mlsanityData to 14 Oct 2026

The Reasoning-Memorization Interplay in Language Models Is Mediated by a Single Direction

arXiv 2503.23084 ↗First posted 29 Mar 2025cs.CL cs.AIcited by 21

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

Large language models (LLMs) excel on a variety of reasoning benchmarks, but previous studies suggest they sometimes struggle to generalize to unseen questions, potentially due to over-reliance on memorized training examples. However, the precise conditions under which LLMs switch between reasoning and memorization during text generation remain unclear. In this work, we provide a mechanistic understanding of LLMs' reasoning-memorization dynamics by identifying a set of linear features in the model's residual stream that govern the balance between genuine reasoning and memory recall. These features not only distinguish reasoning tasks from memory-intensive ones but can also be manipulated to causally influence model performance on reasoning tasks. Additionally, we show that intervening in these reasoning features helps the model more accurately activate the most relevant problem-solving capabilities during answer generation. Our findings offer new insights into the underlying mechanisms of reasoning and memory in LLMs and pave the way for the development of more robust and interpretable generative AI systems.

Key facts. Cited by 21 papers in the map (data through 14 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 37 of its 67 references that are in the map. First posted on arXiv 29 Mar 2025.

Builds on

37 of 67 references.

Closest ancestors · 6

Same direction, newest first

  1. 2024On Memorization of Large Language Models in Logical Reasoning30 Oct
  2. 2024Grokked Transformers are Implicit Reasoners: A Mechanistic Journey to the Edge of Generalization23 May
  3. 2023Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Models23 Oct
  4. 2023Code Llama: Open Foundation Models for Code24 Aug
  5. 2023Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks5 Jul
  6. 2023Large Language Models are In-Context Semantic Reasoners rather than Symbolic Reasoners24 May

Other references · 31

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  5. 2024GPT-4o System Cardcited by 4,588
  6. 2021Program Synthesis with Large Language Modelscited by 3,637
  7. 2023Mistral 7Bcited by 3,164
  8. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,998
  9. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  10. 2022Toy Models of Superpositioncited by 971
  11. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 716
  12. 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 712
19 more
  1. 2023Measuring Faithfulness in Chain-of-Thought Reasoningcited by 703
  2. 2022Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasetscited by 629
  3. 2022Language Models are Multilingual Chain-of-Thought Reasonerscited by 530
  4. 20242 OLMo 2 Furiouscited by 340
  5. 2023Proving Test Set Contamination in Black Box Language Modelscited by 230
  6. 2023Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solvecited by 178
  7. 2024Do Large Language Models Latently Perform Multi-Hop Reasoning?cited by 169
  8. 2023Linear Representations of Sentiment in Large Language Modelscited by 163
  9. 2023Causal Parrots: Large Language Models May Talk Causality But Are Not Causalcited by 150
  10. 2023Emergent and Predictable Memorization in Large Language Modelscited by 148
  11. 2024Physics of Language Models: Part 2.1, Grade-School Math and the Hidden Reasoning Processcited by 108
  12. 2024Hopping Too Late: Exploring the Limitations of Large Language Models on Multi-Hop Queriescited by 78
  13. 2024Attention Heads of Large Language Models: A Surveycited by 67
  14. 2024Robust LLM safeguarding via refusal feature adversarial trainingcited by 64
  15. 2024Demystifying Verbatim Memorization in Large Language Modelscited by 56
  16. 2024Steering Without Side Effects: Improving Post-Deployment Control of Language Modelscited by 46
  17. 2024Intrinsic Test of Unlearning Using Parametric Knowledge Tracescited by 29
  18. 2024Deciphering the Factors Influencing the Efficacy of Chain-of-Thought: Probability, Memorization, and Noisy Reasoningcited by 27
  19. 2024Case-Based or Rule-Based: How Do Transformers Do the Math?cited by 26

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 13; 2026: 8.

21 papers.

Sort:

By area:

  1. 2025AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraintcited by 49
  2. 2026Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Modelscited by 27
  3. 2025MetaFaith: Faithful Natural Language Uncertainty Expression in LLMscited by 16
  4. 2026Spurious Rewards Paradox: Mechanistically Understanding How RLVR Activates Memorization Shortcuts in LLMscited by 14
  5. 2025On Reasoning Strength Planning in Large Reasoning Modelscited by 13
  6. 2025Detecting (Un)answerability in Large Language Models with Linear Directionscited by 13
  7. 2025The Landscape of Memorization in LLMs: Mechanisms, Measurement, and Mitigationcited by 10
  8. 2025MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steeringcited by 10
  9. 2025Hypothesis-Driven Feature Manifold Analysis in LLMs via Supervised Multi-Dimensional Scalingcited by 8
  10. 2025Stochastic Chameleons: Irrelevant Context Hallucinations Reveal Class-Based (Mis)Generalization in LLMscited by 6
  11. 2025Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injectioncited by 5
  12. 2025Diagnosing Memorization in Chain-of-Thought Reasoning, One Token at a Timecited by 4
9 more
  1. 2025ThoughtProbe: Classifier-Guided LLM Thought Space Exploration via Probing Representationscited by 3
  2. 2025Eliciting Chain-of-Thought in Base LLMs via Gradient-Based Representation Optimizationcited by 2
  3. 2025Adaptive Multi-Agent Reasoning via Automated Workflow Generationcited by 1
  4. 2026Memory Dial: A Training Framework for Controllable Memorization in Language Modelscited by 1
  5. 2026The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignmentcited by 1
  6. 2026Multilingual Idioms in Sentences and Conversations Across High-, Medium-, and Low-Resource Languagescited by 0
  7. 2026IsoSci: A Benchmark of Isomorphic Cross-Domain Science Problems for Evaluating Reasoning versus Knowledge Retrieval in LLMscited by 0
  8. 2026Beyond Fixed Directions: Adaptive Representation Analysis of Reasoning and Memorization in LLMscited by 0
  9. 2026How Do Linear Probes Emerge? A Circuit-Tracing Framework with Concept-Targeted Attributioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026MemoReason: Evaluating the Effect of Parametric Memory on Contextual Reasoning in LLMs
  2. 2024Disentangling Memory and Reasoning Ability in Large Language Models
  3. 2025Answer-Centric or Reasoning-Driven? Uncovering the Latent Memory Anchor in LLMs
  4. 2025Advancing Reasoning in Large Language Models: Promising Methods and Approaches
  5. 2026Reasoning as State Transition: A Representational Analysis of Reasoning Evolution in Large Language Models
  6. 2025Reason to Rote: Rethinking Memorization in Reasoning
  7. 2024Unveiling Factual Recall Behaviors of Large Language Models through Knowledge Neurons
  8. 2026PRISM: A Dual View of LLM Reasoning through Semantic Flow and Latent Computation