mlsanityData to 14 Oct 2026

Answer-Centric or Reasoning-Driven? Uncovering the Latent Memory Anchor in LLMs

arXiv 2506.17630 ↗First posted 21 Jun 2025cs.CLcited by 0

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

While Large Language Models (LLMs) demonstrate impressive reasoning capabilities, growing evidence suggests much of their success stems from memorized answer-reasoning patterns rather than genuine inference. In this work, we investigate a central question: are LLMs primarily anchored to final answers or to the textual pattern of reasoning chains? We propose a five-level answer-visibility prompt framework that systematically manipulates answer cues and probes model behavior through indirect, behavioral analysis. Experiments across state-of-the-art LLMs reveal a strong and consistent reliance on explicit answers. The performance drops by 26.90% when answer cues are masked, even with complete reasoning chains. These findings suggest that much of the reasoning exhibited by LLMs may reflect post-hoc rationalization rather than true inference, calling into question their inferential depth. Our study uncovers the answer-anchoring phenomenon with rigorous empirical validation and underscores the need for a more nuanced understanding of what constitutes reasoning in LLMs.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 26 of its 52 references that are in the map. First posted on arXiv 21 Jun 2025.

Builds on

26 of 52 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Patterns Over Principles: The Fragility of Inductive Reasoning in LLMs under Noisy Observations22 Feb
  2. 2024Beyond Math: Stories as a Testbed for Memorization-Constrained Reasoning in LLMs18 Dec
  3. 2024What Do Learning Dynamics Reveal About Generalization in LLM Reasoning?12 Nov
  4. 2024On Memorization of Large Language Models in Logical Reasoning30 Oct

Other references · 22

  1. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  2. 2024GPT-4o System Cardcited by 4,588
  3. 2023A Survey of Large Language Modelscited by 2,576
  4. 2024A Survey on Large Language Models for Code Generationcited by 658
  5. 2024Large Language Models for Mathematical Reasoning: Progresses and Challengescited by 284
  6. 2025Chain-of-Thought Reasoning In The Wild Is Not Always Faithfulcited by 170
  7. 2024Benchmark Data Contamination of Large Language Models: A Surveycited by 147
  8. 2022Holistic Evaluation of Language Modelscited by 131
  9. 2015Yara Parser: A Fast and Accurate Dependency Parsercited by 111
  10. 2025MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbationscited by 89
  11. 2025Missing Premise exacerbates Overthinking: Are Reasoning Models losing Critical Thinking Skill?cited by 58
  12. 2023An Open Source Data Contamination Report for Large Language Modelscited by 51
10 more
  1. 2025Recent Advances in Large Langauge Model Benchmarks against Data Contamination: From Static to Dynamic Evaluationcited by 38
  2. 2024Can Large Language Models Understand Symbolic Graphics Programs?cited by 35
  3. 2025Mathematical Reasoning in Large Language Models: Assessing Logical and Arithmetic Errors across Wide Numerical Rangescited by 24
  4. 2025The Order Effect: Investigating Prompt Sensitivity to Input Order in LLMscited by 19
  5. 2025Recitation over Reasoning: How Cutting-Edge Language Models Can Fail on Elementary School-Level Reasoning Problems?cited by 15
  6. 2025Exploring LLM Reasoning Through Controlled Prompt Variationscited by 10
  7. 2025Predicting the Performance of Black-box LLMs through Follow-up Queriescited by 9
  8. 2025Memorize or Generalize? Evaluating LLM Code Generation with Code Rewritingcited by 7
  9. 2025The Challenge of Identifying the Origin of Black-Box Large Language Modelscited by 5
  10. 2024Neuron-Level Differentiation of Memorization and Generalization in Large Language Modelscited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Beyond the Last Answer: Your Reasoning Trace Uncovers More than You Think
  2. 2024Reasoning or a Semblance of it? A Diagnostic Study of Transitive Reasoning in LLMs
  3. 2025From Reasoning to Answer: Empirical, Attention-Based and Mechanistic Insights into Distilled DeepSeek R1 Models
  4. 2026MemoReason: Evaluating the Effect of Parametric Memory on Contextual Reasoning in LLMs
  5. 2026Same Question, Different Answers: Evaluating LLM Reliability Beyond Accuracy
  6. 2026LLM Reasoning Is Latent, Not the Chain of Thought
  7. 2025Test-Time Reasoners Are Strategic Multiple-Choice Test-Takers
  8. 2025Reasoning Capabilities and Invariability of Large Language Models