It’s not surprising that a language model trained on the entire history of human output can regurgitate some pseudo-spiritual slop.