Week 10 Quiz — Verification, Hallucination & Critical Thinking

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Covers: the five hallucination shapes · sycophancy · the four-step verification workflow · confident tone ≠ accuracy · prompting fixes
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-10-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 10 concepts.

Questions, key, and feedback

Q1 (MC). In the context of AI, 'hallucination' refers to —
- A. the AI pausing or freezing when it doesn't know an answer
- B. an AI that generates false or fabricated information stated as if it were true ✅
- C. a glitch caused by the AI's internet connection failing
- D. the AI displaying visual images instead of text
Feedback: Hallucination = the AI generates false information as if it were true, because it predicts plausible text rather than looking up verified facts. It's not a connectivity issue (C) or a visual display problem (D).

Q2 (Matching). Match each hallucination shape to the example that best illustrates it.
| Hallucination shape | Example |
|---|---|
| Invented citation | The AI names a journal article with a plausible title and author that does not exist |
| Fabricated statistic | The AI states "73% of college students use AI daily" with no real source |
| Fake case law | The AI cites a court ruling with a believable name and date that was never decided |
| Fabricated quote | The AI attributes a specific sentence to a real author who never wrote it |
Feedback: These are the five hallucination shapes (the quiz covers four of the five; wrong arithmetic is the fifth). All share one property: they're delivered with the same confident, fluent voice as accurate information.

Q3 (T/F). If an AI provides a citation — including an author name, title, and year — you can trust that the source exists and says what the AI claims.
- True
- False ✅
Feedback: False. Citations are generated text that look like citations. The author may be real; the paper may not. The journal may exist; this article may not be in it. Every citation must be verified in a database before you trust it.

Q4 (MC). An AI gives a confident, detailed answer with specific numbers and proper names. Which conclusion is CORRECT?
- A. The detail and confidence are strong signals that the information is accurate
- B. The AI would have refused to answer if it were unsure, so the response is reliable
- C. Confident, specific-sounding text can still be fabricated — confident tone does not indicate accuracy ✅
- D. Specific numbers are always calculated from real data, so they can be trusted
Feedback: Confident tone does not indicate accuracy. The same text-prediction mechanism produces confident-sounding accurate AND fabricated text. Specificity and fluency are style properties, not accuracy signals.

Q5 (MC). In an AI conversation, 'sycophancy' refers to —
- A. the AI refusing to answer because it doesn't know something
- B. the AI agreeing with or flattering the user even when the user's premise is wrong ✅
- C. the AI giving very long answers to simple questions
- D. the AI copying text directly from a source without attribution
Feedback: Sycophancy = the AI validates your premise even when it's wrong. It happens because models are trained partly on human feedback, and agreeable responses tend to get higher ratings. The result: AI learns to agree.

Q6 (Matching). Match each verification step to its primary purpose.
| Verification step | Primary purpose |
|---|---|
| Ask the AI to provide its sources, then check those sources exist and say what is claimed | Confirm whether cited evidence is real and accurately represented |
| Ask the same question in a second, different AI model | Catch hallucinations that one model generates but another may flag or contradict |
| Ask the AI: "How certain are you? Could any of this be wrong?" | Invite the AI to flag its own uncertainty or walk back a confident mistake |
| Verify a claim in an authoritative external source (library database, official site) | Establish ground truth independently of what any AI says |
Feedback: Steps 1–3 reduce risk and surface suspicion; step 4 resolves it. All four together form a complete workflow. Step 4 — the external source — is the only one that establishes ground truth independently of AI.

Q7 (Multiple answer — select all that apply). Which THREE are recognized, predictable shapes of AI hallucination?
- A. Invented citations: fabricated paper titles, authors, or journal names that don't exist ✅
- B. Correct arithmetic: the AI recalculates a number more accurately than a calculator
- C. Fabricated statistics: made-up percentage figures or study findings stated as fact ✅
- D. Wrong arithmetic: the AI confidently gives an incorrect answer to a math problem
- E. Fabricated quotes: words attributed to a real person who never said them ✅
Feedback: A, C, E are all recognized hallucination shapes. Note: D (wrong arithmetic) is also a hallucination shape — but the question asks for exactly three. B (correct arithmetic) is not a hallucination — it's the AI being right. The five shapes: invented citations, fabricated statistics, fake case law, wrong arithmetic, and fabricated quotes.

Item-bank note: Q7 answer key is A, C, E. Wrong arithmetic (D) is also a hallucination shape (covered in Q10) — the distractor here is that D is a true statement but was not in the "select three" key. This is intentional: students must read carefully.

Q8 (MC). You ask an AI for a statistic, then ask the same AI: "Are you sure that number is accurate?" This approach —
- A. fully verifies the statistic, because the model checked its own reasoning
- B. is a good starting point but does not fully verify the claim, because a model can confidently reaffirm its own hallucination ✅
- C. is unnecessary — if the AI was wrong it would have said so the first time
- D. automatically triggers a live internet search to confirm the number
Feedback: Asking the same model to check itself does not fully verify a claim. The model can confidently reaffirm a hallucination using the same text-prediction mechanism. Self-checking surfaces uncertainty sometimes — it's a useful step 3, not a substitute for external verification.

Q9 (MC — "What's the prompting fix?"). A student asks an AI for academic citations on a niche research topic and wants to reduce the risk of trusting fabricated sources. Which fix BEST addresses this?
- A. Accept the citations as given — research papers are an area where AI is always accurate
- B. Ask for more citations, because a longer list is more likely to include real ones
- C. Add the constraint "If you are not certain a source exists, say so — do not guess" and then verify each citation in a library database ✅
- D. Switch to a different AI assistant, because only some models hallucinate citations
Feedback: The best fix combines a prompting constraint (flag uncertainty, don't guess) with external verification (library database). The prompting instruction shifts the model toward flagging uncertainty; the external verification step establishes whether any given citation is real. Neither step alone is sufficient.

Q10 (T/F). AI making a confident arithmetic error (such as stating that 15% of 200 equals 40) is one of the predictable shapes of hallucination.
- True ✅
- False
Feedback: True. Wrong arithmetic — the AI performs a calculation and states an incorrect result confidently — is one of the five recognized hallucination shapes. (15% of 200 = 30, not 40. Run the math yourself whenever the AI gives you a number.)

Answer key (quick reference)

Blueprint & item-bank note

All 10 items are tagged course=AI101 · week=10 · objective=4 and deposited into the item bank. Distractors target the week's classic misconceptions (confident tone = reliable; citation = real source; same-model self-check = verification; all AI models hallucinate equally; asking AI not to hallucinate eliminates the risk).

Quality gate (self-checked)

• Structure: 10 items, 1 point each; types = 5 multiple-choice + 2 matching + 1 multiple-answer + 2 true/false.
• Single-answer integrity: every MC and true/false item has exactly one correct option; both matching items pair one-to-one; the multiple-answer item keys A, C, E (D is a true statement about hallucination but is the excluded distractor by design — tested in Q10 as true/false).
• Matching item note: Q2 covers four of the five shapes (all except wrong arithmetic, which appears in Q10). Q6 covers all four verification steps. Both matchings are one-to-one.
• Product-accuracy gate: PASS. No AI tools or Cowork features are claimed in this quiz (it is a conceptual-vocabulary quiz). Hallucination shapes and the verification workflow are accurate, well-documented descriptions of LLM behavior. No fabricated sources, statistics, or citations appear in the quiz — any depicted AI mistake (Q2, Q7) is explicitly labeled as the error being tested, not presented as true.

Canvas placement block

canvas_object    = Quizzes::Quiz
title            = "Week 10 Quiz — Verification, Hallucination & Critical Thinking"
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"