Using Generative AI for Simplifying Official Documents in the Public Accounts Domain

This is the authors' abstract. We don't add key points for this paper.

Governments and societies have understood the benefits of transparency and access to information since the 1950s, aiming for greater social participation and even more control over corruption. In Brazil, initiatives such as the Access to Information Act and the Digital Government Act promote transparency. However, the complex language used in sectors of the public sector, especially the judiciary, prevents the consumption of this information by part of the population. In this context, this practical report aims to present a project to facilitate access to and consumption of the preliminary opinions published by the State Court of Accounts (TCEs), through automatic simplifications using Large Language Models (LLMs), especially the GPT-4. The automatically simplified texts contain the main points of the opinion, highlighting monetary values and percentages, with their respective explanations and laws that support the result of the preliminary opinion. The MVP generated is being validated by a state court of accounts for subsequent incorporation into its transparency framework.