Uma revisão sistemática sobre Modelos de Linguagem Ampla (LLM’s) aplicados a geração de código

Abstract

Large Language Models (LLMs) for code generation represent a significant advancement in programming, boosting efficiency and speed in software development, simplifying repetitive processes, facilitating automated testing, and promoting the adoption of best practices. This work aims to map studies related to LLMs for code generation, using systematic literature review as the research method. The analysis of these studies deepens the understanding of the potential of these innovations. By highlighting the benefits and challenges associated with the use of LLMs, this study contributes to expanding knowledge and continuous advancement in this constantly evolving field. The systematic review protocol involved searching for articles in the Google Scholar database, using keywords related to LLMs and code generation. A total of 112 articles were initially found, of which 15 were selected based on relevance and quality criteria. Of the 15 selected articles, 8 were used for detailed comparative analysis, providing a solid foundation for evaluating different approaches and results. The remaining 7 articles served as an introductory theoretical framework, offering a comprehensive and well-founded overview of the essential theories and concepts that support the field of study.