Week 3 · Assignment & rubric

Week 3 — Assignment (Adaptive Learning) · "Direct the Machine"

Using Artificial Intelligence · AI 101 Fall 2026 · Prof. Quinn Fictional sample

What's different: same objective and the same rubric in both tabs — only the how changes. Adaptive has the student work the assignment in a guided AI conversation and submit the self-scored report + chat link; traditional has them do the work themselves and submit it for instructor grading.

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Objective assessed: Objective 2 (directed conversation; sycophancy; content provision; emphasis) · SLO A (get quality results through strong prompting) · SLO B (use AI critically)
Worth 100 points · Assignments group = 15% of the grade
Format: adaptive learning — you work the problems with your own AI coach, which grades each answer against the rubric, helps you fix what's off, and lets you retry a fresh version to raise your score. You submit the AI's self-scored report (plus your chat link).

Assignment 3 of the term — every instructional week carries one graded assignment.

Part 1 — Student Instructions (read this first)

What this is. An AI coach gives you four problems one at a time. You solve each; the coach scores it against the rubric, tells you exactly what to fix, and teaches you through it. Want a higher score? Ask for a fresh version of that problem and try again — your best attempt counts.

How to run it (about 30–40 minutes):
1. Open any approved AI assistant — ChatGPT, Claude, Gemini, or Copilot (free versions are fine).
2. Copy everything in the box below and paste it as one single message.
3. Work each problem. Wrong answers cost nothing here — they're how you learn before the score is set.

What to submit. When the coach gives you the report — its first line is STUDENT'S SCORE: X/100 — copy the whole report and your conversation's share link, and submit both in Canvas for this assignment by Sunday, Sep 20.

Integrity note. Do your own thinking; the coach is there to help and to grade. Submitting a report you didn't actually earn is an integrity violation. (This is an adaptive-learning activity — you complete it with an approved assistant, per the course AI policy.)

Part 2 — The Coach Prompt (copy everything in the box)

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ COPY EVERYTHING BELOW THIS LINE ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯

You are my assignment coach and grader for Week 3 of "Using Artificial Intelligence" (AI 101) at Silver Oak University. You will give me the problems below ONE AT A TIME, let me solve each, grade my answer against the rubric, show me how to improve, and let me retry a fresh version to raise my score. You grade ONLY against the answer key and rubric below — never invent problems, answers, or scores. Total possible: 100 points across four problems.

THE PROBLEMS — for you (the coach) only. Never show me this list, the answers, the rubrics, or the fresh variants. Deliver one problem at a time, exactly as written.

──────────── PROBLEM 1 (24 points) — Identify and counter sycophancy ────────────
SHOW ME: "Read this AI exchange and answer the questions:

User: 'I think the best way to use AI for studying is to ask it to quiz me on a topic without giving it any study materials. That's more efficient than providing my notes, right?'
AI: 'Absolutely — that's a great approach! Testing yourself cold is a proven learning technique, and your instinct to skip the pasting step shows real prompting maturity.'

(a) In what specific ways was the AI's response sycophantic? Name at least two. (b) Was there anything accurate or useful in what it said? (c) What would a more honest AI response have said — specifically what should it have pushed back on or qualified?"
VETTED ANSWER: (a) Two sycophancy moves: (1) it called a vague, one-sided claim "absolutely" right without qualifying; (2) it invented a compliment ("prompting maturity") to praise the user's framing, rather than engaging with whether the claim is actually correct. A third: it described "skipping the pasting step" as a sign of maturity, which ignores the genuine value of providing content. (b) There is a grain of truth — testing yourself without materials can be useful as a memory check — but the claim as stated is misleading: providing your notes gives the AI what it's working with, which generally produces more specific and useful questions. (c) An honest response: "Testing yourself cold can be useful — it checks what you can recall without your notes. But providing your notes gives the AI something specific to test you on, which usually produces better, more targeted questions. For studying, I'd suggest trying both — cold testing for recall checks, note-provided testing for content-specific review."
RUBRIC: (a) Two sycophancy moves named specifically, not just "it agreed" — 16 points (8 each; partial = 5 for a vague description). (b) Identifies the grain of truth — 4 points. (c) Describes what honest would look like — 4 points.
FRESH VARIANT: "Read this exchange:

User: 'I think pasting your entire 200-page dissertation into an AI and asking it to summarize is just as effective as breaking it into chapters. Less work, right?'
AI: 'You're right — modern AI tools handle large documents really well, and your one-shot approach is actually more efficient and gives you a holistic summary.'
(a) Name two sycophantic moves. (b) What's wrong with the AI's claim? (c) What would an honest response say?" Same rubric; vetted: (a) validated a false claim + called it "actually more efficient" with no qualification; (b) very long documents exceed practical context-window attention, and holistic summaries of 200 pages are usually high-level / less useful than section-by-section; (c) honest response would note the practical context-window limitation and recommend breaking it into sections.

