Where to Steer: Input-Dependent Layer Selection for Steering Improves LLM Alignment

• URL: http://arxiv.org/abs/2604.03867v1
• Date: Sat, 04 Apr 2026 21:16:47 GMT
• Title: Where to Steer: Input-Dependent Layer Selection for Steering Improves LLM Alignment
• Abstract summary: We show that different inputs can require steering at different layers to achieve alignment with a desirable model behavior.<n>We introduce Where to Steer, a framework that adaptively selects the intervention layer conditioned on the input, by learning a mapping from input embeddings to optimal steering layers.
• Score: 17.001194019333052
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Steering vectors have emerged as a lightweight and effective approach for aligning large language models (LLMs) at inference time, enabling modulation over model behaviors by shifting LLM representations towards a target behavior. However, existing methods typically apply steering vectors at a globally fixed layer, implicitly assuming that the optimal intervention layer is invariant across inputs. We argue that this assumption is fundamentally limited, as representations relevant to a target behavior can be encoded at different layers depending on the input. Theoretically, we show that different inputs can require steering at different layers to achieve alignment with a desirable model behavior. We also provide empirical evidence that the optimal steering layer varies substantially across inputs in practice. Motivated by these observations, we introduce Where to Steer (W2S), a framework that adaptively selects the intervention layer conditioned on the input, by learning a mapping from input embeddings to optimal steering layers. Across multiple LLMs and alignment behaviors, W2S consistently outperforms fixed-layer baselines, with improvements in both in-distribution and out-of-distribution settings. Our findings highlight the importance of input-dependent control in LLM alignment and demonstrate that adaptive layer selection is a key design dimension missing in the current methodology of steering vectors.

Related papers

• Beyond Linear Activation Steering: Invertible Latent Transformations for Controlling LLM Behavior [13.073472989807675]
  INNSteer is a nonlinear activation steering framework based on invertible latent transformations.<n>It consistently improves model control over linear, transport-based, and nonlinear steering baselines.
  arXiv  Detail & Related papers  (2026-06-07T05:01:25Z)
• Local Linearity of LLMs Enables Activation Steering via Model-Based Linear Optimal Control [2.752817022620644]
  Inference-time LLM alignment methods offer an alternative to fine-tuning by directly modifying activations during generation.<n>We show that layer-wise dynamics across multiple LLM architectures and scales are well-approximated by locally-linear models.<n>We derive theoretical bounds on setpoint tracking error, enabling formal guarantees on steering performance.
  arXiv  Detail & Related papers  (2026-04-21T03:09:46Z)
• ODESteer: A Unified ODE-Based Steering Framework for LLM Alignment [49.68063561145927]
  We propose a unified ordinary differential equations (ODEs)-based theoretical framework for activation steering.<n>We introduce ODESteer, a kind of ODE-based steering guided by barrier functions.<n>Compared to state-of-the-art activation steering methods, ODESteer achieves consistent empirical improvements.
  arXiv  Detail & Related papers  (2026-02-19T17:13:44Z)
• Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics [81.80010043113445]
  Local weight fine-tuning, LoRA-based adaptation, and activation-based interventions are studied in isolation.<n>We present a unified view that frames these interventions as dynamic weight updates induced by a control signal.<n>Across methods, we observe a consistent trade-off between preference and utility: stronger control increases preference while predictably reducing utility.
  arXiv  Detail & Related papers  (2026-02-02T17:04:36Z)
• GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs [56.93583799109029]
  GrAInS is an inference-time steering approach that operates across both language-only and vision-language models and tasks.<n>During inference, GrAInS hidden activations at transformer layers guided by token-level attribution signals, and normalizes activations to preserve representational scale.<n>It consistently outperforms both fine-tuning and existing steering baselines.
  arXiv  Detail & Related papers  (2025-07-24T02:34:13Z)
• REAL: Reading Out Transformer Activations for Precise Localization in Language Model Steering [26.428347164111926]
  Inference-time steering aims to alter a large language model's responses without changing its parameters.<n>Existing approaches often rely on simplistic cues or ad hoc generalizations.<n>We introduce REAL, a framework for identifying behavior-relevant modules in Transformer models.
  arXiv  Detail & Related papers  (2025-06-10T02:16:50Z)
• Leveraging Importance Sampling to Detach Alignment Modules from Large Language Models [48.15777554876988]
  Traditional alignment methods often require retraining large pretrained models.<n>We propose a novel textitResidual Alignment Model (textitRAM) that formalizes the alignment process as a type of importance sampling.<n>We develop a resampling algorithm with iterative token-level decoding to address the common first-token latency issue in comparable methods.
  arXiv  Detail & Related papers  (2025-05-26T08:53:02Z)
• Guiding Giants: Lightweight Controllers for Weighted Activation Steering in LLMs [8.085475675888045]
  Activation steering provides an alternative for inference-time control.<n>We introduce a novel approach using a lightweight, trainable controller network integrated during inference.
  arXiv  Detail & Related papers  (2025-05-22T01:48:38Z)
• Analyzing Finetuning Representation Shift for Multimodal LLMs Steering [56.710375516257876]
  We propose to map hidden states to interpretable visual and textual concepts.<n>This enables us to more efficiently compare certain semantic dynamics, such as the shift from an original and fine-tuned model.<n>We also demonstrate the use of shift vectors to capture these concepts changes.
  arXiv  Detail & Related papers  (2025-01-06T13:37:13Z)
• Personalized Steering of Large Language Models: Versatile Steering Vectors Through Bi-directional Preference Optimization [34.05163996072159]
  "steering vectors" are extracted from the activations of human preference data. This work proposes an innovative approach that could produce more effective steering vectors through bi-directional preference optimization. Our method is designed to allow steering vectors to directly influence the generation probability of contrastive human preference data pairs.
  arXiv  Detail & Related papers  (2024-05-28T05:10:40Z)
• Semi-supervised Domain Adaptive Structure Learning [72.01544419893628]
  Semi-supervised domain adaptation (SSDA) is a challenging problem requiring methods to overcome both 1) overfitting towards poorly annotated data and 2) distribution shift across domains. We introduce an adaptive structure learning method to regularize the cooperation of SSL and DA.
  arXiv  Detail & Related papers  (2021-12-12T06:11:16Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.