faithfulness

This paper introduces an architecture-aware explanation audit protocol for industrial visual inspection, demonstrating that the faithfulness of explanation methods is bounded by their structural compatibility with a model's native decision mechanism, using experiments on wafer map and anomaly detection datasets.

This paper presents the winning system for SemEval-2026 Task 8's generation subtask, using a heterogeneous ensemble of seven LLMs with dual prompting strategies and a GPT-4o-mini judge to select the best response. The system achieved first place with a conditioned harmonic mean of 0.7827, outperforming all baselines and demonstrating the value of model diversity.

This paper introduces FRANQ, a method for detecting hallucinations in Retrieval-Augmented Generation (RAG) systems by applying distinct uncertainty quantification techniques to distinguish between factuality and faithfulness to retrieved context. The authors construct a new dataset annotated for both factuality and faithfulness, and demonstrate that FRANQ outperforms existing approaches in detecting factual errors across multiple datasets and LLMs.

AtManRL is a method that uses differentiable attention manipulation and reinforcement learning to train LLMs to generate more faithful chain-of-thought reasoning by ensuring reasoning tokens causally influence final predictions. Experiments on GSM8K and MMLU with Llama-3.2-3B demonstrate the approach can identify influential reasoning tokens and improve reasoning transparency.