Template-guided interpretable reasoning with execution feedback for LLM-based program repair

Key Points

• The execution feedback facilitates better program repair outcomes, enhancing the reliability of large language models in coding tasks.
• Precision in execution feedback was demonstrated, with successful repairs increasing by 35% compared to traditional methods.
• Template-guided reasoning was utilized in an innovative algorithm to improve program repair accuracy across various coding languages.
• This approach may enable more interpretable outcomes in software development, with potential applications in educational tools.