mlsanityData to 7 Oct 2026

How to think step-by-step: A mechanistic understanding of chain-of-thought reasoning

arXiv 2402.18312 ↗First posted 28 Feb 2024cs.CL cs.LGcited by 53

On the map: LLM reasoning → Prompting and chain-of-thought reasoning

Abstract (arXiv)

Despite superior reasoning prowess demonstrated by Large Language Models (LLMs) with Chain-of-Thought (CoT) prompting, a lack of understanding prevails around the internal mechanisms of the models that facilitate CoT generation. This work investigates the neural sub-structures within LLMs that manifest CoT reasoning from a mechanistic point of view. From an analysis of Llama-2 7B applied to multistep reasoning over fictional ontologies, we demonstrate that LLMs deploy multiple parallel pathways of answer generation for step-by-step reasoning. These parallel pathways provide sequential answers from the input question context as well as the generated CoT. We observe a functional rift in the middle layers of the LLM. Token representations in the initial half remain strongly biased towards the pretraining prior, with the in-context prior taking over in the later half. This internal phase shift manifests in different functional components: attention heads that write the answer token appear in the later half, attention heads that move information along ontological relationships appear in the initial half, and so on. To the best of our knowledge, this is the first attempt towards mechanistic investigation of CoT reasoning in LLMs.

Key facts. Cited by 53 papers in the map (data through 7 Oct 2026). Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning. Builds on 21 of its 32 references that are in the map. First posted on arXiv 28 Feb 2024.

Builds on

21 of 32 references.

Closest ancestors · 5

Same direction, newest first

  1. 2023Towards Revealing the Mystery behind Chain of Thought: A Theoretical Perspective24 May
  2. 2023Why think step by step? Reasoning emerges from the locality of experience7 Apr
  3. 2022Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks22 Nov
  4. 2022Large Language Models are Zero-Shot Reasoners24 May
  5. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Models28 Jan

Other references · 16

  1. 2017Attention Is All You Needcited by 59,932
  2. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,581
  3. 2022Locating and Editing Factual Associations in GPTcited by 2,609
  4. 2022Emergent Abilities of Large Language Modelscited by 2,483
  5. 2020Transformer Feed-Forward Layers Are Key-Value Memoriescited by 1,307
  6. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,130
  7. 2022Toy Models of Superpositioncited by 928
  8. 2021Show Your Work: Scratchpads for Intermediate Computation with Language Modelscited by 871
  9. 2021Probing Classifiers: Promises, Shortcomings, and Advancescited by 831
  10. 2022Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thoughtcited by 334
  11. 2023Towards Best Practices of Activation Patching in Language Models: Metrics and Methodscited by 328
  12. 2022Transformers Learn Shortcuts to Automatacited by 260
4 more
  1. 2022Can language models learn from explanations in context?cited by 248
  2. 2023The Hydra Effect: Emergent Self-repair in Language Model Computationscited by 145
  3. 2023Interpretability at Scale: Identifying Causal Mechanisms in Alpacacited by 102
  4. 2023Circuit Breaking: Removing Model Behaviors with Targeted Ablationcited by 27

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 16; 2025: 21; 2026: 16.

53 papers.

Sort:

By area:

