xu2024-intra-memory-contradiction-rate

IN premise — summaries/2026/08/24/xu-2024-knowledge-conflicts-survey-sR-references-chunk-2.md

Created 2026-08-25T02:59:04+00:00

Intra-memory contradictory output probability ranges from 15.7% to 22.9% across models (Mündler et al. 2023), with stronger models producing fewer contradictions.

Summary

Language models contradict themselves within a single conversation roughly one time in six to one time in four, which means self-contradiction is a common baseline failure that any system built on top of them must explicitly detect and manage rather than treat as a rare edge case. The fact that larger models reduce but do not eliminate this rate suggests the problem is structural to how these systems generate text, not a simple bug that gets patched away.