Studying Large Language Model Behaviors Under Context-Memory Conflicts With Real Documents

TL;DR

This paper investigates how large language models handle conflicting information from real documents in retrieval-augmented generation, revealing biases and limitations in updating factual knowledge.

Contribution

It introduces a realistic framework for studying knowledge conflicts in LLMs using real documents, highlighting biases and the impact of parametric knowledge on reading behavior.

Findings

Knowledge updates fail less often than previously thought.

Parametric bias increases likelihood of update failure.

Factual parametric knowledge can hinder reading and updating abilities.

Abstract

Retrieval-augmented generation (RAG) mitigates many problems of fully parametric language models, such as temporal degradation, hallucinations, and lack of grounding. In RAG, the model's knowledge can be updated from documents provided in context. This leads to cases of conflict between the model's parametric knowledge and the contextual information, where the model may not always update its knowledge. Previous work studied context-memory knowledge conflicts by creating synthetic documents that contradict the model's correct parametric answers. We present a framework for studying such knowledge conflicts in a realistic setup. We update incorrect parametric knowledge using real conflicting documents. This reflects how knowledge conflicts arise in practice. In this realistic scenario, we find that knowledge updates fail less often than previously reported. In cases where the models still fail to update their answers, we find a parametric bias: the incorrect parametric answer appearing in context makes the knowledge update likelier to fail. These results suggest that the factual parametric knowledge of LLMs can negatively influence their reading abilities and behaviors. Our code is available at https://github.com/kortukov/realistic_knowledge_conflicts/ .