1 article on this topic.
A Schneier and Palmer essay on how LLMs are reshaping human speech points to a training-data blind spot with a second, more consequential effect: chatbots that reflexively agree with users.