VIBE: Annotation-Free Video-to-Text Information Bottleneck Evaluation for TL;DR | Read Paper on Bytez