Demystify large language models — tokens, transformers, attention mechanisms, and why models sometimes 'hallucinate'.