ABSTRACT
The chapter explores methods to assess the credibility of information produced by large language models (LLMs). To address the inherent limitations of these models, a range of approaches are analyzed to mitigate credibility issues. The findings are derived from a literature review, case analysis, and experiments conducted with LLMs. The chapter contributes to the science of information management by proposing a comprehensive approach to obtaining credible information as a predictive outcome, starting with quality-oriented data preparation, followed by language model training, and culminating in advanced prompting of LLMs to generate accurate explanations. Management science practice is addressed by demonstrating the importance of explainability, defined as the ability to clarify how data inform decision-making. For fact-checking techniques such as tracking model training dynamics, analyzing attention maps, and employing advanced prompting strategies, such as Graphs of Thought, are discussed. Ensuring credible information is crucial for achieving Sustainable Development Goal 16, which emphasizes peace, justice, and strong institutions.
