Parametric Knowledge is Not All You Need: Toward Honest Large Language Models via Retrieval of Pretraining Data

The paper introduces a novel benchmark dataset aimed at evaluating the honesty of large language models (LLMs) by leveraging the pretraining data from the open LLM, Pythia. It critiques existing methods for their lack of robustness in assessing LLM knowledge boundaries and proposes a method to enhance LLM honesty by enabling models to acknowledge their limitations instead of generating incorrect responses. This work is significant for practitioners as it provides a framework for developing more reliable LLMs that can better manage uncertainty and improve user trust.