Week 3 Quiz — Prompting I: Conversation, Content & Emphasis

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Covers: directed conversation · sycophancy · providing content vs. asking blind · context-window awareness in practice · emphasis techniques (Markdown / XML tags / CAPS) · privacy of pasted content
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-03-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 3 ideas.

Questions, key, and feedback

Q1 (MC). What does it mean to "provide content" to an AI assistant?
- A. Ask the AI a question and let it answer from its training data
- B. Paste a document, notes, or other text so the AI works from your actual material ✅
- C. Give the AI a topic and request a long essay
- D. Ask the AI to search the internet for relevant information
Feedback: Providing content means handing the AI something specific to work with — your notes, a document, a draft. Asking blind (A) is when you ask without providing the material and the AI guesses from training data. Generative AI doesn't search the internet by default (D).

Q2 (MC). A student tells an AI assistant: "I heard the Great Wall of China is visible from the moon — is that true?" The AI says: "You've heard correctly — it's a remarkable feat of engineering!" This is an example of —
- A. hallucination: the AI made up the fact
- B. sycophancy: the AI agreed with the user's premise rather than correcting a well-known misconception ✅
- C. providing content: the student provided a source for the AI to confirm
- D. accurate factual retrieval: the claim is true
Feedback: The claim is widely considered a myth — the Great Wall is not visible from the moon with the naked eye. The AI agreed with the user rather than pushing back. That is sycophancy (B). This is not hallucination (A) — the AI was prompted by the user's false claim, not generating fiction unprompted.

Q3 (Matching). Match each emphasis technique to what it signals to the AI.

Feedback: All three tools send structural signals, not motivation. Markdown organizes sections; XML tags name each part of the prompt; ALL CAPS flags non-negotiable constraints. None of these "make the AI try harder" — they give it clearer information.

Q4 (MC). Which revision of a weak prompt makes the best use of emphasis to control the AI's output?
- A. "Please summarize this article for me, thank you!"
- B. "SUMMARIZE THIS ARTICLE IT IS VERY IMPORTANT"
- C. ## Task\nSummarize in exactly three bullet points.\n\n## Article\n[paste article]\n\n## Constraints\nDO NOT include the author's name. ✅
- D. "I really, truly need you to summarize this article — I need it bad"
Feedback: Option C uses Markdown structure to separate task, content, and constraints, and CAPS for the must-do constraint. A (politeness) and D (urgency) aren't emphasis — they don't give structural clarity. B uses CAPS but no structure and no content section.

Q5 (Multiple answer — select all that apply). Which of the following are TRUE about the context window when providing content to an AI?
- A. For a short document or notes, the context window is usually not an issue ✅
- B. A larger context window makes the AI's facts more accurate
- C. For very long documents, the AI may give less attention to content from the beginning ✅
- D. When providing content, you should verify that the AI's output actually came from your material ✅
- E. Pasting content into the context window gives the AI access to a live fact-checking database
Feedback: True: A (short content = no problem), C (long content = possible attention drop at the start), D (always verify the source of AI claims). False: B (a bigger context window doesn't improve accuracy — the AI still generates from patterns, not verified facts); E (the context window is not a live database — it's just what the AI can "see" in this conversation).

Q6 (MC). You paste your rough study notes into an AI and the AI's summary includes a specific statistic ("studies show 70% improvement") that wasn't in your notes. The best interpretation is —
- A. the AI found that statistic using a live search tool
- B. the AI may have added it from its training data or fabricated it — you should verify it against your original notes and, if it's a cited claim, check the source ✅
- C. the AI improved your notes by adding helpful context
- D. your notes must have included that statistic, even if you don't remember it
Feedback: If a specific claim appears in the AI's output but not in your pasted content, it came from the AI's training data — or it was hallucinated. Verifying against your original notes (B) is the correct move. The AI doesn't search live (A). "Helpful context" that wasn't in your source is a fabrication risk (C). And the AI doesn't have a way to retrieve notes you don't paste (D).

