Turn pretrained language models into useful assistants through instruction datasets, supervised fine-tuning, conversation templates, and response formatting. Analyze how instruction quality changes model behavior.