analiza aiqbench
This study provides a mechanistic analysis of how large language models internally resolve conflicting facts stored in their parameters, revealing that distinct neural circuits often encode competing pieces of knowledge rather than relying on a single universal mechanism. The research highlights significant differences between synthetic and real-world conflicts, showing that targeted attention-head interventions are more effective at resolving internal inconsistencies than broad layer-wise adjustments. These findings offer critical insights into the structural limitations of current models when handling contradictory information without external context.
Where Knowledge Collides: A Mechanistic Study of Intra-Memory Knowledge Conflict in Language Models
arXiv:2601.09445v2 Announce Type: replace Abstract: In language models (LMs), intra-memory knowledge conflict arises when inconsistent information about the same subject is encoded within the model's parametric knowledge. Prior work has primarily focused on resolving conflicts between a model's internal knowledge and external sources, which is known as context-memory knowledge conflict, through approaches such as fine-tuning or knowledge editing, while the understanding of conflicts that arise internally remains largely unexplored. In this work, we design a framework to identify where internal conflicting knowledge is encoded within LMs. We test our framework on four LMs using both synthetic and real-world knowledge conflicts. We find that internal conflicts often arise and are resolved in the final layers across all models, but that interventions are markedly less effective on real-world knowledge conflicts. Targeted attention-head interventions outperform layer-wise ones, and a filtering analysis shows that heads specialized for a single competing fact are far more common in synthetic conflicts, helping explain this gap. Finally, we find no evidence of a single universal circuit for handling knowledge conflict. Instead, our results suggest that distinct circuits may separately encode competing pieces of knowledge, giving rise to conflict. Our results offer a first mechanistic account of intra-memory conflict resolution and highlight a substantial gap between synthetic and real-world settings.
Abstract:In language models (LMs), intra-memory knowledge conflict arises when inconsistent information about the same subject is encoded within the model's parametric knowledge. Prior work has primarily focused on resolving conflicts between a model's internal knowledge and external sources, which is known as context-memory knowledge conflict, through approaches such as fine-tuning or knowledge editing, while the understanding of conflicts that arise internally remains largely unexplored. In this work, we design a framework to identify where internal conflicting knowledge is encoded within LMs. We test our framework on four LMs using both synthetic and real-world knowledge conflicts. We find that internal conflicts often arise and are resolved in the final layers across all models, but that interventions are markedly less effective on real-world knowledge conflicts. Targeted attention-head interventions outperform layer-wise ones, and a filtering analysis shows that heads specialized for a single competing fact are far more common in synthetic conflicts, helping explain this gap. Finally, we find no evidence of a single universal circuit for handling knowledge conflict. Instead, our results suggest that distinct circuits may separately encode competing pieces of knowledge, giving rise to conflict. Our results offer a first mechanistic account of intra-memory conflict resolution and highlight a substantial gap between synthetic and real-world settings.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2601.09445 [cs.CL]
(or arXiv:2601.09445v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2601.09445 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 302 cuvinte
Subiecte conexe
Performance Profiling on AMD GPUs – Part 5: Profiling-Driven Kernel Optimization with an AI Code-Assist Tool
AMD · 16 iul. 2026
[AINews] not much happened today
Latent Space · 3 oct. 2026
TokenMem: Faithful Knowledge Injection for Frozen LLMs
arXiv · 28 iul. 2026
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · 18 sept. 2026