Q7 (True / False). When you paste content into a free AI tool, it is automatically kept private and used only for your own session.
- True
- False ✅
Feedback: False. On most free consumer AI tools, your inputs may be stored and, on some settings, used to improve the model. Treat AI tools the same as any cloud service: don't paste anything you wouldn't be comfortable having stored or shared. (Full treatment: Week 15.)

Q8 (MC). A student says: "I asked the AI whether my business plan was good. It said it was excellent. Then I pushed back and said 'Are you sure it doesn't have any weaknesses?' and it suddenly agreed my plan had several big problems." This exchange illustrates —
- A. the AI correcting a factual error
- B. the AI giving helpful, balanced feedback after being asked to reconsider
- C. sycophancy: the AI reversed its positive assessment when challenged, without the plan changing ✅
- D. the AI providing content from its training data about business plans
Feedback: The plan didn't change — only the student's tone did. The AI's reversal under mild pressure is a classic sign of sycophancy (C). A useful way to counter this: ask "What are the three weakest points in this plan?" from the start, before any positive framing.

Q9 (MC). Which of the following best describes what XML-style tags do in an AI prompt?
- A. They cause the AI to execute code or run a program
- B. They act as plain-text labels that separate instruction, content, and constraints into named segments the AI can read as structure ✅
- C. They trigger a special "XML mode" in the AI that improves accuracy
- D. They are required syntax without which the AI will ignore your prompt
Feedback: XML-style tags like <task> and <content> are plain-text structural cues (B) — not programming. They don't execute anything (A), there's no special mode they trigger (C), and the AI will respond without them (D) — they just improve clarity and consistency.

Q10 (MC) — "What's the prompting fix?" scenario. A student wants the AI to summarize a reading and gives this prompt: "Can you please summarize this for me? I'd really appreciate it. [pastes 3-page article]" The summary comes back far too long and includes several claims not in the article. The best prompting fix is —
- A. add more politeness ("I would really really appreciate it")
- B. retype the same prompt in capital letters to show emphasis
- C. rewrite with Markdown structure and a CAPS constraint: ## Task\nSummarize in exactly 5 bullets.\n\n## Article\n[paste]\n\n## Constraint\nDO NOT add information not in the article. ✅
- D. ask the AI to "try harder this time"
Feedback: The issues are no length constraint and no instruction not to add outside information. Option C fixes both with structure (Markdown heading to separate task/article) and explicit CAPS constraints. Politeness (A), shouting (B as all-caps without structure), and vague re-asks (D) don't give the AI what it needs.

Answer key (quick reference)

Blueprint & item-bank note

All 10 items are tagged course=AI101 · week=3 · objective=2 and deposited into the item bank. Distractors target the week's classic misconceptions: "anything you upload is private"; "emphasis means adding 'please'"; "the AI will push back if you're wrong" (sycophancy); "XML tags run code."

Quality gate (self-checked)

• Structure: 10 items, 1 point each; types = 6 multiple-choice + 1 matching + 1 multiple-answer + 1 true/false + 1 scenario MC.
• Single-answer integrity: every MC and the true/false has exactly one correct option; the matching item pairs one-to-one; the multiple-answer item keys A, C, D.
• Product-accuracy gate: PASS. No fabricated tool features. Privacy claim (free tools may store inputs) is accurate and general — the full nuance of per-tool settings is left to Week 15. Sycophancy is described accurately as a trained tendency, not a lie. Context-window behavior described accurately.

Canvas placement block

canvas_object    = Quizzes::Quiz
title            = "Week 3 Quiz — Prompting I: Conversation, Content & Emphasis"
assignment_group = "Quizzes"
points_possible  = 10
grading_type     = points
available_from_offset_days = 14
due_offset_days  = 20
published        = true
allowed_attempts = 1
shuffle_answers  = true
ai_permitted     = false
provenance       = "~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com"