Study Tests Whether Small Video-Language Models Truly Use Their Cited Evidence
A new arXiv paper proposes counterfactual tests to check whether grounded reasoning chains in small video-language models—where each step cites a frame span and bounding box—actually depend on the cited evidence.
Read the full story at arXiv cs.CV ↗
Timeline · 1 report
- 2026-11-18 05:00 · arXiv cs.CV
Does the Model Use What It Cites? Counterfactual Tests of Grounded Reasoning Chains in Small Video-Language Models