What about the other publishers whose content was used to train the model?