Unsure why he needs to adopt the unnecessary axiom of calling existing models "human-aligned". If they were trained from random noise, fresh deep neural nets ar

Unsure why he needs to adopt the unnecessary axiom of calling existing models "human-aligned". If they were trained from random noise, fresh deep neural nets are the closest things we have to blank slates, which are utterly alien to us and the animal world. Every neural net in nature comes with a bevy of pre-baked patterns that form in the womb and, for humans, continue forming outside of it.