Comparative Analysis of Loop-Free Function Evaluation Using ChatGPT and Copilot with C Bounded Model Checking

Advanced machine learning models and automated coding helpers have significantly transformed software development and verification techniques in recent years. This paper performs a comparative investigation of two prominent AI-driven code generation tools, ChatGPT and Copilot, with a specific emphasis on their ability to evaluate loop-free functions. By employing C Bounded Model Checking (CBMC) as the verification framework, we use model checking (MC) to systematically compare the accuracy and effectiveness of code produced by both tools. Our approach entails creating function implementations that are free of loops by utilizing established requirements with the assistance of ChatGPT and Copilot. These implementations are then subjected to thorough examination using CBMC to evaluate aspects such as functional correctness, safety, and potential vulnerabilities. Performance measurements encompass coding accuracy, verification time, and error identification. Results reveal notable differences in the performance of ChatGPT and Copilot. While both tools show promise in code generation, distinct strengths and weaknesses emerge in handling complex specifications and ensuring code correctness. This work provides useful insights into coding assistants powered by artificial intelligence and emphasizes the significance of incorporating formal verification methods, such as CBMC, to ensure dependable software development. This comparative analysis adds to the expanding research on AI-assisted programming, offering practical guidance for developers and researchers seeking efficient and dependable code generation tools.