Token-level steering has emerged as a pivotal approach for inference-time alignment, enabling fine-grained control over large language models (LLMs) by modulating their output distributions without parameter updates. While effective, existing methods rely on dense intervention at every decoding step. This persistent manipulation not only incurs substantial computational overhead but also risks compromising generation quality by excessively drifting from the model's intrinsic distribution. In this work, we show that dense intervention is unnecessary and propose Sparse Inference-time Alignment (SIA), which performs sparse junction steering, intervening only at critical decision points along the generation trajectory. Our key insight is that high-entropy junctions are empirically consistent with pivotal decision points in the generation trajectory and are particularly susceptible to misalignment, suggesting that these points benefit from alignment-related reward signal. Extensive experiments across different model families and alignment objectives show that steering only 20%-80% of tokens achieves the superior alignment-efficiency trade-offs. For strong base models such as Qwen3, intervening on as few as 20% of tokens matches or even surpasses heavily post-trained instruct models. This sparsity enables stronger guidance while better preserving the model's native distribution, integrates seamlessly with search-based methods (e.g., Best-of-N), and improves efficiency by reducing the required sampling budget while yielding practical wall-clock speedups under matched-quality comparisons.
On the Limitations of Steering in Language Model Alignment
Birla Institute of Technology and Science Pilani(比拉理工科学与 Pilani 学院);National University of Singapore(新加坡国立大学);Institute for Data Science, National University of Singapore(新加坡国立大学数据科学研究院)
CoSToM:Causal-oriented Steering for Intrinsic Theory-of-Mind Alignment in Large Language Models
National Engineering Research Center for Big Data Technology and System, Services Computing Technology and System Lab, Cluster and Grid Computing Lab, Huazhong University of Science and Technology(华中科技大学国家大数据技术与系统工程技术研究中心、服务计算技术与系统实验室、集群与网格计算实验室);Singapore Management University(新加坡管理大学)
ODESteer: A Unified ODE-Based Steering Framework for LLM Alignment
University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);Northwestern University(西北大学);William & Mary(威廉与玛丽学院);University of Washington(华盛顿大学);Stanford University(斯坦福大学)
Adaptive Multi-Subspace Representation Steering for Attribute Alignment in Large Language Models
Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)(莫扎德·本·泽德人工智能大学);Data Analytics (PRADA) Lab(数据分析(PRADA)实验室);King Abdullah University of Science(卡瓦尔大学);Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai, China(上海先进研究院,中国科学院,上海,中国);University of Chinese Academy of Sciences, Beijing, China(中国科学院大学,北京,中国);University of Copenhagen, Copenhagen, Denmark(哥本哈根大学,哥本哈根,丹麦);University of Science(科学大学)
Beyond Token-Level Guidance: Inference-Time Alignment of Specialized LLMs via Cross-Family Representation Steering
College of Computing and Data Science, Nanyang Technological University(南洋理工大学计算与数据科学学院);College of Cryptology and Cyber Science, Nankai University(南开大学密码与网络科学学院)
VISPA: Pluralistic Alignment via Automatic Value Selection and Activation
University of Waterloo(滑铁卢大学);University of Melbourne(墨尔本大学);University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);MBZUAI(马克斯·普朗克智能系统研究所);Macquarie University(麦考瑞大学)
Agenda-based Narrative Extraction: Steering Pathfinding Algorithms with Large Language Models
Department of Computing and Systems Engineering, Universidad Católica del Norte, Antofagasta, Chile(智利天主教大学北部校区计算与系统工程系,安托法加斯塔,智利);Department of Industrial Engineering, Universidad Católica del Norte, Antofagasta, Chile(智利天主教大学北部校区工业工程系,安托法加斯塔,智利);Department of Chemical and Environmental Engineering, Universidad Católica del Norte, Antofagasta, Chile(智利天主教大学北部校区化学与环境工程系,安托法加斯塔,智利)
PIXEL: Adaptive Steering Via Position-wise Injection with eXact Estimated Levels under Subspace Calibration
University of Queensland(昆士兰大学);Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)(穆罕默德·本·扎耶德人工智能大学);Provable Responsible AI and Data Analytics (PRADA) Lab(可证责任AI与数据分析师实验室);King Abdullah University of Science and Technology (KAUST)(国王 Abdullah 科学与技术大学)
Minimizing Collateral Damage in Activation Steering
Department of Electrical \& Computer Engineering, Rice University, Houston, USA;Department of Computational;Applied Mathematics, Rice University, Houston, USA;Department of Electrical \& Computer Engineering, The University of Texas at Austin, Austin, USA