Curveball Steering: The Right Direction To Steer Isn't Always Linear
This addresses the problem of inconsistent behavior in controlling LLMs for researchers and practitioners, offering a more effective steering approach, though it is incremental as it builds on existing activation steering methods.
The paper challenges the Linear Representation Hypothesis in activation steering for LLMs by showing that activation spaces have substantial geometric distortions, and introduces Curveball steering, a nonlinear method based on polynomial kernel PCA that consistently outperforms linear PCA-based steering, especially in high-distortion regimes.
Activation steering is a widely used approach for controlling large language model (LLM) behavior by intervening on internal representations. Existing methods largely rely on the Linear Representation Hypothesis, assuming behavioral attributes can be manipulated using global linear directions. In practice, however, such linear interventions often behave inconsistently. We question this assumption by analyzing the intrinsic geometry of LLM activation spaces. Measuring geometric distortion via the ratio of geodesic to Euclidean distances, we observe substantial and concept-dependent distortions, indicating that activation spaces are not well-approximated by a globally linear geometry. Motivated by this, we propose "Curveball steering", a nonlinear steering method based on polynomial kernel PCA that performs interventions in a feature space, better respecting the learned activation geometry. Curveball steering consistently outperforms linear PCA-based steering, particularly in regimes exhibiting strong geometric distortion, suggesting that geometry-aware, nonlinear steering provides a principled alternative to global, linear interventions.