Sparse autoencoders isolate unstable features in reward model representations and enable two mitigation techniques that reduce preference errors on perturbed inputs without retraining.
Refusal in language models is mediated by a single direction.Advances in Neural Information Processing Systems, 37:136037–136083
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
years
2026 3roles
background 1polarities
background 1representative citing papers
Injecting brief safety-plausible phrases into robot audio triggers LLM safety halts, enabling semantic denial-of-service attacks where prompt defenses trade attack suppression for impaired genuine hazard detection.
RET learns temporally consistent macrovariables from LLM activations via self-supervised learning to support interpretability, early behavioral prediction, and causal intervention.
citing papers explorer
-
Preference Instability in Reward Models: Detection and Mitigation via Sparse Autoencoders
Sparse autoencoders isolate unstable features in reward model representations and enable two mitigation techniques that reduce preference errors on perturbed inputs without retraining.
-
Semantic Denial of Service in LLM-controlled robots
Injecting brief safety-plausible phrases into robot audio triggers LLM safety halts, enabling semantic denial-of-service attacks where prompt defenses trade attack suppression for impaired genuine hazard detection.
-
Towards Effective Theory of LLMs: A Representation Learning Approach
RET learns temporally consistent macrovariables from LLM activations via self-supervised learning to support interpretability, early behavioral prediction, and causal intervention.