Where Knowledge Collides: A Mechanistic Study of Intra-Memory Knowledge Conflict in Language Models
This addresses a foundational issue in AI interpretability for researchers, though it is incremental as it builds on prior work on knowledge conflict resolution.
The paper tackles the problem of intra-memory knowledge conflict in language models, where inconsistent information is encoded during pre-training, by developing a framework using mechanistic interpretability to localize and control these conflicts, demonstrating causal intervention at inference time.
In language models (LMs), intra-memory knowledge conflict largely arises when inconsistent information about the same event is encoded within the model's parametric knowledge. While prior work has primarily focused on resolving conflicts between a model's internal knowledge and external resources through approaches such as fine-tuning or knowledge editing, the problem of localizing conflicts that originate during pre-training within the model's internal representations remain unexplored. In this work, we design a framework based on mechanistic interpretability methods to identify where and how conflicting knowledge from the pre-training data is encoded within LMs. Our findings contribute to a growing body of evidence that specific internal components of a language model are responsible for encoding conflicting knowledge from pre-training, and we demonstrate how mechanistic interpretability methods can be leveraged to causally intervene in and control conflicting knowledge at inference time.