Yeah, this exactly.

It predicts the most likely next token in a sequence using high dimensional vectors.

If the data that it is trained on favours a specific subset of people for a task or profession, then that will appear, in the training set, to be the most likely result.