A Skill-augmented Agentic Framework and Benchmark for Multi-Video Understanding
The episode discusses a paper introducing MVXBench, a benchmark for evaluating Multimodal Large Language Models (MLLMs) on complex multi-video understanding tasks. The hosts detail the SAMA framework, which uses an Agent Core with an LLM Planner and a Verification Layer to guide the model using external visual tools. This approach aims to improve cross-video analysis by enabling iterative reasoning and conflict detection.