The phenomenon where a machine learning model loses previously learned information upon learning new information, particularly in neural networks and language models