[anktor]

Isn't the paper discussing precisely the opposite? That chatgpt predicts text in a way that, with each version, resembles more and more human logic, both with its succeseses and errors?