StyliTruth separates style and truth directions in the activations of selected attention heads, then steers each token along the disentangled subspaces to preserve truthfulness during stylization.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Balancing Stylization and Truth via Disentangled Representation Steering
StyliTruth separates style and truth directions in the activations of selected attention heads, then steers each token along the disentangled subspaces to preserve truthfulness during stylization.