When is Your LLM Steerable?

• URL: http://arxiv.org/abs/2606.11599v1
• Date: Wed, 10 Jun 2026 02:55:34 GMT
• Title: When is Your LLM Steerable?
• Abstract summary: Activation steering offers a lightweight approach to control language models' behavior at inference time.<n>Finding the regime and boundaries of successful steering typically requires expensive grid searches.<n>We investigate whether steerability can be predicted from the model's internal states at the beginning of the generation process.
• Score: 56.656180566692946
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Activation steering offers a lightweight approach to control language models' behavior at inference time, but whether it succeeds or fails heavily depends on the prompt, concept, model, and steering configuration. Finding the regime and boundaries of successful steering typically requires expensive grid searches and post-hoc evaluation of full autoregressive rollouts. In this work, we investigate whether steerability can be predicted from the model's internal states at the beginning of the generation process, e.g., after generating the first few tokens, and how to leverage such a predictor to improve steering success rate. To this end, we first introduce ASTEER, a testbed including 1.4M steered generations, spanning 150 concepts with each steering success/failure labeled. Leveraging this testbed, we analyze the model's early decoding dynamics by extracting features that compare hidden states before and after steering across layers and initial decoding steps. These features help us understand how steering's effects propagate along layers and token positions, which provide key information for steerability prediction. We then train a Gradient Boosting Decision Trees (GBDT) classifier on these features to predict whether an intervention will under-steer, succeed, or over-steer without requiring full rollout. Our predictor achieves around 0.7 macro-F1 score on unseen concepts, demonstrating that early hidden states encode substantial, structured information about eventual steering efficacy. We further leverage this steerability predictor as guidance for steering strength searching, achieving near-optimal performance with a small fraction of decoding cost.

Related papers

• Predicting Future Behaviors in Reasoning Models Enables Better Steering [35.89008737242842]
  Test-time steering controls large reasoning models (LRMs) by intervening on their hidden representations, but it can degrade output quality.<n>We argue that prior steering work implicitly relies on internal features that detect behavior in already generated text.<n>We train activation probes to predict future behavior likelihoods from intermediate reasoning steps.<n>These probes predict the most likely behavior with 64%-91% accuracy, revealing a separate type of internal prediction features.
  arXiv  Detail & Related papers  (2026-06-09T17:49:24Z)
• Statistically Grounded Sparse-Feature Interventions for Activation-Space Control in Large Language Models [2.464003792743989]
  Activation steering offers a lightweight alternative to fine-tuning for behavioral control of large language models.<n>We introduce a transparent SAE-feature steering pipeline that ranks sparse features through an unweighted Borda consensus.<n>Across three Gemma-family models, four behavioral domains, and 356 layer-strength configurations, the method produces measurable domain-specific shifts.
  arXiv  Detail & Related papers  (2026-06-05T19:27:31Z)
• When Is Rank-1 Steering Cheap? Geometry, Granularity, and Budgeted Search [61.73232505100331]
  Activation steering offers a lightweight way to control LLMs without retraining, but its effectiveness varies sharply across concepts.<n>We formalize rank-1 steering as a budget-constrained optimization over intervention layer and coefficient.<n>We present textitGRACE, a Granularity- and Representation-Aware Concept Engineering framework that uses activation geometry to diagnose the dominant source of steering difficulty.
  arXiv  Detail & Related papers  (2026-05-09T14:26:49Z)
• Mechanistic Indicators of Steering Effectiveness in Large Language Models [3.635648354808971]
  Activation-based steering enables Large Language Models to exhibit targeted behaviors by intervening on intermediate activations without retraining.<n>Despite its widespread use, the mechanistic factors that govern when steering succeeds or fails remain poorly understood.<n>We investigate whether the reliability of steering can be diagnosed using internal model signals.
  arXiv  Detail & Related papers  (2026-02-02T06:56:22Z)
• KV Cache Steering for Controlling Frozen LLMs [80.50365534625438]
  cache steering is a lightweight method for implicit steering of language models.<n>We apply cache steering to induce chain-of-thought reasoning in small language models.
  arXiv  Detail & Related papers  (2025-07-11T17:59:36Z)
• Poutine: Vision-Language-Trajectory Pre-Training and Reinforcement Learning Post-Training Enable Robust End-to-End Autonomous Driving [19.48508500497233]
  We present Poutine, a method that uses an off-the-shelf vision-language model (VLM) to achieve robust end-to-end autonomous driving.<n>To learn strong base driving capabilities, we first train Poutine-Base using self-supervised next-token prediction over vision, language, and trajectory (VLT) tokens.<n>The final Poutine model achieves an RFS of 7.99 on the test set, placing 1st in the 2025 Vision-Based End-to-End Driving Challenge.
  arXiv  Detail & Related papers  (2025-06-12T19:14:00Z)
• ReCogDrive: A Reinforced Cognitive Framework for End-to-End Autonomous Driving [49.07731497951963]
  ReCogDrive is a novel Reinforced Cognitive framework for end-to-end autonomous driving.<n>We introduce a hierarchical data pipeline that mimics the sequential cognitive process of human drivers.<n>We then address the language-action mismatch by injecting the VLM's learned driving priors into a diffusion planner.
  arXiv  Detail & Related papers  (2025-06-09T03:14:04Z)
• Analyzing the Generalization and Reliability of Steering Vectors [8.253773195379166]
  We show that steering vectors have substantial limitations both in- and out-of-distribution.<n>In-distribution, steerability is highly variable across different inputs.<n>Out-of-distribution, while steering vectors often generalise well, for several concepts they are brittle to reasonable changes in the prompt.
  arXiv  Detail & Related papers  (2024-07-17T08:32:03Z)
• Think Twice before Driving: Towards Scalable Decoders for End-to-End Autonomous Driving [74.28510044056706]
  Existing methods usually adopt the decoupled encoder-decoder paradigm. In this work, we aim to alleviate the problem by two principles. We first predict a coarse-grained future position and action based on the encoder features. Then, conditioned on the position and action, the future scene is imagined to check the ramification if we drive accordingly.
  arXiv  Detail & Related papers  (2023-05-10T15:22:02Z)
• Control-Aware Prediction Objectives for Autonomous Driving [78.19515972466063]
  We present control-aware prediction objectives (CAPOs) to evaluate the downstream effect of predictions on control without requiring the planner be differentiable. We propose two types of importance weights that weight the predictive likelihood: one using an attention model between agents, and another based on control variation when exchanging predicted trajectories for ground truth trajectories.
  arXiv  Detail & Related papers  (2022-04-28T07:37:21Z)
• Value-driven Hindsight Modelling [68.658900923595]
  Value estimation is a critical component of the reinforcement learning (RL) paradigm. Model learning can make use of the rich transition structure present in sequences of observations, but this approach is usually not sensitive to the reward function. We develop an approach for representation learning in RL that sits in between these two extremes. This provides tractable prediction targets that are directly relevant for a task, and can thus accelerate learning the value function.
  arXiv  Detail & Related papers  (2020-02-19T18:10:20Z)

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.