Towards Understanding Steering Strength
This work addresses a critical but underexplored aspect of post-training control for LLMs, providing foundational insights that could improve steering techniques across various applications.
The authors tackled the problem of determining the optimal magnitude for steering latent representations in large language models, revealing non-monotonic effects and deriving qualitative laws that govern key metrics like next token probability and cross-entropy.
A popular approach to post-training control of large language models (LLMs) is the steering of intermediate latent representations. Namely, identify a well-chosen direction depending on the task at hand and perturbs representations along this direction at inference time. While many propositions exist to pick this direction, considerably less is understood about how to choose the magnitude of the move, whereas its importance is clear: too little and the intended behavior does not emerge, too much and the model's performance degrades beyond repair. In this work, we propose the first theoretical analysis of steering strength. We characterize its effect on next token probability, presence of a concept, and cross-entropy, deriving precise qualitative laws governing these quantities. Our analysis reveals surprising behaviors, including non-monotonic effects of steering strength. We validate our theoretical predictions empirically on eleven language models, ranging from a small GPT architecture to modern models.