Preceptron model updating with back propagation has become the routine of deep learning. Continu-ous feed forward procedure is required in order for backward propagate to function properly. Doubt-ing the underlying physical interpretation on transformer based models such as GPT brought aboutby the routine explaination, a new method of training is proposed in order to keep self-consistencyof the physics. By treating the GPT model as a space-time diagram, and then trace the worldlinesof signals, identifing the possible paths of signals in order fot a self-attention event to occure. Witha slight modification, self-attention can be viewed as an ising model interaction, which enables thegoal to be designed as energy of system. Target is treated as an external magnetic field inducing sig-nals modeled as magnetic dipoles. A probability network is designed to pilot input signals travellingat constant speed through different routes. A rule of updating the probabilities is designed in orderto form constructive interference at target locations so that instantaneous energy can be maximised.Experiment is conducted on a 4-class classification problem extracted from MNIST. The results ex-hibit interesting but expected behavours, which do not exist in a bp updated network, but more likelearning in a real human, especially in the few-shot scenario.