Prompting & fine-tuning
P10.llm-essentials.04 · Audience: guest, it-ml, language-pro · Prerequisites: Reliability & grounding
Take it one idea at a time: what prompting is and why wording changes the output so much, the difference between fine-tuning a model and just instructing it, and what professional prompt engineering really involves. Each step has a worked example. No maths required.
ⓘ Concept: One mental model to keep
Prompting steers the model at the moment of use without changing it; fine-tuning actually changes the model by training it on many examples. Reach for the cheap, flexible tool (a good prompt) first — and treat prompt design as writing a precise brief, not hunting for magic words.
Why it matters — Reach for the cheap, flexible tool first (a good prompt), and treat prompt design as writing a precise brief rather than hunting for magic words.
Prompting is giving the model instructions and context in natural language to steer its output. Phrasing matters enormously because the model conditions everything it produces on your exact words, tone, and examples. A clearer, more specific prompt narrows the space of likely continuations toward what you actually want.
Worked example — vague brief vs precise brief, same task
Both prompts ask the model to translate the same marketing line into French:
Vague prompt
"Translate this: Get started today!"Output: « Commencez aujourd'hui ! »
Correct, but generic — tone, audience, and formality are all guesses.
Precise brief
"You are a marketing copywriter for a premium
B2B brand. Translate to formal French (vous),
keep it punchy (≤ 5 words), keep the CTA feel.
Term base: 'Get started' → 'Lancez-vous'.
Text: Get started today!"Output: « Lancez-vous dès aujourd'hui. »
Hits the register, length, CTA feel, and your approved term.
Same task, very different result — driven entirely by how it was asked. What helps most: stating the role, audience, and constraints explicitly, and giving examples of the desired output (usually more effective than clever wording). This is why prompt design is a real skill for language professionals, not a gimmick.
⚡ Interview Ref — the quick-scan Reference face
In plain language: what prompting is and why wording changes the output so much, the difference between fine-tuning a model and just instructing it, and what professional prompt engineering really involves. No maths required.
1 — What prompting is, and why phrasing changes output (l9)
Prompting gives the model instructions and context in natural language; it conditions everything on your exact words, tone, and examples, so a clearer, more specific prompt narrows the likely continuations toward what you want. What helps most: state the role, audience, and constraints and give examples of the desired output — usually more effective than clever wording.
2 — Fine-tuning vs giving instructions (l12)
Prompting instructs at use time without changing the model; fine-tuning trains it on many examples so a behaviour/style/format is built in. Prompting is instant, flexible, cheap (your first tool); fine-tuning is slower and needs data
- upkeep (a consistent voice/format at scale). Fine-tuning teaches how to behave, not fresh facts — prompt first.
3 — Professional prompt engineering, beyond tricks (l27)
At a professional level it is clear specification (task, audience, constraints, format) + good examples + systematic iteration and testing — closer to writing a precise brief and a test plan than to finding magic words. Done well it becomes a documented, reusable asset the team reuses.
Interview one-liners
- Prompting steers at use time; the model conditions on your exact words + examples, so specificity wins.
- Fine-tuning changes the model (teaches behaviour, not facts) — worth it for a consistent voice/format at scale; prompt first.
- Professional prompt engineering = a precise brief + examples + a test plan, documented and reused — not magic words.
📚 Go Further
Plain-language explainers on prompting and adaptation.
| Type | Resource |
|---|---|
| Guide | OpenAI / Anthropic prompt-engineering guides — role, examples, structure |
| Article | Language Models are Few-Shot Learners — Brown et al. (2020) — why examples steer output |
| In-app | P10.llm-systems-frontier.04 — Alignment & adaptation, for how fine-tuning and RLHF specialise a model at the systems level |
Ask the mentor about this module
Ask a question about this content. The mentor explains and grounds its answer in what you are studying; asking is recorded as a learning signal, not a grade.
Where next?
Go up a level