Evaluating Syntactic Robustness of Large Language Model-based Code Generation

Large language models (LLMs) like GPT-3.5 and GPT-4 are not syntactically robust for code generation tasks, but their syntactic robustness can be significantly improved using a prompt pre-processing step that simplifies the mathematical formulas in the prompts.