This research analyzes how conversational LLMs like ChatGPT perform when automatically fixing bugs in software code.

• Evaluates the effectiveness of LLM-powered automated program repair (APR)
• Identifies patterns in both successful and failed repair attempts
• Examines how conversational capabilities enhance iterative patch improvement
• Explores limitations that still exist in current LLM-based repair systems

For engineering teams, this research provides valuable insights into when and how to leverage LLMs in development workflows, potentially reducing debugging time while understanding where human intervention remains necessary.

Studying and Understanding the Effectiveness and Failures of Conversational LLM-Based Repair