Search papers, labs, and topics across Lattice.
1
0
3
A surprisingly effective defense against LVLM image attacks can be built without any training, using only image transformations and strategic LLM prompting.