Search papers, labs, and topics across Lattice.
1
0
3
2
VLA models can achieve 80% F1-score in task complexity detection using only visual embeddings and 5% of training data, suggesting language may be surprisingly redundant for this specific capability.
