I wouldn't be surprised if the negative developer sentiment toward CSS is reflected in training datasets.