Interpretable LLM Guardrails via Sparse Representation Steering
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866909901260849152 |
|---|---|
| author | He, Zeqing Wang, Zhibo Xu, Huiyu Lin, Hejun Zhang, Wenhui Chu, Zhixuan |
| author_facet | He, Zeqing Wang, Zhibo Xu, Huiyu Lin, Hejun Zhang, Wenhui Chu, Zhixuan |
| contents | Large language models (LLMs) exhibit impressive capabilities in generation tasks but are prone to producing harmful, misleading, or biased content, posing significant ethical and safety concerns. To mitigate such risks, representation engineering, which steer model behavior toward desired attributes by injecting carefully designed steering vectors into LLM's representations at inference time, has emerged as a promising alternative to fine-tuning approaches. However, due to the semantically entangled nature of LLM's representation, existing representation engineering methods still suffer from several limitations: limited fine-grained controllability, content quality degradation, and conflict in multi-attribute control. To overcome these challenges, we propose Sparse Representation Steering (SRS), a novel framework that achieves fine-grained and interpretable control over LLM behavior by first disentangling internal activations into a sparse, semantically meaningful representation space, and then selectively steering relevant dimensions. Specifically, SRS leverages a pretrained Sparse Autoencoder (SAE) to transform dense, entangled activation patterns into a sparse monosemantic feature space. To identify relevant features, SRS contrasts sparse activations from positive and negative prompt pairs and measures their bidirectional KL divergence to locate dimensions most associated with the target attribute. We conduct comprehensive experiments on Gemma-2 series model across three alignment dimensions, i.e., safety, fairness, and truthfulness, to evaluate the effectiveness of SRS. Results show that SRS consistently outperforms existing steering methods, which achieves significantly improved controllability across both single and multiple attribute settings, while preserving high linguistic quality and general ability. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_16851 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Interpretable LLM Guardrails via Sparse Representation Steering He, Zeqing Wang, Zhibo Xu, Huiyu Lin, Hejun Zhang, Wenhui Chu, Zhixuan Cryptography and Security Computation and Language Large language models (LLMs) exhibit impressive capabilities in generation tasks but are prone to producing harmful, misleading, or biased content, posing significant ethical and safety concerns. To mitigate such risks, representation engineering, which steer model behavior toward desired attributes by injecting carefully designed steering vectors into LLM's representations at inference time, has emerged as a promising alternative to fine-tuning approaches. However, due to the semantically entangled nature of LLM's representation, existing representation engineering methods still suffer from several limitations: limited fine-grained controllability, content quality degradation, and conflict in multi-attribute control. To overcome these challenges, we propose Sparse Representation Steering (SRS), a novel framework that achieves fine-grained and interpretable control over LLM behavior by first disentangling internal activations into a sparse, semantically meaningful representation space, and then selectively steering relevant dimensions. Specifically, SRS leverages a pretrained Sparse Autoencoder (SAE) to transform dense, entangled activation patterns into a sparse monosemantic feature space. To identify relevant features, SRS contrasts sparse activations from positive and negative prompt pairs and measures their bidirectional KL divergence to locate dimensions most associated with the target attribute. We conduct comprehensive experiments on Gemma-2 series model across three alignment dimensions, i.e., safety, fairness, and truthfulness, to evaluate the effectiveness of SRS. Results show that SRS consistently outperforms existing steering methods, which achieves significantly improved controllability across both single and multiple attribute settings, while preserving high linguistic quality and general ability. |
| title | Interpretable LLM Guardrails via Sparse Representation Steering |
| topic | Cryptography and Security Computation and Language |
| url | https://arxiv.org/abs/2503.16851 |