Current generation AI technology requires partial training by real human beings in data centers using reinforcement learning from human feedback (RLHF), where humans select between model outputs to train preference signals into the system.