Learning to Steer: Input-dependent Steering for Multimodal LLMs

TL;DR

This paper introduces L2S, a method for input-dependent steering of multimodal LLMs, which uses a learned auxiliary module to generate input-specific guidance, reducing hallucinations and improving safety.

Contribution

It proposes a novel input-specific steering technique for multimodal LLMs using a learned auxiliary module, addressing limitations of static steering methods.

Findings

Reduces hallucinations in MLLMs

Enforces safety in model responses

Outperforms static baseline steering methods

Abstract

Steering has emerged as a practical approach to enable post-hoc guidance of LLMs towards enforcing a specific behavior. However, it remains largely underexplored for multimodal LLMs (MLLMs); furthermore, existing steering techniques, such as mean steering, rely on a single steering vector, applied independently of the input query. This paradigm faces limitations when the desired behavior is dependent on the example at hand. For example, a safe answer may consist in abstaining from answering when asked for an illegal activity, or may point to external resources or consultation with an expert when asked about medical advice. In this paper, we investigate a fine-grained steering that uses an input-specific linear shift. This shift is computed using contrastive input-specific prompting. However, the input-specific prompts required for this approach are not known at test time. Therefore, we propose to train a small auxiliary module to predict the input-specific steering vector. Our approach, dubbed as L2S (Learn-to-Steer), demonstrates that it reduces hallucinations and enforces safety in MLLMs, outperforming other static baselines. Our code is publicly available at https://jayneelparekh.github.io/learn-to-steer/