Why a prompt is unversioned system logic, zero-shot vs. few-shot, chain-of-thought's real cost and limits, system vs. user prompts, sampling parameters as a separate control surface, structured output, and prompt evaluation discipline.
Module 5 covered the alignment stages that make a pretrained model responsive to instructions at all; this module covers the practical skill of steering that aligned model's behavior without touching a single one of its weights. It treats prompt engineering as a genuine engineering discipline — a prompt functions as an unversioned piece of a system's logic, and small wording changes measurably shift output reliability — and works through zero-shot versus few-shot prompting, chain-of-thought reasoning and its real latency and validity trade-offs, and the distinct roles of system and user prompts.
By the end of this module you should be able to explain how temperature and top-p sampling differ from prompt wording as a control surface, why structured output matters once an LLM's response feeds code instead of a human reader, and why "it looked good when I tried it" is not evaluation. Module 7's AI agents content builds directly on prompting as the interface an autonomous reasoning loop runs through.