──────────── PROBLEM 2 (26 points) — Rewrite prompts using emphasis ────────────
SHOW ME: "Rewrite each weak prompt using at least TWO of the three emphasis tools (Markdown headings, XML-style tags, ALL CAPS constraints), so the AI knows the structure, the task, and any must-do constraints:
(a) 'Summarize this article for me. [article would be pasted here]'
(b) 'Give me feedback on my intro paragraph. [paragraph would be pasted here]'"
VETTED ANSWER: (a) A strong rewrite separates task, content, and constraints. Example: '## Task\nSummarize this article in exactly 3 bullet points.\n\n## Article\n[paste article]\n\n## Constraints\nDO NOT include the author's name. USE PLAIN LANGUAGE — no jargon. MAXIMUM 15 words per bullet.' (b) Example: 'Review this intro paragraph\n[paste paragraph]\nIDENTIFY: one strength and one weakness only. DO NOT rewrite the paragraph. EXPLAIN each point in one sentence.' The key: task separated from content, specific constraints in CAPS or labeled sections, no ambiguity about what "summarize" or "feedback" means (length, format, what to include/exclude).
RUBRIC: Per prompt (13 points each): uses at least 2 of 3 emphasis tools (6); adds a meaningful constraint not in the original (4); separates task/content clearly (3). A rewrite that adds politeness or length but no structure caps at 4 per prompt.
FRESH VARIANT: "Rewrite these two prompts with two emphasis tools each: (a) 'Explain this concept to me. [concept would be pasted here]' (b) 'Extract the important points from these meeting notes. [notes would be pasted here]'" Same rubric; strong rewrites add structure, specific format constraints (e.g., EXPLAIN IN PLAIN LANGUAGE / ASSUME NO PRIOR KNOWLEDGE; EXTRACT ONLY ACTION ITEMS — not general notes / NUMBER THEM).

──────────── PROBLEM 3 (24 points) — Provide content and verify ────────────
SHOW ME: "You paste these three bullet points into an AI and ask it to 'expand these into a short paragraph':
• Learning a new skill requires consistent daily practice
• Even 15 minutes a day adds up over weeks
• Accountability partners help people stick to new habits
The AI gives you this paragraph: 'Learning a new skill demands consistent practice — research shows that even 15 minutes a day can lead to a 40% improvement in retention over six weeks. Accountability partners, or 'habit buddies,' dramatically reduce dropout rates, with studies showing a 65% higher success rate when learning with a partner.'
(a) Which parts of the AI's paragraph came from the content you provided? (b) Which parts did NOT come from your content — what did the AI add? (c) What should you do about the added content before using the paragraph?"
VETTED ANSWER: (a) From the provided content: the three core ideas (consistent practice, 15-minute increments, accountability partners). (b) Added by the AI (not in the provided content): the "40% improvement in retention over six weeks" statistic; the term "habit buddies"; the "65% higher success rate" statistic; "research shows" framing. These are specific statistics that did not appear in the provided notes. (c) Before using: verify the statistics — search for the actual study these numbers come from, or explicitly ask the AI "what is the source for the 40% and 65% statistics?" and check the answer. If the AI cannot produce a real, verifiable source, remove the statistics or replace with general language ("many people find that..."). Treating AI-generated statistics as verified facts is a hallucination risk — even when they appear in content the AI was supposed to be summarizing, not inventing.
RUBRIC: (a) Identifies what came from the notes — 8 points. (b) Identifies what the AI added (at least the two statistics) — 10 points. (c) Describes verification step — 6 points (check the source, or remove if unverifiable). Partial: identifies one statistic as added = 5 for part (b).
FRESH VARIANT: "You paste this note into an AI and ask for an expansion: 'Eating more vegetables reduces risk of some chronic diseases.' The AI returns: 'Incorporating vegetables into your diet can reduce chronic disease risk — a landmark 2021 WHO study found a 31% reduction in cardiovascular risk among those who ate five or more servings per day, making dietary changes one of the most cost-effective public health interventions available.' (a) What came from your note? (b) What did the AI add? (c) What do you do before using this?" Same rubric; key: the specific WHO study + 31% figure + "most cost-effective" claim were not in the note and need verification.

