Helpnetsecurity GitHub Launches ReviewBench for AI Code Review Benchmarking
Article Content
- •ReviewBench evaluates AI code review tools using 219 pull requests.
- •Metrics reported include precision, recall, and F1 scores for comparison.
- •GitHub's Copilot code review tool demonstrated improved performance in tests.
GitHub has introduced ReviewBench, an open benchmark designed to evaluate the performance of AI code review tools. The benchmark includes 219 pull requests from 187 public repositories across 19 programming languages. It allows users to compare the effectiveness of different AI reviewers in identifying issues and minimizing false positives. ReviewBench reports metrics such as precision, recall, and F1 scores, helping developers understand the strengths and weaknesses of various AI code review agents. GitHub's Copilot code review tool has shown improvements in performance metrics when evaluated using ReviewBench. The benchmark is currently available in research preview on GitHub's website.
Ask AI about this cluster
Answers cite the sources they use