It could be, but without evidence that remains a just so story, no particular reason to think it's required or useful or even harmless to the model performance or anything other than an artifact of some early silicon valley writing style being injected into the model and continuous retraining on the output of older models.