mlsanityData to 14 Oct 2026

Spotlight Your Instructions: Instruction-following with Dynamic Attention Steering

arXiv 2505.12025 ↗First posted 17 May 2025cs.LGcited by 21

Not on the map. Why?

Abstract (arXiv)

In many real-world applications, users rely on natural language instructions to guide large language models (LLMs) across a wide range of tasks. These instructions are often complex, diverse, and subject to frequent change. However, LLMs do not always attend to these instructions reliably, and users lack simple mechanisms to emphasize their importance beyond modifying prompt wording or structure. To address this, we present an inference-time method that enables users to emphasize specific parts of their prompt by steering the model's attention toward them, aligning the model's perceived importance of different prompt tokens with user intent. Unlike prior approaches that are limited to static instructions, require significant offline profiling, or rely on fixed biases, we dynamically update the proportion of model attention given to the user-specified parts--ensuring improved instruction following without performance degradation. We demonstrate that our approach improves instruction following across a variety of tasks involving multiple instructions and generalizes across models of varying scales.

Key facts. Cited by 21 papers in the map (data through 14 Oct 2026). Builds on 52 of its 62 references that are in the map. First posted on arXiv 17 May 2025.

Builds on

52 of 62 references.

References · 52

  1. 2017Attention Is All You Needcited by 60,361
  2. 2022Training language models to follow instructions with human feedbackcited by 15,971
  3. 2024The Llama 3 Herd of Modelscited by 14,715
  4. 2024Qwen2.5 Technical Reportcited by 4,942
  5. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,780
  6. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,754
  7. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,720
  8. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,455
  9. 2022Scaling Instruction-Finetuned Language Modelscited by 3,024
  10. 2024Mixtral of Expertscited by 1,823
  11. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,651
  12. 2023A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questionscited by 1,601
38 more
  1. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  2. 2023Instruction-Following Evaluation for Large Language Modelscited by 1,228
  3. 2023Instruction Tuning with GPT-4cited by 686
  4. 2023Instruction Tuning for Large Language Models: A Surveycited by 623
  5. 2024The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructionscited by 493
  6. 2022ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detectioncited by 422
  7. 2024StruQ: Defending Against Prompt Injection with Structured Queriescited by 375
  8. 2023Evaluating Large Language Models at Evaluating Instruction Followingcited by 302
  9. 2024WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Modelscited by 223
  10. 2024Programming Refusal with Conditional Activation Steeringcited by 206
  11. 2023Efficient Guided Generation for Large Language Modelscited by 200
  12. 2024Improving Instruction-Following in Language Models through Activation Steeringcited by 160
  13. 2024Benchmarking Complex Instruction-Following with Multiple Constraints Compositioncited by 152
  14. 2024InFoBench: Evaluating Instruction Following Ability in Large Language Modelscited by 135
  15. 2023Training Language Models with Language Feedback at Scalecited by 128
  16. 2023Do LLMs Understand User Preferences? Evaluating LLMs On User Rating Predictioncited by 116
  17. 2024The Art of Saying No: Contextual Noncompliance in Language Modelscited by 105
  18. 2024Guiding LLMs The Right Way: Fast, Non-Invasive Constrained Generationcited by 101
  19. 2024Found in the Middle: Calibrating Positional Attention Bias Improves Long Context Utilizationcited by 101
  20. 2023Tell Your Model Where to Attend: Post-hoc Attention Steering for LLMscited by 91
  21. 2024FOFO: A Benchmark to Evaluate LLMs' Format-Following Capabilitycited by 80
  22. 2023Contrastive Decoding Improves Reasoning in Large Language Modelscited by 74
  23. 2023Evaluating Large Language Models on Controlled Generation Taskscited by 73
  24. 2024Attention Heads of Large Language Models: A Surveycited by 67
  25. 2023Explaining black box text modules in natural language with language modelscited by 62
  26. 2023Focus on Your Instruction: Fine-grained and Multi-instruction Image Editing by Attention Modulationcited by 62
  27. 2023When "A Helpful Assistant" Is Not Really Helpful: Personas in System Prompts Do Not Improve Performances of Large Language Modelscited by 53
  28. 2024Measuring and Controlling Instruction (In)Stability in Language Model Dialogscited by 53
  29. 2024Granite-Function Calling Model: Introducing Function Calling Abilities via Multi-task Learning of Granular Taskscited by 52
  30. 2024A Language Model's Guide Through Latent Spacecited by 43
  31. 2024Extending Activation Steering to Broad Skills and Multiple Behaviourscited by 40
  32. 2023AtMan: Understanding Transformer Predictions Through Memory Efficient Attention Manipulationcited by 31
  33. 2024Do LLMs "know" internally when they follow instructions?cited by 26
  34. 2024Attend First, Consolidate Later: On the Importance of Attention in Different LLM Layerscited by 25
  35. 2024Context Steering: Controllable Personalization at Inference Timecited by 22
  36. 2023EchoPrompt: Instructing the Model to Rephrase Queries for Improved In-context Learningcited by 13
  37. 2021On-the-Fly Attention Modulation for Neural Generationcited by 12
  38. 2024RefuteBench: Evaluating Refuting Instruction-Following for Large Language Modelscited by 11

50 most cited of 52 All 52 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 6; 2026: 15.

21 papers.

Sort:

By area:

  1. 2025ReasonIF: Large Reasoning Models Fail to Follow Instructions During Reasoningcited by 13
  2. 2026Enhancing Instruction Following of LLMs via Activation Steering with Dynamic Rejectioncited by 6
  3. 2025When Instructions Multiply: Measuring and Estimating LLM Capabilities of Multiple Instructions Followingcited by 5
  4. 2026Improving Chain-of-Thought for Logical Reasoning via Attention-Aware Interventioncited by 3
  5. 2026Looking Is Not Picking: An Attention-Segment Account of Tool-Selection Failures in LLM Agentscited by 3
  6. 2026PRISM-Delta: Differential Subspace Steering for Prompt Highlighting in Large Language Modelscited by 2
  7. 2026Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Rewardcited by 2
  8. 2026Memory Inception: Latent-Space KV Cache Manipulation for Steering LLMscited by 2
  9. 2026Steering Instruction Hierarchies at Inference Timecited by 2
  10. 2025Eyes-on-Me: Scalable RAG Poisoning through Transferable Attention-Steering Attractorscited by 1
  11. 2026Inference-Time Attention Steering for Vision-Language-Action Driving Modelscited by 1
  12. 2025Enhancing Trustworthy Clinical Diagnosis Decision-Making in Large Language Models via Etiology-Aware Attention Supervisioncited by 0
9 more
  1. 2025Context-Aware Hierarchical Learning: A Two-Step Paradigm towards Safer LLMscited by 0
  2. 2025Financial Instruction Following Evaluation (FIFE)cited by 0
  3. 2026Steer-to-Detect: Probing Hidden Representations for Detection of LLM-Generated Textscited by 0
  4. 2026Steered Generation via Gradient-Based Optimization on Sparse Query Featurescited by 0
  5. 2026Prompt Coverage Adequacycited by 0
  6. 2026Attention is Case-Sensitivecited by 0
  7. 2026The Safety Operator: Modulating the Expression of Safety Instructions via Spectral Optimizationcited by 0
  8. 2026Suppressing Pressure, Amplifying Evidence: Self-Guided Attention Steering to Mitigate Sycophancy and Stubbornnesscited by 0
  9. 2026Lend Me Your Eyes: Instruction-Aware Text Embeddings via Attention Relaycited by 0