One SPACE to Rule Them All: Jointly Mitigating Factuality and Faithfulness Hallucinations in LLMs

Pengbo Wang, Chaozhuo Li, Chenxu Wang, Liwen Zheng, Litian Zhang, Xi Zhang

Keywords: LLM, Hallucination, Safety

TL;DR: We uncover a critical factuality-faithfulness trade-off in LLM hallucinations and propose SPACE, a novel framework theoretically and experimentally validated on TruthfulQA and PDTB to effectively address this issue.

Abstract: LLMs have demonstrated unprecedented capabilities in natural language processing, yet their practical deployment remains hindered by persistent factuality and faithfulness hallucinations. While existing methods address these hallucination types independently, they inadvertently induce performance trade-offs, as interventions targeting one type often exacerbate the other. Through empirical and theoretical analysis of activation space dynamics in LLMs, we reveal that these hallucination categories share overlapping subspaces within neural representations, presenting an opportunity for concurrent mitigation. To harness this insight, we propose SPACE, a unified framework that jointly enhances factuality and faithfulness by editing shared activation subspaces. SPACE establishes a geometric foundation for shared subspace existence through dual-task feature modeling, then identifies and edits these subspaces via a hybrid probe strategy combining spectral clustering and attention head saliency scoring. Experimental results across multiple benchmark datasets demonstrate the superiority of our approach.

Primary Area: Applications (e.g., vision, language, speech and audio, Creative AI)

Submission Number: 12545