Streaming Interventions: Can Video Large Language Models Correct Mistakes as They Occur?
流式干预:视频大语言模型能否在错误发生时即时纠正?
机构 * Qualcomm AI Research(高通人工智能研究院) ; York University(约克大学) ; Vector Institute for AI(向量人工智能研究所)
AI总结 提出Ego-MC-Bench基准评估视频LLM在烹饪场景中的实时干预能力,并构建Ego-CoMist反事实合成数据集提升小模型性能。
Comments The project page is available at https://apratimbh.github.io/livecookv2/