Search papers, labs, and topics across Lattice.
1
0
2
Stop blindly steering all layers of your LVLM - this new method uses attribution to apply targeted interventions only where hallucinations originate, preserving performance on general tasks.
