Search papers, labs, and topics across Lattice.
1
0
2
Hallucinations in LVLMs can be effectively mitigated without training by leveraging per-instance subspaces that adaptively suppress specific hallucination modes.
