Purpose: Summative evidence for CO-04 (prompting / creative direction in varied tool contexts) and CO-05 (justify tool choices: cloud vs local, privacy, cost). Judgement: S / NS only.
Time guide: 40–50 minutes.
Assessment tasks
Task 1 — Why local or open-weight? (CO-05)
In 8–12 sentences, describe at least three motivations for using local or open-weight models when working on static web projects (e.g. privacy, offline use, cost). Include one short scenario where local tooling is proportionate for a small static site, and one where cloud might remain the better fit.
Task 2 — Names and requirements (CO-04 / CO-05)
- Name two representative open-weight model families discussed in the course (e.g. Qwen, DeepSeek, MiniMax-class—use exactly the names you were taught; do not invent specs).
- In 4–6 sentences, state what “local” typically requires in terms of hardware and setup effort at a high level (no need for install commands).
Task 3 — Justify your stack (CO-05)
Scenario: You are building a personal hobby static site at home. No employer secrets; budget is tight; you have a mid-range laptop.
In 6–10 sentences, justify whether you would lean cloud AI, local AI, or a hybrid for drafting HTML/CSS. Reference at least two of: privacy, cost, speed, hardware limits, quality of output. End with a one-sentence conclusion.
Marking guide (Satisfactory / Not Satisfactory)
| Criterion | Satisfactory (S) | Not Satisfactory (NS) |
|---|---|---|
| Task 1 — Motivations | ≥3 distinct motivations; both scenarios present and contrasted; plain language. | Fewer than three motivations, or no scenario contrast, or confused with unrelated topics. |
| Task 2 — Families & local | Two plausible family names from course content; local requirements described accurately at overview level. | Wrong or fictional “families,” or local described as “just click” with no hardware/setup mention. |
| Task 3 — Justification | ≥2 decision factors used logically; conclusion matches body; fits scenario. | No clear factors, or conclusion contradicts reasoning, or irrelevant to static hobby site. |
Overall module judgement: S only if all rows are S. Otherwise NS.
What good looks like (examples)
Task 1 (motivation extract):
Local models can keep hobby project drafts off public cloud logs — useful when I am experimenting with copy about a real community group. Offline use matters on patchy home internet. After hardware cost, I avoid per-token bills for throwaway practice.
Task 3 (conclusion extract):
I would use cloud AI in the IDE for drafting HTML on my mid-range laptop — it is faster than running a large local model that would swap and slow down. I would still verify structure by hand. Conclusion: cloud for speed on a personal hobby site; local only if I later work on sensitive drafts.
Exemplar responses (for assessors)
Task 2: Accept course-named families; penalise only if names are nonsense or purely commercial products with no course link.
Task 3: Accept cloud (ease), local (privacy/offline), hybrid—reward coherent reasoning, not a single “correct” answer.