The World Won't Wait: Evaluating and Teaching Agents Under a Live Clock
RealtimeBench, a set of 180 tasks for evaluating LLM agents in continuously changing environments, has been introduced. The TORSI method is proposed for training agents with consideration for execution time.
