Is Chain-of-Thought Really Not Explainability? Chain-of-Thought Can Be Faithful without Hint Verbalization

TL;DR

This paper challenges the notion that chain-of-thought explanations are unfaithful if they omit hints, showing that many are actually faithful and that evaluation metrics should be broader.

Contribution

It introduces a new faithful@k metric, applies causal mediation analysis, and argues for a broader interpretability toolkit beyond hint-based evaluations.

Findings

Larger inference budgets increase hint verbalization up to 90%.

Many CoTs flagged as unfaithful are actually faithful according to other metrics.

Hint omission alone does not prove unfaithfulness.

Abstract

Recent work, using the Biasing Features metric, labels a CoT as unfaithful if it omits a prompt-injected hint that affected the prediction. We argue this metric adopts a narrow notion of faithfulness and confuses unfaithfulness with incompleteness, the lossy compression needed to turn distributed transformer computation into a linear natural language narrative. On multi-hop reasoning tasks with instruct-tuned and reasoning models, many CoTs flagged as unfaithful by Biasing Features are judged faithful by other metrics, exceeding 50% in some models. With a new faithful@k metric, we show that larger inference-time budgets greatly increase hint verbalization (up to 90% in some settings), suggesting much apparent unfaithfulness is due to tight token limits. Using Causal Mediation Analysis, we further show that even non-verbalized hints can causally mediate prediction changes through the CoT. We therefore caution against relying solely on hint-based evaluations and advocate a broader interpretability toolkit, including causal mediation and corruption-based metrics. We do not claim all CoTs are faithful, only that the absence of hint words alone does not prove unfaithfulness.