LSTN is an experimental text generation engine that models language not as static probabilities, but as a dynamic and "liquid" neural network. Each trigram (3-character sequence) acts as an individual neuron within a continuous temporal dynamic system.