Curveball Steering: The Right Direction To Steer Isn't Always Linear
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866917356161204224 |
|---|---|
| author | Raval, Shivam Song, Hae Jin Wu, Linlin Harrasse, Abir Phillips, Jeff M. Barez, Fazl Abdullah, Amirali |
| author_facet | Raval, Shivam Song, Hae Jin Wu, Linlin Harrasse, Abir Phillips, Jeff M. Barez, Fazl Abdullah, Amirali |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_09313 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Curveball Steering: The Right Direction To Steer Isn't Always Linear Raval, Shivam Song, Hae Jin Wu, Linlin Harrasse, Abir Phillips, Jeff M. Barez, Fazl Abdullah, Amirali Artificial Intelligence I.2.6; I.2.7 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. |
| title | Curveball Steering: The Right Direction To Steer Isn't Always Linear |
| topic | Artificial Intelligence I.2.6; I.2.7 |
| url | https://arxiv.org/abs/2603.09313 |