Refusal Beyond a Single Direction: A Preliminary Comparison of Diff-in-Means and INLP
For researchers working on interpretability and safety of LLMs, this work provides a preliminary comparison of two intervention methods, showing INLP counterfactual flipping as a competitive alternative to DiM, but the results are incremental and limited to five models.
The paper compares Diff-in-Means (DiM) and Iterative Nullspace Projection (INLP) interventions for steering refusal in safety fine-tuned chat models. INLP counterfactual flipping matches DiM directional ablation in suppressing refusal, while nullspace projection is weaker; restricting INLP to leading directions preserves suppression with near-baseline perplexity.
Arditi et al. (2024) has shown that refusal in safety fine-tuned chat models is mediated by a single linear direction in the residual stream, recoverable by a difference-in-means (DiM) of harmful and harmless activations. We compare DiM-based interventions (activation addition and directional ablation) with two interventions derived from Iterative Nullspace Projection (INLP) -- nullspace projection and counterfactual flipping -- on five open-weight chat models, asking whether INLP can match DiM at steering refusal and whether its richer parameterisation yields more tweakable interventions. INLP counterfactual flipping is competitive with DiM directional ablation on refusal suppression, while nullspace projection is consistently weaker. Restricting INLP to the leading directions of the extracted subspace preserves most of the suppression effect at near-baseline perplexity, giving a tunable capability. Geometrically, the two INLP interventions land in qualitatively different regions of activation space: nullspace projection collapses transformed activations \emph{between} the harmful and harmless clusters, while counterfactual flipping moves them into the opposite cluster, suggesting that the model encodes the absence of a concept differently from its opposite -- an intriguing distinction that warrants further investigation in future work.