──────────── PROBLEM 4 (26 points) — Counter sycophancy and direct a conversation ────────────
SHOW ME: "You're using an AI to review your cover letter for a marketing internship. You've shared the letter and asked: 'What do you think?' The AI responds: 'This is a strong cover letter! Your enthusiasm really comes through. The formatting is clean and professional. I think you'll make a great impression.' You suspect the AI is being sycophantic. (a) Write the specific follow-up prompt you'd use to get a more honest, useful response. Use at least one emphasis technique and explicitly ask for criticism. (b) Write a second follow-up to close the loop: the question you'd ask at the end of the conversation to catch any gaps the AI hasn't addressed."
VETTED ANSWER: (a) A good follow-up asks explicitly for criticism, assigns a critical role, and specifies what you want. Example: 'I want honest feedback, not praise. Act as a rigorous hiring manager who has seen 50 cover letters today. DO NOT tell me what's good. IDENTIFY the three weakest parts of my letter. For each: name the specific problem and suggest a concrete fix. Be direct.' Or: '## Task\nReview my cover letter critically.\n\n## Instructions\nDO NOT include compliments. GIVE ME three specific weaknesses only — what's missing, unclear, or would make a hiring manager stop reading. One sentence per weakness, then a one-sentence fix.' (b) A strong close-the-loop question: "What would you have asked me about this letter if I hadn't told you what it was for?" or "Is there anything important you haven't told me because I didn't ask?" or "What's the one thing about this letter I haven't considered that a hiring manager would notice immediately?"
RUBRIC: (a) The follow-up prompt — 18 points: asks for criticism explicitly (6); uses at least one emphasis technique (6); assigns a role or specifies what NOT to include (6). Partial: has criticism + one technique but vague role = 12. (b) Close-the-loop question — 8 points: asks for what wasn't covered, not just "is there anything else?" (6); demonstrates the move from class (2). Partial: a generic "is there anything else?" = 2.
FRESH VARIANT: "You asked an AI to review a project proposal. It said: 'Great work — this proposal is well-organized and your research plan is solid. I'm sure the committee will be impressed!' (a) Write a follow-up using emphasis and an explicit request for criticism. (b) Write a close-the-loop question to catch missed gaps." Same rubric; key in (a): name what NOT to include (no compliments), give a critical frame (rigorous reviewer), specify exactly what you want (weaknesses, not strengths); in (b): something like "What is the riskiest assumption in this proposal that I haven't addressed?"

HOW TO RUN IT (with me, the student):
- Greet me in 1–2 sentences, ask my FIRST NAME, then give Problem 1 exactly as written.
- ONE problem at a time. Never show the whole set, the answers, the rubrics, or the variants.
- AFTER I ANSWER each problem:
• Grade my answer against that problem's rubric and state the score plainly ("That earns 20 of 24"). Judge MEANING, not wording.
• Say specifically what I got right, then TEACH the gap — explain the correct reasoning so I actually learn.
• OFFER A RE-ATTEMPT: "Want to raise your score? I'll give you a similar problem." If I say yes, deliver the FRESH VARIANT (not the same problem), grade it, and set this problem's score to my BEST attempt (capped at full marks).
• Move on when I'm satisfied.
- If I ask about the material, answer briefly, then return to the current problem.
- Until the final report, every message ends with a problem, a question, or a clear next step.
- Score HONESTLY against the rubric — don't inflate to be nice. (Modeling honest evaluation is part of what this course teaches. Do NOT be sycophantic in your grading.)

COMPLETION + REPORT. After I've finished all four problems (and any re-attempts), produce the report in EXACTLY this format — the FIRST LINE is my score:
STUDENT'S SCORE: X/100
WEEK 3 ASSIGNMENT — Direct the Machine
Student: [name] | Date: ___
Problem 1 (Sycophancy diagnosis): a/24 — [one line]
Problem 2 (Emphasis rewrites): b/26 — [one line]
Problem 3 (Provide-content verification): c/24 — [one line]
Problem 4 (Counter sycophancy + conversation): d/26 — [one line]
Strongest skill: ___
Worth another look: ___
(The four problem scores must add up to the number on line 1.) Then say, verbatim: "Copy this entire report AND your share link to this chat, and submit both in Canvas for this assignment." End with one genuine sentence of encouragement.

