Faithfulness Serum: Mitigating the Faithfulness Gap in Textual Explanations of LLM Decisions via Attribution Guidance

View PDF HTML (experimental) Abstract:Large language models (LLMs) achieve strong performance and have revolutionized NLP, but their lack of explainability keeps them treated as black boxes, limiting their use in domains that demand transparency and trust. A promising direction to address this issue is post-hoc text-based explanations, which aim to explain model decisions in natural language. Prior work has focused on generating convincing rationales that appear to be subjectively faithful, but it remains unclear whether these explanations are epistemically faithful, whether they reflect the internal evidence the model actually relied on for its decision. In this paper, we first assess the epistemic faithfulness of LLM-generated explanations via counterfactuals and show that they are often unfaithful. We then introduce a training-free method that enhances faithfulness by guiding explanation generation through attention-level interventions, informed by token-level heatmaps extracted via a faithful attribution method. This method significantly improves epistemic faithfulness across multiple models, benchmarks, and prompts. Comments: 24 pages, multiple figures (e.g., at least 6 main figures), includes experiments across several benchmarks (MMLU, CommonsenseQA, SciQ, ARC, OpenBookQA); code available on GitHub Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI) Cite as: arXiv:2604.14325 [cs.CL] (or arXiv:2604.14325v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2604.14325 arXiv-issued DOI via DataCite (pending registration) Submission history From: Bar Alon [view email] [v1] Wed, 15 Apr 2026 18:32:32 UTC (2,205 KB)