When Context Leads but Parametric Memory Follows in Large Language Models

TL;DR

This paper examines how large language models balance local context and global parameters when answering questions, revealing consistent reliance patterns and the impact of context size on hallucinations, with implications for model robustness.

Contribution

Introduces the WikiAtomic dataset and systematically analyzes how nine LLMs prioritize context versus parametric knowledge in knowledge-consistent scenarios.

Findings

Models rely approximately 70% on context and 30% on parameters.

Increasing context size reduces hallucinations.

Patterns are consistent across different models.

Abstract

Large language models (LLMs) have demonstrated remarkable progress in leveraging diverse knowledge sources. This study investigates how nine widely used LLMs allocate knowledge between local context and global parameters when answering open-ended questions in knowledge-consistent scenarios. We introduce a novel dataset, WikiAtomic, and systematically vary context sizes to analyze how LLMs prioritize and utilize the provided information and their parametric knowledge in knowledge-consistent scenarios. Additionally, we also study their tendency to hallucinate under varying context sizes. Our findings reveal consistent patterns across models, including a consistent reliance on both contextual (around 70%) and parametric (around 30%) knowledge, and a decrease in hallucinations with increasing context. These insights highlight the importance of more effective context organization and developing models that use input more deterministically for robust performance.