method · first seen 4 Oct 2026
safety feature discovery in transcoder features
Also: CSFD
A filtering approach that identifies candidate safety-relevant features from transcoder features without exhaustive ablation.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Don't Judge an LLM Only by Its Activations: Discovering Suppressed Safety Features via Counterfactual Activation PotentialPapers · 4 Oct 2026 · Swadesh Swain, Sanghamitra Dutta
Latest sources
- Don't Judge an LLM Only by Its Activations: Discovering Suppressed Safety Features via Counterfactual Activation PotentialPapers · 4 Oct 2026 · Swadesh Swain, Sanghamitra Dutta