Type: url
Original: https://arxiv.org/abs/2404.05427
Status: generated
Summary
The arXiv abstract page for AutoCodeRover, an automated approach for solving GitHub issues to autonomously achieve program improvement - the software maintenance and evolution work the authors distinguish from coding itself. It combines LLMs with code search over a program representation, the abstract syntax tree, exploiting classes and methods to sharpen the model’s understanding of an issue’s root cause, and uses spectrum-based fault localization where a test suite is available. On SWE-bench-lite (300 real-life GitHub issues) the authors report 19% efficacy at an average cost of $0.43 USD per issue, higher than the reported efficacy of SWE-agent.