Search papers, labs, and topics across Lattice.
1
0
2
VLMs exhibit a staggering drop in performance when forced to rely solely on image content, with open models plummeting to just 1-9% accuracy on the most challenging questions.
