Better AI code comment detector

A researcher rebuilt an AI comment classifier using public data instead of private data, achieving 77% balanced accuracy in distinguishing human-written from AI-generated code comments. The new classifier is accessible as a web tool that provides calibrated confidence scores and interactive feature visualization, allowing users to see which text elements influenced its predictions. The classifier's false positive rate drops to 5% when confidence is high (80%+), making it reliable for confident predictions while maintaining transparency about uncertainty.