paper-with-me

홈 › Papers

Faithfulness as Information Flow: Evaluating and Training Faithful Chain-of-Thought Reasoning

2026-05-22 · Jinghan Jia, Joe Benton, Eric Easley arxiv

Chain-of-thought (CoT) reasoning is useful for monitoring language models only when the reasoning trace faithfully reflects the computation that produces the final answer. However, models can rely on prompt-to-answer shortcuts that bypass the CoT, making the visible reasoning trace misleading even when it appears plausible. We study CoT faithfulness through a structural information-flow perspective: faithful reasoning should route answer-relevant information through the mediated path from prompt to CoT to answer, rather than through a direct prompt-to-answer shortcut. This perspective yields a task-agnostic framework based on three complementary properties, sufficiency, completeness, and necessity, which we instantiate with entropy-based, masked-KL, and gradient-based diagnostics. We show that these metrics recover externally judged faithfulness differences in hinted reasoning, and identify a low-entropy failure mode of KL-based diagnostics where gradient-based measures remain more stable. Building on this analysis, we introduce update-time interventions for verifier-based on-policy RL, including attention masking, backward-only gradient masking, CoT gradients, and adversarial perturbations of prompt representations. Across hinted arithmetic, reward-hackable code repair, and DAPO-Math models trained without hints but evaluated under wrong-hint injection, our interventions shift behavioral and structural indicators toward stronger CoT mediation. In particular, they make shortcut and reward-hacking behavior more transparent in the CoT and improve task-agnostic faithfulness metrics, while in some settings also reducing wrong-hint susceptibility. Our results suggest that controlling information flow during training is a practical route toward more faithful and monitorable CoT reasoning. Code is available at https://github.com/safety-research/faithful-cot.

📄 PDF Abstract BibTeX arXiv:2605.24286

Code (0)

등록된 구현이 없습니다.

Tasks

Code Repair

Similar Papers 제목 키워드 기반

FaithCAMERA: Construction of a Faithful Dataset for Ad Text Generation

2024-10-04 · Akihiko Kato, Masato Mita, Soichiro Murakami, Ukyo Honda 외

In ad text generation (ATG), desirable ad text is both faithful and informative. That is, it should be faithful to the input document, while at the same time containing important information that appeals to potential cus…

InformativenessSentenceText Generation

ED-FAITH: Evaluating Dialogue Summarization on Faithfulness

2022-11-15 · Sicong Huang, Asli Celikyilmaz, Haoran Li

Abstractive summarization models typically generate content unfaithful to the input, thus highlighting the significance of evaluating the faithfulness of generated summaries. Most faithfulness metrics are only evaluated …

Abstractive Text SummarizationLanguage ModelingLanguage Modelling

Faithful-First Reasoning, Planning, and Acting for Multimodal LLMs

2025-11-11 · Junxian Li, Xinyue Xu, Sai Ma, Di Zhang 외 arxiv

Multimodal Large Language Models (MLLMs) frequently suffer from unfaithfulness, generating reasoning chains that drift from visual evidence or contradict final predictions. We propose Faithful-First Reasoning, Planning, …

Multimodal Reasoning

FEQA: A Question Answering Evaluation Framework for Faithfulness Assessment in Abstractive Summarization

2020-05-07 · ACL 2020 6 · Esin Durmus, He He, Mona Diab

Neural abstractive summarization models are prone to generate content inconsistent with the source document, i.e. unfaithful. Existing automatic metrics do not capture such mistakes effectively. We tackle the problem of …

Abstractive Text SummarizationQuestion AnsweringReading Comprehension

Semantic Faithfulness and Entropy Production Measures to Tame Your LLM Demons and Manage Hallucinations

2025-12-04 · Igor Halperin arxiv

Evaluating faithfulness of Large Language Models (LLMs) to a given task is a complex challenge. We propose two new unsupervised metrics for faithfulness evaluation using insights from information theory and thermodynamic…

Answer Generation