mlsanityData to 14 Oct 2026

Understanding Reasoning in Thinking Language Models via Steering Vectors

arXiv 2506.18167 ↗First posted 22 Jun 2025cs.LG cs.AIcited by 72

On the map: LLM reasoning → Test-time compute and efficient thinking

also in 1 more
Abstract (arXiv)

Recent advances in large language models (LLMs) have led to the development of thinking language models that generate extensive internal reasoning chains before producing responses. While these models achieve improved performance, controlling their reasoning processes remains challenging. This work presents a steering approach for thinking LLMs by analyzing and manipulating specific reasoning behaviors in DeepSeek-R1-Distill models. Through a systematic experiment on 500 tasks across 10 diverse categories, we identify several reasoning behaviors exhibited by thinking models, including expressing uncertainty, generating examples for hypothesis validation, and backtracking in reasoning chains. We demonstrate that these behaviors are mediated by linear directions in the model's activation space and can be controlled using steering vectors. By extracting and applying these vectors, we provide a method to modulate specific aspects of the model's reasoning process, such as its tendency to backtrack or express uncertainty. Our approach offers practical tools for steering reasoning processes in thinking models in a controlled and interpretable manner. We validate our steering method using three DeepSeek-R1-Distill models, demonstrating consistent control across different model architectures.

Key facts. Cited by 72 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 14 of its 22 references that are in the map. First posted on arXiv 22 Jun 2025.

Builds on

14 of 22 references.

References · 14

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  2. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  3. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  4. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  5. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  6. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  7. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  8. 2023Steering Language Models With Activation Engineeringcited by 922
  9. 2021Show Your Work: Scratchpads for Intermediate Computation with Language Modelscited by 886
  10. 2021Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigmcited by 606
  11. 2022Extracting Latent Steering Vectors from Pretrained Language Modelscited by 225
  12. 2023Attribution Patching Outperforms Automated Circuit Discoverycited by 209
2 more
  1. 2024How to think step-by-step: A mechanistic understanding of chain-of-thought reasoningcited by 53
  2. 2025Activation Control for Efficiently Eliciting Long Chain-of-thought Ability of Language Modelscited by 13

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 26; 2026: 46.

72 papers.

Sort:

By area:

