INNSteer learns an invertible neural network to map LLM activations into a latent space where linear steering becomes more effective, then applies the inverse map to produce nonlinear interventions in the original space.
hub
A unified understanding and evaluation of steering methods.arXiv preprint arXiv:2502.02716
12 Pith papers cite this work. Polarity classification is still indexing.
12
Pith papers citing it
hub tools
verdicts
UNVERDICTED 12representative citing papers
First systematic test shows activation steering robustness drops sharply (up to 64%) under adversarial input perturbations across multiple extraction methods, models, and personas.
A geometric decomposition framework shows that affine transformations best recover prompt-induced task geometry and behavior in language and vision models across multiple datasets.
FishBack derives a closed-form minimum-distortion steering direction from the pullback Fisher metric of the softmax layer, outperforming Euclidean baselines on GPT-2 verb-morphology tasks with lower off-target KL divergence.
Prompt-boundary directional alignment enables geometry-guided search that cuts trials to 95% best utility by 39.8% on average, while concept granularity predicts remaining difficulty via directional heterogeneity.
Activation steering produces synthetic safety-violating data that improves downstream classifiers over prompting on most tested concepts when a harmonic mean of alignment, coherence, and diversity is optimized.
VSPO samples rollouts at varying steering intensities to improve behavioral control in LLMs while preserving task accuracy.
Steer-to-Detect learns a steering vector injected into LLM hidden states to boost class separability and applies hypothesis testing with finite-sample Type I/II error guarantees for generated-text detection.
LLMs perform in-context learning as trajectories through a structured low-dimensional conceptual belief space, with the structure visible in both behavior and internal representations and causally manipulable via interventions.
K-Steering uses a non-linear multi-label classifier on activations to compute gradient-based intervention directions for unified multi-attribute control in LLMs, outperforming linear baselines on ToneBank and DebateMix benchmarks across three model families.
SALSA adapts speech-aware LLMs via supervised layer-wise steering vectors, reporting up to 46.8% relative gains over zero-shot on out-of-domain speech benchmarks.
Prototype-Based Sparse Steering decomposes query activations with SAEs and optimizes sparse features via gradients to steer LLM outputs toward specific behaviors.
citing papers explorer
-
Decomposing how prompting steers behavior
A geometric decomposition framework shows that affine transformations best recover prompt-induced task geometry and behavior in language and vision models across multiple datasets.