Refusal in language models is mediated by a single direction in residual stream activations that can be erased to disable safety or added to elicit refusal.
Tradeoffs between alignment and helpfulness in language models
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
fields
cs.LG 3representative citing papers
Value-filtered decoding steers LLM outputs for safety at decoding time using a value criterion with an explicit bound on false interventions controlled by one threshold hyperparameter.
Causal localization via attribution and patching identifies a temporal preference subgraph in mid-to-upper layers of Qwen3-4B-Instruct-2507, with time-horizon geometry in the residual stream and initial evidence for steering-vector control.
citing papers explorer
-
Refusal in Language Models Is Mediated by a Single Direction
Refusal in language models is mediated by a single direction in residual stream activations that can be erased to disable safety or added to elicit refusal.
-
Selective Safety Steering via Value-Filtered Decoding
Value-filtered decoding steers LLM outputs for safety at decoding time using a value criterion with an explicit bound on false interventions controlled by one threshold hyperparameter.
-
Temporal Preference Concepts and their Functions in a Large Language Model
Causal localization via attribution and patching identifies a temporal preference subgraph in mid-to-upper layers of Qwen3-4B-Instruct-2507, with time-horizon geometry in the residual stream and initial evidence for steering-vector control.