Does chatGPT update its training set with its own conversations? If so, shouldn't we be really careful about tormenting and publicly embarrassing it lest it remember?