What Can We Actually Steer? A Multi-Behavior Study of Activation Control
This provides empirically grounded guidance for implementing activation steering in LLMs, addressing the need for precise behavior control in safe deployment, though it is incremental in analyzing existing methods.
The study investigated how activation steering effectiveness varies across 50 different behavior types in large language models, finding that effectiveness differs significantly by behavior type, with trait expression showing an inverted-U curve in response to steering strength.
Large language models (LLMs) require precise behavior control for safe and effective deployment across diverse applications. Activation steering offers a promising approach for LLMs' behavioral control. We focus on the question of how steering effectiveness varies across different behavior types and whether the nature of target behaviors can predict steering success. We address this through empirical analysis of activation steering across 50 behaviors that span persona archetypes, personality traits, misalignment behaviors, style cues, and impersonation of public figures. We present a set of comprehensive experiments on coefficient optimization, vector properties, and data requirements to provide comprehensive guidance for the implementation of activation steering. Our analysis demonstrates that steering effectiveness varies significantly by behavior type, with different behavioral categories exhibiting distinct response patterns to intervention strength. We find that trait expression follows an inverted-U curve with a steering coefficient strength. We also show that vector separation metrics do not predict steering success, but larger training datasets enable more aggressive steering. These findings provide empirically grounded guidance for implementing activation steering and demonstrate that steering effectiveness is heavily influenced by behavior type.