Research of Neural Networks ChatGPT Used to Generate Code in Python Programming Language

• Published in: 2024 Intelligent Technologies and Electronic Devices in Vehicle and Road Transport Complex (TIRVED) pp. 1 - 6
• Main Authors: Leokhin, Yuri, Fatkhulin, Timur, Kozhanov, Mikhail
• Format: Conference Proceeding
• Language: English
• Published: IEEE 13.11.2024
• Subjects: artificial intelligence; automation; Chatbots; Codes; Complexity theory; development; Measurement; methodology; neural network; Neural networks; Optimization; program code; Programming; programming language; Python; Roads; Software development management
• DOI: 10.1109/TIRVED63561.2024.10769794

The article deals with the problem associated with the influence of a prompt composed by users on the efficiency of the generated program code. The aim of the research is to work out a methodology allowing to analyze the operation and to estimate the effectiveness of online neural network services generating program code. The object of the study is neural network online services that allow generating program code. The subject of the study is the quality indicators of the generated program code. The relevance of the study is due to the increasing use of neural network technologies, including software development. To achieve this goal, the article considers online neural network services "ChatGPT 3.5", "ChatGPT 4" and "ChatGPT 4o", that allow generating program code in a programming language Python in question-and-answer form, besides, in this study, a methodology has been developed that allows determining the efficiency of neural network tools used to generate program code. In order to test this methodology, tasks of different levels of complexity were introduced into the above-mentioned neural network services and the program code generated by them was analyzed. In conclusion, a comparative analysis of the experimental results is conducted, the advantages and disadvantages of the considered neural network techniques in generating program code are identified. The methodological basis of the article includes the following methods: theoretical analysis, description, comparison and experiment.