EMNLP Findings2025
A Closer Look at Bias and Chain-of-Thought Faithfulness of Large (Vision) Language Models
Sriram Balasubramanian, Samyadeep Basu, Soheil Feizi
Abstract
Chain-of-thought (CoT) reasoning enhances performance of large language models, but questions remain about whether these reasoning traces faithfully reflect the internal processes of the model. We present the first comprehensive study of CoT faithfulness in large vision-language models (LVLMs), investigating how both text-based and previously unexplored image-based biases affect reasoning and bias articulation. Our work introduces a novel, fine-grained evaluation pipeline for categorizing bias articulation patterns, enabling significantly more precise analysis of CoT reasoning than previous methods. This framework reveals critical distinctions in how models process and respond to different types of biases, providing new insights into LVLM CoT faithfulness. Our findings reveal that subtle image-based biases are rarely articulated compared to explicit text-based ones, even in models specialized for reasoning. Additionally, many models exhibit a previously unidentified phenomenon we term ``inconsistent'' reasoning - correctly reasoning before abruptly changing answers, serving as a potential canary for detecting biased reasoning from unfaithful CoTs. We then apply the same evaluation pipeline to revisit CoT faithfulness in LLMs across various levels of implicit cues. Our findings reveal that current language-only reasoning models continue to struggle with articulating cues that are not overtly stated.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
2 top-tier papers have cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- DecepChain: Inducing Deceptive Reasoning in Large Language Models cites Chain-of-Thought Reasoning In The Wild Is Not Always Faithful
- Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting cites Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them
- Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs cites Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting
- On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Cited by top-tier papers2
Builds on13
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
- Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought PromptingMiles Turpin, Julian Michael, Ethan Perez, Samuel R. BowmanNeurIPS 2023 · 1,792 citations
- Distributionally Robust Neural NetworksShiori Sagawa, Pang Wei Koh, Tatsunori B. Hashimoto, Percy LiangICLR 2020 · 1,578 citations
- Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve ThemMirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann et al.ACL Findings 2023 · 253 citations
- Chain-of-Thought Reasoning In The Wild Is Not Always FaithfulIván Arcuschin, Jett Janiak, Robert Krzyzanowski, Senthooran Rajamanoharan et al.ICML 2026 · 175 citations
Related papers
- Visual Thoughts: A Unified Perspective of Understanding Multimodal Chain-of-ThoughtZihui Cheng, Qiguang Chen, Xiao Xu, Jiaqi Wang et al.NeurIPS 2025 · 38 citations
- Biases in the Blind Spot: Detecting What LLMs Fail to MentionIván Arcuschin, David Chanin, Adrià Garriga-Alonso, Oana-Maria CamburuICML 2026 · 7 citations
- Unveiling Confirmation Bias in Chain-of-Thought ReasoningYue Wan, Xiaowei Jia, Xiang Lorraine LiACL Findings 2025 · 1 citation
- FaithCoT-Bench: Benchmarking Instance-Level Faithfulness of Chain-of-Thought ReasoningXu Shen, Song Wang, Zhen Tan, Laura Yao et al.ICLR 2026 · 28 citations
- ImgCoT: Compressing Long Chain of Thought into Compact Visual Tokens for Efficient Reasoning of Large Language ModelXiaoshu Chen, sihang zhou, KE LIANG, Taichun Zhou et al.ICML 2026 · 1 citation
