SAE-SSV: Supervised Steering in Sparse Representation Spaces for Reliable Control of Language Models
This addresses the problem of controlling LLM behavior for users in open-ended generation settings, representing an incremental improvement over existing methods.
The paper tackles the challenge of reliably controlling large language models in open-ended generation by introducing a supervised steering approach that operates in sparse, interpretable representation spaces, achieving higher success rates with minimal degradation in generation quality across tasks like sentiment, truthfulness, and politics polarity steering.
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 politics 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.