Prompt highlighting steers a large language model to prioritize user-specified text spans during generation. A key challenge of existing Key-editing approaches is extracting steering directions that capture the difference between relevant and irrelevant contexts, rather than shared structural patterns common to both. We propose PRISM-$Δ$ (Projection-based Relevance-Informed Steering Method), which decomposes the difference between positive and negative cross-covariance matrices to maximize discriminative energy while eliminating shared directions. Each attention head receives a continuous softplus importance weight, letting weak-but-useful heads contribute at reduced strength. The framework extends naturally to Value representations, capturing content-channel signal that Key-only methods leave unused. Across four benchmarks and five models, our methods match or exceed the best existing method on 19 of 20 configurations, with relative gains up to +13.0%, while halving the fluency cost of steering. Our methods also scale to long-context retrieval, outperforming the best existing method by up to +5.1% relative gain. PRISM-$Δ$ is compatible with FlashAttention and adds negligible memory overhead.
Steering Language Generation: Harnessing Contrastive Expert Guidance and Negative Prompting for Coherent and Diverse Synthetic Data Generation
Mathematical Sciences Institute, Australian National University(澳大利亚国立大学数学科学研究所);Research School of Astronomy & Astrophysics, Australian National University(澳大利亚国立大学天文与天体物理研究学院);School of Computing, Australian National University(澳大利亚国立大学计算学院);Department of Astronomy, The Ohio State University(俄亥俄州立大学天文学系)
STEER-BENCH: A Benchmark for Evaluating the Steerability of Large Language Models
Department of Computer Science, University of Southern California(计算机科学系,南加州大学);Information Sciences Institute, University of Southern California(信息科学研究所,南加州大学)
When Do Language Models Endorse Limitations on Human Rights Principles?
Max Planck Institute for Intelligent Systems(马克斯·普朗克智能系统研究所);Jinesis AI Lab, University of Toronto & Vector Institute(Jinesis AI实验室,多伦多大学及向量研究所);University of Michigan(密歇根大学);University of Copenhagen(哥本哈根大学);EuroSafeAI