Describir: Do Language Models Understand the Cognitive Tasks Given to Them? Investigations with the N-Back Paradigm