Search papers, labs, and topics across Lattice.
1
0
3
A single, optimized text snippet can fool CLIP into thinking it's a good caption for almost any image, revealing a surprising vulnerability in cross-modal understanding.
