Early Access Video: Training AI Without Writing A Reward Function, with Reward Modelling