Week 5 Quiz — Prompting III: Examples, Structure & Control

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Covers: zero/one/few-shot definitions · examples for voice, format, PII · control toolkit (count, structure, constraints, expansion, regenerate, sources, guidance) · catching drift
Format: 10 auto-graded items (multiple-choice, multiple-answer, matching, true/false) · 10 points (1 each) · allowed attempts: 1 · No AI on this quiz.

This is the human-readable quiz with its vetted answer key and one-line feedback. The import-ready Classic QTI 1.2 is in F-quiz-week-05-qti.xml (generated by a validated Python script — parses with 10 items, every single-answer item exactly one correct). Reminder: AI is not permitted on quizzes — this checks that you understand the Week 5 ideas.

Questions, key, and feedback

Q1 (MC). A "zero-shot" prompt is one that —
- A. includes zero words so the AI can fill in everything
- B. gives the AI no examples and asks it to perform the task from the instruction alone ✅
- C. uses exactly zero constraints so the output is unrestricted
- D. provides zero context, making the AI guess what you need
Feedback: Zero-shot = no examples before the task. The AI relies on the instruction alone. It says nothing about how many words, constraints, or how much context is provided.

Q2 (MC). Which statement about "few-shot" prompting is CORRECT?
- A. Few-shot means you provide exactly one example to guide the AI
- B. Few-shot means you provide a small number of examples — typically two to five — so the AI learns a pattern from them ✅
- C. Few-shot and zero-shot are the same technique with different names
- D. Few-shot means the AI generates a few short outputs and you pick one
Feedback: Few-shot = several examples (typically two to five). One example is one-shot, not few-shot. This distinction matters: more examples help the AI generalize the pattern rather than over-fitting to a single sample.

Q3 (Matching). Match each prompting technique to its best description.
| Technique | Best description |
|---|---|
| Zero-shot | Task given with no examples — AI uses only the instruction |
| One-shot | Task given with exactly one example to set the pattern |
| Few-shot | Task given with several examples (typically 2–5) to teach format or voice |
| Regenerate | Ask the AI to produce a new version — does not fix underlying facts |
Feedback: The shot vocabulary is a count of examples: zero, one, or a few. Regenerating is a control move — useful for variety, not for correcting factual errors.

Q4 (MC). A student pastes three examples of her own social-media captions and asks an AI to write two more in the same voice. This is an example of —
- A. zero-shot prompting, because she is not explaining the rules
- B. few-shot prompting using examples to teach a specific voice or style ✅
- C. meta-prompting, because she is asking the AI to write about writing
- D. regenerating, because she is asking for new versions of existing content
Feedback: Three examples = few-shot. The examples demonstrate the voice; no verbal description of "the rules" is needed — that's the advantage of few-shot. Meta-prompting (asking the AI to help write the prompt) is a different Week 4 technique.

Q5 (True/False). True or False: If you ask an AI to "provide sources," the links and citations it lists are guaranteed to be real, retrievable documents.
- True
- False ✅
Feedback: False. AI assistants can produce perfectly formatted, plausible-sounding citations that do not exist. Every link must be opened and verified independently — a fabricated DOI or journal name is a common AI failure mode.

Q6 (MC). A student prompts: "Write me a short bio." The AI returns a generic five-paragraph bio that is too long and formal. What is the BEST prompting fix?
- A. Regenerate — the AI will automatically improve it the second time
- B. Paste two or three examples of the exact length and tone you want, then ask again (few-shot) ✅
- C. Add the word "please" at the start and resubmit
- D. Use a different AI tool — this one clearly cannot write bios
Feedback: The best fix is to show the AI what you want with examples — few-shot prompting. Regenerating will likely produce another generic bio. Politeness and switching tools don't address the root problem: the AI doesn't know your specific format or voice.

Q7 (MC). You ask an AI for a list of five research articles and the citations look suspicious. You click "Regenerate." What is MOST LIKELY to happen?
- A. The new response will contain the correct, verified citations
- B. Regenerating asks the model to search the internet for real articles
- C. The AI produces a different set of citations — which may be equally invented ✅
- D. Regenerating fixes the facts because the model rechecks its training data
Feedback: Regenerating produces a different output — the model doesn't go verify; it generates again from the same patterns. A new list of fabricated citations is still fabricated. The fix is manual verification, not regeneration.

Q8 (Multiple answer — select all that apply). Which of the following are legitimate techniques for controlling AI output structure and content? Select all that apply.
- A. Specifying a number ("give me exactly five bullet points") ✅
- B. Requesting a specific format ("return a Markdown table with two columns") ✅
- C. Setting a constraint ("do not use jargon; keep each item under 15 words") ✅
- D. Asking for expansion ("take point 3 and write a full paragraph") ✅
- E. Typing in all-caps to make the AI try harder
Feedback: A, B, C, and D are all real, effective control moves. E is false — using all-caps is an emphasis technique from Week 3 that can create attention, but it does not make the AI "try harder" in any meaningful sense. Count, format, constraints, and expansion are the control toolkit.

Q9 (MC). You want the AI to rewrite a sensitive email but you must remove personal details first. The BEST approach is to —
- A. paste the full original email so the AI can decide what is sensitive
- B. replace names, IDs, and specific details with placeholders (e.g., [NAME], [DATE]) before pasting, then ask for the rewrite ✅
- C. ask the AI to automatically detect and strip PII from the email
- D. use a paid AI plan, which protects all data by default
Feedback: You do the scrubbing before pasting. The AI may miss PII or infer it from context. Paid plans may offer better data policies, but they don't guarantee automatic PII detection. The placeholder method is the only approach that keeps sensitive data off the tool entirely.

Q10 (MC). After receiving a bulleted outline from an AI, you type: "Expand bullet 2 into a full paragraph with a concrete example." This technique is called —
- A. regenerating
- B. zero-shot prompting
- C. asking for expansion — a follow-up that deepens one part of a previous output ✅
- D. meta-prompting
Feedback: Asking for expansion deepens one specific part of a previous output — it's a different move from regenerating (which replaces the whole output) and from meta-prompting (which has the AI help design the prompt itself).

Answer key (quick reference)

Blueprint & item-bank note

All 10 items are tagged course=AI101 · week=5 · objective=2 and deposited into the item bank for future per-term ($39) regenerations. Distractors target the week's classic misconceptions (few-shot = exactly one example; regenerate fixes facts; requesting sources guarantees real ones; paid plans scrub PII automatically; all-caps = more effort).

Quality gate (self-checked)

• Structure: 10 items, 1 point each; types = 7 multiple-choice + 1 matching + 1 multiple-answer + 1 true/false. Includes ≥1 matching item (Q3: technique→description) and ≥1 "what's the prompting fix?" scenario item (Q6).
• Single-answer integrity: every MC and the true/false item has exactly one correct option; the matching item pairs one-to-one; the multiple-answer item keys A, B, C, D (E is the distractor).
• Classic confusion addressed: Q2 and Q3 directly engineer around the "few-shot means exactly one example" confusion called out in the brief.
• Product-accuracy gate: PASS. No AI tool features are described or asserted; all quiz content covers prompting techniques and general AI behavior that are accurate across tools. No fabricated features, statistics, or citations.

Canvas placement block

canvas_object    = Quizzes::Quiz
title            = "Week 5 Quiz — Prompting III: Examples, Structure & Control"
assignment_group = "Quizzes"
points_possible  = 10
grading_type     = points
available_from_offset_days = 0
due_offset_days  = 6
published        = true
allowed_attempts = 1
shuffle_answers  = true
ai_permitted     = false
provenance       = "~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com"