The question is how machine will iterate the stuff. Like, it gets new info about surroundings and add to the code immediately and completely changing it's behavior on the outcome. Or just collects the data and then reprocess words into bigger salad.

Currently, gpt4 can return to original incorrect answers because it keeps iterating the salad until the user is satisfied.