To be fair, most of the training data likely came from GitHub.