Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics

TL;DR

This paper unifies various language model control methods into a single framework, analyzing their effects on preference and utility, and introduces a new approach called SPLIT that balances these aspects.

Contribution

It provides a unified conceptual framework for different control techniques and proposes a novel method, SPLIT, to improve preference while maintaining utility.

Findings

Control methods trade off preference and utility predictably.

Control shifts representations along target directions, affecting model behavior.

SPLIT improves preference with better utility preservation.

Abstract

Methods for controlling large language models (LLMs), including local weight fine-tuning, LoRA-based adaptation, and activation-based interventions, are often studied in isolation, obscuring their connections and making comparison difficult. In this work, we present a unified view that frames these interventions as dynamic weight updates induced by a control signal, placing them within a single conceptual framework. Building on this view, we propose a unified preference-utility analysis that separates control effects into preference, defined as the tendency toward a target concept, and utility, defined as coherent and task-valid generation, and measures both on a shared log-odds scale using polarity-paired contrastive examples. Across methods, we observe a consistent trade-off between preference and utility: stronger control increases preference while predictably reducing utility. We further explain this behavior through an activation manifold perspective, in which control shifts representations along target-concept directions to enhance preference, while utility declines primarily when interventions push representations off the model's valid-generation manifold. Finally, we introduce a new steering approach SPLIT guided by this analysis that improves preference while better preserving utility. Code is available at https://github.com/zjunlp/EasyEdit/blob/main/examples/SPLIT.md.