S. Lapuschkin

The results suggest that gradient-based attribution can identify data responsible for subliminal learning in some settings, but that some approximations are more reliable than others.

Steering isn't just a trick; it's a fundamentally different way to adapt language models, offering localized, reversible control that traditional fine-tuning can't match.

Forget IoU, measuring the structural compactness of attribution maps with Minimum Spanning Trees reveals fundamental differences in how models explain themselves.

CLIP models exhibit surprising reliance on latent components encoding polysemous words, visual typography, and dataset artifacts, revealing hidden biases that can be amplified in downstream tasks.