I am not sure if we can interpret the language output like they were human. What inner state were the models in? What inner state were the text to illicit?