Tag
This paper investigates whether high-quality Natural Language Explanations (NLEs) generated by LLMs from XAI outputs actually improve task performance, finding they do not aid accuracy but inflate confidence, revealing a quality-usefulness gap.