SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Models

TL;DR

This paper presents SAE-SSV, a supervised steering method in sparse, interpretable latent spaces of language models, enabling more reliable and targeted control over model behaviors with minimal quality loss.

Contribution

Introduces a novel supervised steering approach using sparse autoencoders and linear classifiers to control language model outputs in an interpretable, efficient manner.

Findings

Higher success rates in steering tasks

Minimal degradation in generation quality

Effective control with small subspace dimensions

Abstract

Large language models (LLMs) have demonstrated impressive capabilities in natural language understanding and generation, but controlling their behavior reliably remains challenging, especially in open-ended generation settings. This paper introduces a novel supervised steering approach that operates in sparse, interpretable representation spaces. We employ sparse autoencoders (SAEs) to obtain sparse latent representations that aim to disentangle semantic attributes from model activations. Then we train linear classifiers to identify a small subspace of task-relevant dimensions in latent representations. Finally, we learn supervised steering vectors constrained to this subspace, optimized to align with target behaviors. Experiments across sentiment, truthfulness, and political polarity steering tasks with multiple LLMs demonstrate that our supervised steering vectors achieve higher success rates with minimal degradation in generation quality compared to existing methods. Further analysis reveals that a notably small subspace is sufficient for effective steering, enabling more targeted and interpretable interventions. Our implementation is publicly available at https://github.com/Ineedanamehere/SAE-SSV.