FYI "大语言模型可以通过生成推理痕迹和任务特定行动来实现更大的协同作用。" 这句话的原文是："In this paper, we explore the use of LLMs to generate both reasoning traces and task-specific actions in an interleaved manner, allowing for greater synergy between the two"。 后面还有半句："reasoning traces help the model induce, track, and update action plans as well as handle exceptions, while actions allow it to interface with and gather additional information from external sources such as knowledge bases or environments."