Analysing Chain of Thought Dynamics: Active Guidance or Unfaithful Post-hoc Rationalisation?

2025/08/27 by Lewis-Lim, Samuel, Tan, Xingwei, Zhao, Zhixue +1 · 3 citations
#Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences

paper · doi:10.48550/arxiv.2508.19827

Abstract

Recent work has demonstrated that Chain-of-Thought (CoT) often yields limited gains for soft-reasoning problems such as analytical and commonsense reasoning. CoT can also be unfaithful to a model's actual reasoning. We investigate the dynamics and faithfulness of CoT in soft-reasoning tasks across instruction-tuned, reasoning and reasoning-distilled models. Our findings reveal differences in how these models rely on CoT, and show that CoT influence and faithfulness are not always aligned.

Citations

• Qwen3 Technical Report
• Reasoning Models Don't Always Say What They Think
• Truthful or Fabricated? Using Causal Attribution to Mitigate Reward Hacking in Explanations
• Are DeepSeek R1 And Other Reasoning Models More Faithful?
• Qwen2.5 Technical Report
• Mind Your Step (by Step): Chain-of-Thought can Reduce Performance on Tasks where Thinking Makes Humans Worse
• RULEBREAKERS: Challenging LLMs at the Crossroads between Formal Logic and Human-like Reasoning
• To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning
• MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark
• How Likely Do LLMs with CoT Mimic Human Reasoning?
• Chain-of-Thought Unfaithfulness as Disguised Accuracy
• GPQA: A Graduate-Level Google-Proof Q&A Benchmark
• Efficient Memory Management for Large Language Model Serving with PagedAttention
• Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting
• Training language models to follow instructions with human feedback
• Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies
• CommonsenseQA: A Question Answering Challenge Targeting Commonsense Knowledge

Cited by

• Enhancing Logical Reasoning in Language Models via Symbolically-Guided Monte Carlo Process Supervision
• Large Language Model Agents Are Not Always Faithful Self-Evolvers
• The Tell-Tale Trace: Detecting Reasoning Failures in LLMs Using Chain-of-Thought Dynamics
• Reasoning Dynamics and the Limits of Monitoring Modality Reliance in Vision-Language Models

Related

• Post-Hoc Reasoning in Chain of Thought: Decoding and Steering Pre-Committed Answers
• Measuring Chain of Thought Faithfulness by Unlearning Reasoning Steps
• Measuring Faithfulness in Chain-of-Thought Reasoning
• CoT is Not True Reasoning, It Is Just a Tight Constraint to Imitate: A Theory Perspective
• Towards Better Chain-of-Thought: A Reflection on Effectiveness and Faithfulness