Making LLMs Say What They Think: Measuring and Improving CoT-Interpretability Alignment
cs.CL, cs.AI
Submitted: 2026-09-30
Updated: 2026-10-09
Comments: 28 pages, 9 figures, 10 tables
Code: https://github.com/yihuaihong/CIA-minimal-repro
License: http://creativecommons.org/licenses/by/4.0/
The gist: Chain-of-thought (CoT) traces often serve as a proxy for how Large Language Models (LLMs) arrive at their answers.
Terminology
Abstract
Chain-of-thought (CoT) traces often serve as a proxy for how Large Language Models (LLMs) arrive at their answers. However, growing evidence shows that models' CoT often fails to reflect their internal computations and can be changed without affecting their final answers. In this work, we measure and improve the alignment between the reasoning described in an LLM's CoT and what it computes internally. We propose CoT-Interpretability Alignment (CIA), a metric that measures the agreement between a model's CoT traces and its internal reasoning strategies as detected by interpretability tools. We evaluate CIA on three tasks (two-hop question answering, hint intervention, and integer multiplication) across three LLMs, finding that LLMs exhibit limited alignment across all tasks (44.8-75.9%). We then experiment with improving CIA via post-training, setting both the task accuracy and parametric faithfulness signals as a reward. Experiments show that we can substantially improve CoT parametric faithfulness while maintaining or improving the task accuracy. We provide rich analysis, such as their generalization patterns. Our work provides both a framework for auditing CoT parametric faithfulness and a pathway toward making models' explicit reasoning more trustworthy. Code and data are available at https://github.com/yihuaihong/CIA-minimal-repro.
Sources
- Why Can't Transformers Learn Multiplication? Reverse-Engineering Reveals Long-Range Dependency Pitfalls
- Eliciting Latent Predictions from Transformers with the Tuned Lens
- Eight Things to Know about Large Language Models
- Reasoning Models Don't Always Say What They Think
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- Gemma 2: Improving Open Language Models at a Practical Size
- The Llama 3 Herd of Models
- Counterfactual Simulation Training for Chain-of-Thought Faithfulness
- Measuring Faithfulness in Chain-of-Thought Reasoning
- Parcae: Scaling Laws For Stable Looped Language Models
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- Is Chain-of-Thought Really Not Explainability? Chain-of-Thought Can Be Faithful without Hint Verbalization
- Scaling Latent Reasoning via Looped Language Models
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering