LARGE LANGUAGE MODELS PARA RECUPERAÇÃO DE INFORMAÇÃO EM DOCUMENTOS DIGITALIZADOS um estudo com o modelo GPT-4

Technological advancements that have expanded access to digital information have driven scientific, technical, artistic, and cultural production. However, the vast amount of available information also presents challenges, particularly in retrieving relevant and accessible information for people with different needs andabilities. Digitized textual documents, common in institutional collections, amplify these challenges as they often lack machine-readable characters. This study investigated the use of the GPT-4 model for information retrieval in digitized documents from institutional repositories. The applied and exploratory research adopted a qualitative and quantitative approach to evaluate character recognition and semantic searches using a customized GPT. Twenty theses from the Federal University of Minas Gerais repository were analyzed using five prompts. The model achieved 98% precise and coherent responses, demonstrating high performance, although technical challenges still limit its large-scale application.