First it will dream up the interaction frame by frame. Next, to improve efficiency, it will cache those interaction representations. What better way to do that

First it will dream up the interaction frame by frame. Next, to improve efficiency, it will cache those interaction representations. What better way to do that than through a code representation. While I think current AI can’t come close to anything remotely usable, this is a plausible direction for the future. Like you, I shudder.