English

Predicting Where Steering Vectors Succeed

Machine Learning 2026-04-20 v1 Computation and Language

Abstract

Steering vectors work for some concepts and layers but fail for others, and practitioners have no way to predict which setting applies before running an intervention. We introduce the Linear Accessibility Profile (LAP), a per-layer diagnostic that repurposes the logit lens as a predictor of steering vector effectiveness. The key measure, AlinA_{\mathrm{lin}}, applies the model's unembedding matrix to intermediate hidden states, requiring no training. Across 24 controlled binary concept families on five models (Pythia-2.8B to Llama-8B), peak AlinA_{\mathrm{lin}} predicts steering effectiveness at ρ=+0.86\rho = +0.86 to +0.91+0.91 and layer selection at ρ=+0.63\rho = +0.63 to +0.92+0.92. A three-regime framework explains when difference-of-means steering works, when nonlinear methods are needed, and when no method can work. An entity-steering demo confirms the prediction end-to-end: steering at the LAP-recommended layer redirects completions on Gemma-2-2B and OLMo-2-1B-Instruct, while the middle layer (the standard heuristic) has no effect on either model.

Keywords

Cite

@article{arxiv.2604.15557,
  title  = {Predicting Where Steering Vectors Succeed},
  author = {Jayadev Billa},
  journal= {arXiv preprint arXiv:2604.15557},
  year   = {2026}
}

Comments

19 pages, incl. 10 appendix pages, 4 figures, 20 tables

R2 v1 2026-07-01T12:13:37.113Z