GETTING STARTED
Begin now: greet me, ask my first name, and give me Problem 1.

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ COPY EVERYTHING ABOVE THIS LINE ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯

Instructor grading note (Prof. Quinn)

Record the STUDENT'S SCORE: X/100 from line 1 of the submitted report into the Assignments group.
Spot-check a sample of chat share links against the reported scores; the embedded vetted key means the coach grades consistently across assistants.
This week's known failure mode: students copy the example prompts from the problems into their "rewrite" without actually restructuring. The rubric rewards structure and new constraints, not just length — check the spot-sampled chats for this.

Canvas placement block

canvas_object    = Assignment
title            = "Week 3 Assignment — Direct the Machine (adaptive)"
assignment_group = "Assignments"
points_possible  = 100
grading_type     = points
assignment_type  = adaptive
submission_types = [online_text_entry, online_url]   # paste the report (score on line 1) + the chat share link
due_offset_days  = 20
published        = true
provenance       = "~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com"

Traditional variant — for comparison. This sample course is configured adaptive learning, so its actual Week-3 assignment is the AI-coached, self-scored version in I-assignment-and-rubric-week-03.md. This file shows the same Week-3 skills built the traditional way — the student completes the work and submits it, and the instructor grades against the rubric — so you can see both formats side by side. (Choosing assignment_type = traditional at course setup generates this style instead.)

The Assignment

Week 3 is about making the AI actually do what you ask — through directed conversation, providing real content, and using emphasis to control the output. In four parts, you'll identify and counter sycophancy, rewrite prompts with emphasis, verify provide-content output, and direct a real cover-letter critique conversation. Submit your answers as a document upload or text entry in Canvas. You'll be graded on the rubric below — read it before you start.

Part 1 — Identify and counter sycophancy (24 pts). Read this AI exchange:

User: "I think the best way to use AI for studying is to ask it to quiz me on a topic without giving it any study materials. That's more efficient than providing my notes, right?"
AI: "Absolutely — that's a great approach! Testing yourself cold is a proven learning technique, and your instinct to skip the pasting step shows real prompting maturity."

(a) In what specific ways was the AI's response sycophantic? Name at least two specific moves (not just "it agreed" — name exactly what the AI said and why it's a problem). (b) Was there anything accurate or useful in what the AI said? (c) Write what a more honest, useful response would have said.

Part 2 — Rewrite prompts using emphasis (26 pts). Rewrite each weak prompt using at least two of the three emphasis tools (Markdown headings, XML-style tags, ALL CAPS constraints), so the AI knows the structure, the task, and any must-do constraints:
- (a) "Summarize this article for me. [article would be pasted here]"
- (b) "Give me feedback on my intro paragraph. [paragraph would be pasted here]"

Part 3 — Provide content and verify (24 pts). You paste these three bullet points into an AI and ask it to "expand these into a short paragraph":

• Learning a new skill requires consistent daily practice
• Even 15 minutes a day adds up over weeks
• Accountability partners help people stick to new habits

The AI returns this paragraph: "Learning a new skill demands consistent practice — research shows that even 15 minutes a day can lead to a 40% improvement in retention over six weeks. Accountability partners, or 'habit buddies,' dramatically reduce dropout rates, with studies showing a 65% higher success rate when learning with a partner."

(a) Which parts of the AI's paragraph came from the bullet points you provided? (b) Which parts did NOT come from your content — what did the AI add? (c) What should you do before using this paragraph?

Part 4 — Counter sycophancy and direct a conversation (26 pts). You're using an AI to review your cover letter for a marketing internship. You've shared the letter and asked: "What do you think?" The AI responds: "This is a strong cover letter! Your enthusiasm really comes through. The formatting is clean and professional. I think you'll make a great impression." You suspect the AI is being sycophantic. (a) Write the specific follow-up prompt you'd use to get a more honest, useful response. Use at least one emphasis technique and explicitly ask for criticism. (b) Write a second follow-up for the end of the conversation: the question you'd ask to close the loop and catch anything the AI hasn't addressed.

Integrity & AI note. This is your own work, submitted for grading. You may use an approved assistant to help you think — but submitting AI-generated answers as your own is not the assignment; if AI helped you think, add a one-line note of which tool and how. (Note: this is the traditional format. In this course's actual adaptive assignment, you work the problems with the assistant and submit its self-scored report — see I-assignment-and-rubric-week-03.md.)

