Beacon: Knowing When and How to Perform Agentic Visual Reasoning
The episode discusses the paper "Beacon: Knowing When and How to Perform Agentic Visual Reasoning." Hosts discuss how existing models struggle with tool use by focusing on Mode Adaptiveness and Tool Effect. The paper proposes improvements using a new RL framework with Necessity-Aware Adaptive Reward (NAAR) and Hint-Guided Capability Expansion (HCE) to encourage selective, effective tool invocation.