Advanced Prompt Engineering

The five principles from the last lesson cover most everyday prompting. This lesson covers named, more deliberate techniques used in professional and technical work — the same ones companies use when building AI-powered products, not just chatting casually.

You give the AI a task with no examples, relying entirely on what it already learned during training. This works well for common, familiar tasks:

Zero-shot fails more often on unusual or highly specific tasks — that's exactly when the next technique helps.

You provide 2–5 example input/output pairs before the real task, showing the model the exact pattern you want:

Use few-shot whenever the model keeps giving inconsistent results or the wrong format — showing examples is far more reliable than describing the format in words.

Telling the model to think step by step before answering dramatically improves accuracy on reasoning, maths, and logic:

A system prompt is an instruction given to the AI before the actual conversation begins, setting its role, personality, and constraints. In a normal chat interface, you can simulate one by putting detailed instructions right at the start of the conversation:

The underlying AI model is identical either way — the system prompt is what turns a generic assistant into a specialised tool. This is exactly how companies build AI-powered products: a customer-service bot, a coding assistant, and a writing coach can all be the same base model with a different system prompt.

Professional workflows often need AI output in a specific format another tool can actually use — not free-flowing prose. Ask for it explicitly:

This turns an AI assistant from a chat tool into a data-processing step that can feed directly into other software — genuinely useful the moment you're combining AI with any other tool or workflow.

Sometimes it's just as important to say what you don't want: "explain this without using any technical jargon," or "list ideas, but don't include anything that requires a budget." Stating a constraint directly is often more reliable than hoping the model infers it.

Beyond the prompt itself, most AI tools — especially when used through an API or a "playground" — expose settings that control how the model picks its next word. Knowing these helps explain why the same prompt can give a different answer twice, and lets you dial an AI toward consistency or creativity on purpose.

You won't see all of these in every AI chat app — many hide them behind a simple "creative / balanced / precise" toggle. But the underlying idea is the same: temperature is the one worth knowing by name, since it's the setting you'll most often see exposed directly.

Chain-of-thought covers most everyday cases. A few related techniques go further for harder problems — worth recognising by name even if you reach for them rarely.

You now have a genuinely professional prompting toolkit. The next two lessons put all of this to work on one of AI's most visible capabilities: generating images.