Where Knowledge Collides: A Mechanistic Study of Intra-Memory Knowledge Conflict in Language Models

Minh Vu Pham, Hsuvas Borkakoty, Yufang Hou · 2026

Read Where Knowledge Collides: A Mechanistic Study of Intra-Memory Knowledge Conflict in Language Models by Minh Vu Pham, Hsuvas Borkakoty, Yufang Hou (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-09445.txt.

Description

Minh Vu Pham, Hsuvas Borkakoty, and Yufang Hou look inside language models for clashes in what they know. Their 2026 arXiv paper studies intra-memory knowledge conflict, where one model stores inconsistent facts about the same event. Earlier work mostly handled clashes between a model's knowledge and outside sources.

Fixes like fine-tuning or knowledge editing target that outside-facing problem. Far less is known about conflicts baked in during pre-training, inside the model's own representations. The authors study how to localize such conflicts mechanically. Knowing where a contradiction lives is a first step toward models that can catch their own mixed-up facts.

Catalog record

CREATORMinh Vu Pham, Hsuvas Borkakoty, Yufang Hou
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-09445

Browse the full library · About the Tower of Babel