AI Can’t Adulterate its Own Writing
My friend John Gruber is incensed by the idea that AI watermarking schemes adjust the output of a given LLM so that the output can later be assessed as likely coming from it in particular: They say “imperceptible” and “doesn’t change the me…