Text this: Probing Maximally Divergent Conceptual Regions in Large Language Models Through Prompt Engineering