Search papers, labs, and topics across Lattice.
1
0
3
VLMs can't "see" squares: their spatial reasoning collapses when transcribing binary grids lacking textual cues, revealing a surprising reliance on text recognition over native visual processing.
