It could also be that there is a dirth of high quality CSS training data in comparison to JavaScript et al.

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