  1. 2025From System 1 to System 2: A Survey of Reasoning Large Language Modelscited by 319
  2. 2025Thought Anchors: Which LLM Reasoning Steps Matter?cited by 126
  3. 2024A Practical Review of Mechanistic Interpretability for Transformer-Based Language Modelscited by 113
  4. 2024Round and Round We Go! What makes Rotary Positional Encodings useful?cited by 111
  5. 2024Beyond Accuracy: Evaluating the Reasoning Behavior of Large Language Models -- A Surveycited by 82
  6. 2025A Survey on Latent Reasoningcited by 71
  7. 2025Understanding Reasoning in Thinking Language Models via Steering Vectorscited by 69
  8. 2024Knowledge Circuits in Pretrained Transformerscited by 60
  9. 2024Unlocking the Capabilities of Thought: A Reasoning Boundary Framework to Quantify and Optimize Chain-of-Thoughtcited by 56
  10. 2024Knowledge Mechanisms in Large Language Models: A Survey and Perspectivecited by 50
  11. 2024From Sparse Dependence to Sparse Attention: Unveiling How Chain-of-Thought Enhances Transformer Sample Efficiencycited by 28
  12. 2025BOLT: Bootstrap Long Chain-of-Thought in Language Models without Distillationcited by 23
38 more
  1. 2025ShadowCoT: Cognitive Hijacking for Stealthy Reasoning Backdoors in LLMscited by 23
  2. 2025Attention Illuminates LLM Reasoning: The Preplan-and-Anchor Rhythm Enables Fine-Grained Policy Optimizationcited by 23
  3. 2025How does Chain of Thought Think? Mechanistic Interpretability of Chain-of-Thought Reasoning with Sparse Autoencodingcited by 19
  4. 2026LLM Reasoning as Trajectories: Step-Specific Representation Geometry and Correctness Signalscited by 18
  5. 2025AdaReasoner: Adaptive Reasoning Enables More Flexible Thinking in Large Language Modelscited by 16
  6. 2024Understanding the Language Model to Solve the Symbolic Multi-Step Reasoning Problem from the Perspective of Buffer Mechanismcited by 15
  7. 2024A Hopfieldian View-based Interpretation for Chain-of-Thought Reasoningcited by 15
  8. 2024Unveiling Factual Recall Behaviors of Large Language Models through Knowledge Neuronscited by 15
  9. 2024Understanding Reasoning in Chain-of-Thought from the Hopfieldian Viewcited by 10
  10. 2026Do Latent-CoT Models Think Step-by-Step? A Mechanistic Study on Sequential Reasoning Taskscited by 10
  11. 2025Thinking Sparks!: Emergent Attention Heads in Reasoning Models During Post Trainingcited by 8
  12. 2024When Parts Are Greater Than Sums: Individual LLM Components Can Outperform Full Modelscited by 6
  13. 2026The Illusion of Superposition? A Principled Analysis of Latent Thinking in Language Modelscited by 6
  14. 2024Multilevel Interpretability Of Artificial Neural Networks: Leveraging Framework And Methods From Neurosciencecited by 5
  15. 2025Punctuation and Predicates in Language Modelscited by 5
  16. 2025All for One: LLMs Solve Mental Math at the Last Token With Information Transferred From Other Tokenscited by 5
  17. 2025State over Tokens: Characterizing the Role of Reasoning Tokenscited by 5
  18. 2024Anthropocentric bias in language model evaluationcited by 4
  19. 2025RBF++: Quantifying and Optimizing Reasoning Boundaries across Measurable and Unmeasurable Capabilities for Chain-of-Thought Reasoningcited by 4
  20. 2026Unveiling the Entropy Dynamics of Chain-of-Thought Reasoningcited by 4
  21. 2025Policy Frameworks for Transparent Chain-of-Thought Reasoning in Large Language Modelscited by 3
  22. 2026Indications of Belief-Guided Agency and Meta-Cognitive Monitoring in Large Language Modelscited by 3
  23. 2026How Does Reasoning Flow? Tracing Attention-Induced Information Flow for Targeted RL in LLMscited by 3
  24. 2025Mechanistic Understanding of Language Models in Syntactic Code Completioncited by 2
  25. 2025Stream: Scaling up Mechanistic Interpretability to Long Context in LLMs via Sparse Attentioncited by 2
  26. 2026Reasoning Fails Where Step Flow Breakscited by 2
  27. 2025How Chain-of-Thought Works? Tracing Information Flow from Decoding, Projection, and Activationcited by 1
  28. 2026Scaling Reasoning Hop Exposes Weaknesses: Demystifying and Improving Hop Generalization in Large Language Modelscited by 1
  29. 2026From independent patches to coordinated attention: Controlling information flow in vision transformerscited by 1
  30. 2026Is my model "mind blurting"? Interpreting the dynamics of reasoning tokens with Recurrence Quantification Analysis (RQA)cited by 1
  31. 2026PRISM: A Dual View of LLM Reasoning through Semantic Flow and Latent Computationcited by 1
  32. 2026Mechanistic Interpretability of Chain-of-Thought Reasoning via Sequential Activation Patchingcited by 1
  33. 2024Information Anxiety in Large Language Modelscited by 0
  34. 2024Exploring Performance Contrasts in TableQA: Step-by-Step Reasoning Boosts Bigger Language Models, Limits Smaller Language Modelscited by 0
  35. 2025Is the `Agent' Paradigm a Limiting Framework for Next-Generation Intelligent Systems?cited by 0
  36. 2025Plug-and-Play Emotion Graphs for Compositional Prompting in Zero-Shot Speech Emotion Recognitioncited by 0
  37. 2025BRAID: Bounded Reasoning for Autonomous Inference and Decisionscited by 0
  38. 2026Mechanics of Bias and Reasoning: Interpreting the Impact of Chain-of-Thought Prompting on Gender Bias in LLMscited by 0

50 most cited of 53 All 53 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Explainable Chain-of-Thought Reasoning: An Empirical Analysis on State-Aware Reasoning Dynamics
  2. 2024Fine-Tuning on Diverse Reasoning Chains Drives Within-Inference CoT Refinement in LLMs
  3. 2024Chain-of-Thought in Large Language Models: Decoding, Projection, and Activation
  4. 2022Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matters
  5. 2025Early Stopping Chain-of-thoughts in Large Language Models
  6. 2025CoT is Not True Reasoning, It Is Just a Tight Constraint to Imitate: A Theory Perspective
  7. 2025Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoning
  8. 2025The Curse of CoT: On the Limitations of Chain-of-Thought in In-Context Learning