ke-creates-im-and-increases-hallucination
IN premise — summaries/2026/08/24/xu-2024-knowledge-conflicts-survey-s2-context-memory-conflict.md
Created 2026-08-25T02:59:02+00:00
Knowledge editing (e.g., ROME, MEMIT) can create intra-memory conflicts and increase hallucination as a side effect (Li et al. 2023f; Pinter & Elhadad, 2023).
Summary
Patching a language model's stored facts with direct-editing methods can leave the model's internal representations in a contradictory state, like writing over one word in a sentence and breaking the meaning of the whole sentence. The practical concern is that "fixing" a single wrong fact may make the model less reliable overall, because it tends to fabricate details more often as a side effect of the edit.