Knowledge-based Consistency Testing of Large Language Models

TL;DR

This paper introduces KonTest, an automated framework that uses knowledge graphs to test and improve the consistency and knowledge coverage of large language models, revealing significant gaps and errors.

Contribution

We propose KonTest, a novel knowledge-based testing framework that systematically exposes inconsistencies and knowledge gaps in large language models and offers a mitigation strategy.

Findings

KonTest identifies 19.2% error-inducing inputs in tested LLMs.

It reveals a 16.5% knowledge gap across models.

Mitigation reduces LLM knowledge gap by 32.48%.

Abstract

In this work, we systematically expose and measure the inconsistency and knowledge gaps of Large Language Models (LLMs). Specifically, we propose an automated testing framework (called KonTest) which leverages a knowledge graph to construct test cases. KonTest probes and measures the inconsistencies in the LLM's knowledge of the world via a combination of semantically-equivalent queries and test oracles (metamorphic or ontological oracle). KonTest further mitigates knowledge gaps via a weighted LLM model ensemble. Using four state-of-the-art LLMs (Falcon, Gemini, GPT3.5, and Llama2), we show that KonTest generates 19.2% error inducing inputs (1917 errors from 9979 test inputs). It also reveals a 16.5% knowledge gap across all tested LLMs. A mitigation method informed by KonTest's test suite reduces LLM knowledge gap by 32.48%. Our ablation study further shows that GPT3.5 is not suitable for knowledge-based consistency testing because it is only 60%-68% effective in knowledge construction.