Search papers, labs, and topics across Lattice.
2
0
6
0
VideoChat3 achieves unprecedented generalization in video understanding while maintaining high efficiency, outperforming larger models with just 4 billion parameters.
Artifact-centered evaluations reveal that LLM agents can achieve an 88.6% success rate in structural engineering tasks, but still struggle with invalid inputs and model consistency.
