When Context Leads but Parametric Memory Follows in Large Language Models
This addresses the problem of knowledge allocation and hallucination in LLMs for AI researchers, but it is incremental as it analyzes existing models without proposing new methods.
The study investigated how nine large language models allocate knowledge between local context and global parameters when answering open-ended questions, finding consistent reliance on contextual (around 70%) and parametric (around 30%) knowledge, with hallucinations decreasing as context size increased.
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.