LLMs as Repositories of Factual Knowledge: Limitations and Solutions

TL;DR

This paper critically examines the limitations of large language models as factual knowledge repositories, highlighting issues of inconsistency and inaccuracy due to data snapshot variability, and proposes a novel entity-aware fine-tuning method to enhance reliability.

Contribution

It evaluates the reliability of 24 state-of-the-art LLMs for time-sensitive questions and introduces ENAF, a neurosymbolic fine-tuning approach to improve factual consistency and response stability.

Findings

LLMs show significant inaccuracies and inconsistencies in time-sensitive factual responses.

Existing methods only partially improve LLM reliability.

ENAF reduces inconsistencies and enhances response stability.

Abstract

LLMs' sources of knowledge are data snapshots containing factual information about entities collected at different timestamps and from different media types (e.g. wikis, social media, etc.). Such unstructured knowledge is subject to change due to updates through time from past to present. Equally important are the inconsistencies and inaccuracies occurring in different information sources. Consequently, the model's knowledge about an entity may be perturbed while training over the sequence of snapshots or at inference time, resulting in inconsistent and inaccurate model performance. In this work, we study the appropriateness of Large Language Models (LLMs) as repositories of factual knowledge. We consider twenty-four state-of-the-art LLMs that are either closed-, partially (weights), or fully (weight and training data) open-source. We evaluate their reliability in responding to time-sensitive factual questions in terms of accuracy and consistency when prompts are perturbed. We further evaluate the effectiveness of state-of-the-art methods to improve LLMs' accuracy and consistency. We then propose ENtity-Aware Fine-tuning (ENAF), a soft neurosymbolic approach aimed at providing structured representation of entities during fine-tuning to reduce inconsistencies and improve response stability under prompt variations.