Image de couverture de livre

How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoning

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Chen, Haoyang, Liu, Yi, Shao, Jianzhi, Zhang, Tao, Huo, Chengfu, Hu, Wei
Format:	Preprint
Publié:	2026
Sujets:	Computation and Language Artificial Intelligence
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Read Quietly, Think Aloud: Decoupling Comprehension and Reasoning in LLMs
par: Wang, Yuanxin, et autres
Publié: (2025)

Demystifying Reasoning Dynamics with Mutual Information: Thinking Tokens are Information Peaks in LLM Reasoning
par: Qian, Chen, et autres
Publié: (2025)

Do Thinking Tokens Help or Trap? Towards More Efficient Large Reasoning Model
par: Ding, Bowen, et autres
Publié: (2025)

Dynamic Thinking-Token Selection for Efficient Reasoning in Large Reasoning Models
par: Guo, Zhenyuan, et autres
Publié: (2026)

Beyond the Last Answer: Your Reasoning Trace Uncovers More than You Think
par: Hammoud, Hasan Abed Al Kader, et autres
Publié: (2025)

Do LLMs Really Think Step-by-step In Implicit Reasoning?
par: Yu, Yijiong
Publié: (2024)

Do LLMs Encode Functional Importance of Reasoning Tokens?
par: Singh, Janvijay, et autres
Publié: (2026)

When Can Large Reasoning Models Save Thinking? Mechanistic Analysis of Behavioral Divergence in Reasoning
par: Zhu, Rongzhi, et autres
Publié: (2025)

Sandwich Reasoning: An Answer-Reasoning-Answer Approach for Low-Latency Query Correction
par: Zhang, Chen, et autres
Publié: (2026)

How Do LLMs Perform Two-Hop Reasoning in Context?
par: Guo, Tianyu, et autres
Publié: (2025)

Are LLMs Capable of Data-based Statistical and Causal Reasoning? Benchmarking Advanced Quantitative Reasoning with Data
par: Liu, Xiao, et autres
Publié: (2024)

Multiplex Thinking: Reasoning via Token-wise Branch-and-Merge
par: Tang, Yao, et autres
Publié: (2026)

Reverse Thinking Makes LLMs Stronger Reasoners
par: Chen, Justin Chih-Yao, et autres
Publié: (2024)

Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thought
par: Xiang, Violet, et autres
Publié: (2025)

GSM-Infinite: How Do Your LLMs Behave over Infinitely Increasing Context Length and Reasoning Complexity?
par: Zhou, Yang, et autres
Publié: (2025)

Can LLMs Grade Short-Answer Reading Comprehension Questions : An Empirical Study with a Novel Dataset
par: Henkel, Owen, et autres
Publié: (2023)

How Likely Do LLMs with CoT Mimic Human Reasoning?
par: Bao, Guangsheng, et autres
Publié: (2024)

SwiReasoning: Switch-Thinking in Latent and Explicit for Pareto-Superior Reasoning LLMs
par: Shi, Dachuan, et autres
Publié: (2025)

Read Before You Think: Mitigating LLM Comprehension Failures with Step-by-Step Reading
par: Han, Feijiang, et autres
Publié: (2025)

LLMs are Single-threaded Reasoners: Demystifying the Working Mechanism of Soft Thinking
par: Wu, Junhong, et autres
Publié: (2025)

Reasoning or Retrieval? A Study of Answer Attribution on Large Reasoning Models
par: Wang, Yuhui, et autres
Publié: (2025)

Tracing How Annotators Think: Augmenting Preference Judgments with Reading Processes
par: de Langis, Karin, et autres
Publié: (2025)

Beyond Answers: Transferring Reasoning Capabilities to Smaller LLMs Using Multi-Teacher Knowledge Distillation
par: Tian, Yijun, et autres
Publié: (2024)

Tokenization Constraints in LLMs: A Study of Symbolic and Arithmetic Reasoning Limits
par: Zhang, Xiang, et autres
Publié: (2025)

Reasoning Pattern Alignment Merging for Adaptive Reasoning
par: Zhong, Zhaofeng, et autres
Publié: (2026)

LLMs for Relational Reasoning: How Far are We?
par: Li, Zhiming, et autres
Publié: (2024)

RAG over Thinking Traces Can Improve Reasoning Tasks
par: Arabzadeh, Negar, et autres
Publié: (2026)

Think Clearly: Improving Reasoning via Redundant Token Pruning
par: Choi, Daewon, et autres
Publié: (2025)

Say Anything but This: When Tokenizer Betrays Reasoning in LLMs
par: Ayoobi, Navid, et autres
Publié: (2026)

Do Cognitively Interpretable Reasoning Traces Improve LLM Performance?
par: Bhambri, Siddhant, et autres
Publié: (2025)

Preemptive Answer "Attacks" on Chain-of-Thought Reasoning
par: Xu, Rongwu, et autres
Publié: (2024)

KAG-Thinker: Interactive Thinking and Deep Reasoning in LLMs via Knowledge-Augmented Generation
par: Zhang, Dalong, et autres
Publié: (2025)

Code to Think, Think to Code: A Survey on Code-Enhanced Reasoning and Reasoning-Driven Code Intelligence in LLMs
par: Yang, Dayu, et autres
Publié: (2025)

ThinkPilot: Steering Reasoning Models via Automated Think-prefixes Optimization
par: Li, Sunzhu, et autres
Publié: (2025)

Reading or Reasoning? Format Decoupled Reinforcement Learning for Document OCR
par: Zhong, Yufeng, et autres
Publié: (2025)

MR-Ben: A Meta-Reasoning Benchmark for Evaluating System-2 Thinking in LLMs
par: Zeng, Zhongshen, et autres
Publié: (2024)

Smooth Reading: Bridging the Gap of Recurrent LLM to Self-Attention LLM on Long-Context Tasks
par: Liu, Kai, et autres
Publié: (2025)

Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Space
par: Zhang, Zhen, et autres
Publié: (2025)

Draft-Thinking: Learning Efficient Reasoning in Long Chain-of-Thought LLMs
par: Cao, Jie, et autres
Publié: (2026)

Controlling Thinking Speed in Reasoning Models
par: Lin, Zhengkai, et autres
Publié: (2025)