AI RESEARCH
GridProbe: Posterior-Probing for Adaptive Test-Time Compute in Long-Video VLMs
arXiv CS.AI
•
ArXi:2605.10762v1 Announce Type: cross Long-video understanding in VLMs is bottlenecked by a single monolithic forward pass over thousands of frames at quadratic attention cost. A common mitigation is to first select a small subset of informative frames before the forward pass; common for