Search papers, labs, and topics across Lattice.
2
0
3
Video-LLMs fail to effectively learn and apply skills from long video memories, revealing a fundamental gap in their capabilities.
Reinforcement learning enables video-LLMs to re-watch and refine answers without the costly overhead of chain-of-thought training, achieving better performance with less computation.
