Search papers, labs, and topics across Lattice.
1
0
3
2
Current video understanding benchmarks and post-training datasets are riddled with linguistic biases, meaning VLMs might be acing tests without actually "watching" the video.
