Search papers, labs, and topics across Lattice.
1
0
3
Real-world images can actually *harm* a vision-language model's ability to understand abstract concepts, as models latch onto irrelevant visual details.
