I am not convinced, largely because ChatGPT has to have been trained on exactly this kind of pre-existing hallucinatory happy talk. It has always struck me as

I am not convinced, largely because ChatGPT has to have been trained on exactly this kind of pre-existing hallucinatory happy talk. It has always struck me as not coincidental that it writes, long-form, like a college admissions essay, apologises for massive error breezily like a tech support person, and has the confident certain tone of a Californian software developer. There were, after all, these humans involved its development.