Steer2Edit: From Activation Steering to Component-Level Editing

TL;DR

Steer2Edit introduces a training-free, component-level editing framework that transforms inference-time steering signals into interpretable weight modifications, improving safety, truthfulness, and reasoning efficiency in large language models.

Contribution

It provides a novel, theoretically grounded method to convert steering vectors into weight edits, enabling more precise and interpretable model behavior control without retraining.

Findings

Improves safety by up to 17.2%

Increases truthfulness by 9.8%

Reduces reasoning length by 12.2%

Abstract

Steering methods influence Large Language Model behavior by identifying semantic directions in hidden representations, but are typically realized through inference-time activation interventions that apply a fixed, global modification to the model's internal states. While effective, such interventions often induce unfavorable attribute-utility trade-offs under strong control, as they ignore the fact that many behaviors are governed by a small and heterogeneous subset of model components. We propose Steer2Edit, a theoretically grounded, training-free framework that transforms steering vectors from inference-time control signals into diagnostic signals for component-level rank-1 weight editing. Instead of uniformly injecting a steering direction during generation, Steer2Edit selectively redistributes behavioral influence across individual attention heads and MLP neurons, yielding interpretable edits that preserve the standard forward pass and remain compatible with optimized parallel inference. Across safety alignment, hallucination mitigation, and reasoning efficiency, Steer2Edit consistently achieves more favorable attribute-utility trade-offs: at matched downstream performance, it improves safety by up to 17.2%, increases truthfulness by 9.8%, and reduces reasoning length by 12.2% on average. Overall, Steer2Edit provides a principled bridge between representation steering and weight editing by translating steering signals into interpretable, training-free parameter updates. Our code is available at https://github.com/Trustworthy-ML-Lab/Steer2Edit