3 papers across 2 sessions
Perform input dependent steering of MLLMs by learning to predict steering vectors from input context