  1. 2025Thought Anchors: Which LLM Reasoning Steps Matter?cited by 129
  2. 2026Reasoning Theater: Disentangling Model Beliefs from Chain-of-Thoughtcited by 40
  3. 2025RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMscited by 34
  4. 2025Base Models Know How to Reason, Thinking Models Learn Whencited by 27
  5. 2026Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Modelscited by 27
  6. 2025Reasoning-Finetuning Repurposes Latent Representations in Base Modelscited by 26
  7. 2025Attention Illuminates LLM Reasoning: The Preplan-and-Anchor Rhythm Enables Fine-Grained Policy Optimizationcited by 25
  8. 2025ThinKV: Thought-Adaptive KV Cache Compression for Efficient Reasoning Modelscited by 21
  9. 2025The Geometry of Reasoning: Flowing Logics in Representation Spacecited by 21
  10. 2025Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thoughtcited by 15
  11. 2025MONICA: Real-Time Monitoring and Calibration of Chain-of-Thought Sycophancy in Large Reasoning Modelscited by 12
  12. 2026Do Sparse Autoencoders Identify Reasoning Features in Language Models?cited by 12
38 more
  1. 2026Reasoning Models Generate Societies of Thoughtcited by 12
  2. 2025SALT: Steering Activations towards Leakage-free Thinking in Chain of Thoughtcited by 11
  3. 2025Enhancing LLM Steering through Sparse Autoencoder-Based Vector Refinementcited by 10
  4. 2025MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steeringcited by 10
  5. 2026Causal Evidence that Language Models use Confidence to Drive Behaviorcited by 10
  6. 2025Small Vectors, Big Effects: A Mechanistic Study of RL-Induced Reasoning via Steering Vectorscited by 8
  7. 2025Schoenfeld's Anatomy of Mathematical Reasoning by Language Modelscited by 8
  8. 2026Steered LLM Activations are Non-Surjectivecited by 8
  9. 2026What Drives Representation Steering? A Mechanistic Case Study on Steering Refusalcited by 7
  10. 2026The Value Axis: Language Models Encode Whether They're on the Right Trackcited by 7
  11. 2025Bridging the Knowledge-Prediction Gap in LLMs on Multiple-Choice Questionscited by 6
  12. 2025Internal states before wait modulate reasoning patternscited by 6
  13. 2026Towards a Mechanistic Understanding of Large Reasoning Models: A Survey of Training, Inference, and Failurescited by 6
  14. 2025PII Jailbreaking in LLMs via Activation Steering Reveals Personal Information Leakagecited by 5
  15. 2025DISCO: Disentangled Communication Steering for Large Language Modelscited by 5
  16. 2025Fantastic Reasoning Behaviors and Where to Find Them: Unsupervised Discovery of the Reasoning Processcited by 5
  17. 2026ReasonOps: Operator Segmentation for LLM Reasoning Tracescited by 5
  18. 2026From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoningcited by 5
  19. 2025Rank-1 LoRAs Encode Interpretable Reasoning Signalscited by 4
  20. 2025ThinkPilot: Steering Reasoning Models via Automated Think-prefixes Optimizationcited by 3
  21. 2025Modeling Hierarchical Thinking in Large Reasoning Modelscited by 3
  22. 2026Reasoning Shift: How Context Silently Shortens LLM Reasoningcited by 3
  23. 2026How Does Reasoning Flow? Tracing Attention-Induced Information Flow for Targeted RL in LLMscited by 3
  24. 2025A Concrete Roadmap towards Safety Cases based on Chain-of-Thought Monitoringcited by 2
  25. 2025Stream: Scaling up Mechanistic Interpretability to Long Context in LLMs via Sparse Attentioncited by 2
  26. 2026Bifrost: Steering Strategic Trajectories to Bridge Contextual Gaps for Self-Improving Agentscited by 2
  27. 2026Emerging Human-like Strategies for Semantic Memory Foraging in Large Language Modelscited by 2
  28. 2026Nudging Hidden States: Training-Free Model Steering for Chain-of-Thought Reasoning in Large Audio-Language Modelscited by 2
  29. 2026The Tell-Tale Norm: ℓ2 Magnitude as a Signal for Reasoning Dynamics in Large Language Modelscited by 2
  30. 2025Annotating the Chain-of-Thought: A Behavior-Labeled Dataset for AI Safetycited by 1
  31. 2025Can Small and Reasoning Large Language Models Score Journal Articles for Research Quality and Do Averaging and Few-shot Help?cited by 1
  32. 2026Is my model "mind blurting"? Interpreting the dynamics of reasoning tokens with Recurrence Quantification Analysis (RQA)cited by 1
  33. 2026How Uncertainty Estimation Scales with Sampling in Reasoning Modelscited by 1
  34. 2026PRISM: A Dual View of LLM Reasoning through Semantic Flow and Latent Computationcited by 1
  35. 2026Reliable Control-Point Selection for Steering Reasoning in Large Language Modelscited by 1
  36. 2026The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignmentcited by 1
  37. 2026HarmThoughts: A Benchmark for Fine-Grained Harmful Behavior Detection in Reasoning Tracescited by 1
  38. 2026How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoningcited by 1

50 most cited of 72 All 72 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Time
  2. 2025Effectively Controlling Reasoning Models through Thinking Intervention
  3. 2025Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models
  4. 2026Can We Break LLMs Out of Self-Loops? Fine-Grained Reasoning Control with Activation Steering
  5. 2025Implicit Reasoning in Large Language Models: A Comprehensive Survey
  6. 2025Advancing Reasoning in Large Language Models: Promising Methods and Approaches
  7. 2026On the Generalization of Steering Vectors for Chain-of-Thought Faithfulness
  8. 2025KV Cache Steering for Controlling Frozen LLMs