ST-Lite: Training-Free KV Cache Compression with Spatio-Trajectory Guidance for Long-Horizon GUI Agents
The episode discusses the paper 'ST-Lite,' which introduces a method for training-free KV Cache Compression in GUI agents. This technology allows AI agents to maintain complex context over very long periods, overcoming previous limitations where they would forget past events. By using Spatio-Trajectory Guidance, the system selectively preserves essential memory while significantly improving performance and reducing computational overhead.