It was interesting to me that OpenAI admitted (according to the article) that they used Bibliotik to train the models.