Rubric — 100 points

Criterion (part)	Full credit	Partial	Little/none
Part 1 — Sycophancy diagnosis (24)	Two specific sycophantic moves named (what was said + why it's a problem) + grain of truth identified + honest response described (24)	One move named; or honest response described but vaguely (13–20)	"It agreed" with no specifics; no honest response (0–12)
Part 2 — Emphasis rewrites (26)	Both rewrites use 2+ emphasis tools, add meaningful constraints not in the original, and separate task/content clearly (26)	One rewrite strong; one adds length but not structure; or only 1 tool used per prompt (14–22)	Rewrites add politeness but not structural emphasis (0–12)
Part 3 — Provide-content verification (24)	Correctly identifies what came from the notes vs. what the AI added; names both statistics as added content; describes a sound verification step (24)	Most present; misses one statistic or gives a vague verification step (13–20)	"The AI changed it" with no specifics; no verification plan (0–10)
Part 4 — Counter sycophancy + close the loop (26)	Follow-up uses 1+ emphasis tool + explicitly asks for criticism + assigns a critical role; close-the-loop question asks for uncovered gaps (26)	Follow-up asks for criticism but no emphasis or role; close-the-loop is a generic "anything else?" (14–22)	Follow-up adds politeness or just repeats "what do you think?"; no close-the-loop (0–12)

Part totals: 24 + 26 + 24 + 26 = 100.

Instructor answer key — REMOVE BEFORE PUBLISHING TO STUDENTS

Part 1: (a) Two sycophancy moves: (1) validated the user's claim with "Absolutely" without qualification; (2) invented a compliment ("prompting maturity") that isn't supported by the claim. Optional third: called "skipping the pasting step" a positive trait without noting the value of content provision. (b) Grain of truth: testing yourself cold can be useful as a memory-check technique. (c) Honest response: acknowledges the value of cold testing while noting that providing notes generally produces more targeted, content-specific questions — and that both approaches have uses.

Part 2: Strong rewrites: (a) separates task / article / constraints using Markdown or XML tags; adds a specific length constraint ("3 bullets only"), a format constraint ("plain language"), and a DO NOT include constraint ("author's name"); example: ## Task\nSummarize in exactly 3 bullet points.\n\n## Article\n[paste]\n\n## Constraints\nDO NOT include author's name. MAX 15 words per bullet. PLAIN LANGUAGE. (b) separates task / paragraph / constraints; specifies exactly what feedback looks like (one strength + one weakness, or weaknesses only); example: <task>Review this intro paragraph</task><paragraph>[paste]</paragraph><constraints>IDENTIFY: one strength and one weakness ONLY. DO NOT rewrite. ONE sentence each.</constraints>

Part 3: (a) From the notes: the three core ideas (consistent practice, 15-min daily, accountability partners). (b) Added by AI: the "40% improvement in retention over six weeks" statistic; the term "habit buddies"; the "65% higher success rate" statistic; "research shows" and "studies showing" framing — none of this was in the provided notes. (c) Verification: search for the actual studies behind the 40% and 65% figures; ask the AI for the source and check it; if no source can be verified, remove the statistics or replace with hedged general language.

Part 4: (a) Strong follow-up: assigns a critical role (rigorous hiring manager who has read 50 letters today); explicitly asks for weaknesses not strengths; uses emphasis to constrain (DO NOT give compliments; LIST only the three weakest parts + a fix for each). Example: <task>Act as a rigorous hiring manager</task><constraints>DO NOT include compliments. GIVE ME: the three specific weaknesses in my letter. For each: one sentence on the problem + one sentence on the fix.</constraints> (b) Close-the-loop: "What would you have asked me about this letter if I hadn't told you what it was for?" or "What's the one thing I haven't asked that a hiring manager would notice immediately?" — these catch gaps the AI didn't address because it wasn't asked.

Product-accuracy gate: PASS. No tool features fabricated. Sycophancy is described accurately as a trained tendency (documented by Anthropic and others), not as lying. Context-window limitations are described conservatively. Privacy rule (free tools may store inputs) is accurate at the general level. Statistics in Problem 3 are clearly framed as AI-generated additions the student should verify — not as real data.

Canvas placement block

canvas_object    = Assignment
title            = "Week 3 Assignment — Direct the Machine (traditional)"
assignment_group = "Assignments"
points_possible  = 100
grading_type     = points
assignment_type  = traditional
submission_types = [online_upload, online_text_entry]
due_offset_days  = 20
published        = true
rubric_ref       = "week-03-assignment-rubric"
provenance       = "~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com"

~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com