GitHub is using machine learning to enhance its "good first issues" feature, designed to help new contributors find beginner-friendly tasks in open-source projects. Initially launched in 2019, this feature relied on manually curated labels to identify suitable issues, but only covered about 40% of recommended repositories. To expand coverage to roughly 70%, GitHub introduced a machine learning model that predicts suitable issues by analyzing issue titles and bodies, using both classical and deep learning methods such as neural networks implemented in TensorFlow. The model employs a weakly-supervised approach with a heavily imbalanced training set, and prioritizes precision over recall to avoid false positives. Data processing and inference run daily, ensuring recommendations remain current. Future plans include improving classifier models, enhancing repository recommendations, and allowing maintainers to manage ML-based suggestions, while also exploring personalized recommendations for contributors.