Steering CLIP's vision transformer with sparse autoencoders

Re-assign community