LLMs predict text sequences; prompt engineering shapes responses via specific instructions, examples, and temperature settings (0=deterministic, 1=creative).