Credit: Mike Silver
Artificial intelligence, trained on a vast dataset of written text, does a poor job in the back-and-forth of human-like conversation. The solution may be to train large language models more substantially on transcribed spoken conversations