The rise of Large Language Models (LLMs) over the past years and their use in software development has raised questions and concerns about the quality of the code they generate. While LLMs such as ChatGPT have the potential to accelerate the development of new software and boost code productivity, their ability to generate high-quality code is debatable. This study presents an investigation into the code quality that is generated by LLMs, specifically focusing on iOS applications generated by ChatGPT.

To investigate the code quality of the generated iOS applications, the study implemented three movie applications in Swift with ChatGPT-3.5 and ChatGPT-4, along with a similar application written by a human developer. To measure the code quality, a set of code quality metrics was chosen to make a fair evaluation of the results.

The findings showed that ChatGPT can generate iOS applications with code quality comparable to a human-written application, and ChatGPT-4 in particular showed this capability. However, the ChatGPT-3.5 model showed inconsistent results, suggesting potential limitations in the current model. The level of human intervention required for development varies depending on the complexity of the task, with simpler tasks requiring less intervention. Overall, the study suggests that human intervention and guidance are essential for developing a working application.

Keywords: Code generation, Code Quality, Generative AI, iOS code quality, iOS development, Large Language Models