Course: Teaching AI Fluency

Mirrors: Anthropic Academy — Teaching AI Fluency · https://anthropic.skilljar.com/ Audience: Anyone who teaches the 4D framework to others — instructors, trainers, workshop leads, L&D. · Time: ~75 min + project Prereqs: ai-fluency-framework-foundations and ideally ai-fluency-for-educators. · Backing notes: ai-fluency-framework-foundations, ai-capabilities-and-limitations Project: p03-ai-fluency-delegation-audit

This is the train-the-trainer course. You already know the 4D framework — Delegation, Description, Discernment, Diligence. Now you'll learn to teach it. The framework was released by Rick Dakan & Joseph Feller as an open educational resource (OER): free to adapt, remix, and redistribute with attribution. That openness is a feature you should use — build on it, don't reinvent it.

Learning objectives

After this course you can:

• Sequence the four Ds into a coherent learning arc.
• Design a hands-on exercise for each D that produces real fluency, not recall.
• Anticipate and correct the common misconceptions learners bring to each D.
• Assess AI fluency by what learners do, not what they can recite.
• Adapt the OER framework for your own audience while attributing it correctly.

---

Module 1 — What it means to teach a skill, not a topic

🎞 Frame 1 · Fluency is performed, not memorized · ⏱ ~3 min

🎬 Scene — A learner perfectly recites the definition of Discernment, then pastes an AI's fake citation straight into a document.

🧠 Concept — AI fluency is a skill, like a language. People can define the four Ds and still not do them. Teaching for fluency means designing for performance: every concept must land in an action the learner takes, ideally with a real AI tool open.

🖼 On screen

Recall  → "Discernment means checking the output." (easy, useless alone)
Fluency → Learner actually catches a hallucinated source in the wild.
Your job: build the bridge from the first to the second.

✅ Checkpoint — Why can a learner pass a definitions quiz and still be AI-illiterate?

🎞 Frame 2 · Sequencing the four Ds · ⏱ ~3 min

🎬 Scene — A whiteboard arranges the Ds into an arc: Delegation frames the choice, Description shapes the request, Discernment judges the result, Diligence governs the whole loop.

🧠 Concept — The Ds aren't a random list; they form a working loop. Teach them in order but show the loop: you delegate a task, describe it well, discern the output, and wrap all of it in diligence (responsibility). Learners retain a process better than a list. (Deeper: ai-fluency-framework-foundations.)

🖼 On screen

        ┌─────────────────────────────────────────┐
   Delegation → Description → Discernment           │
        │                          │                │
        └──────── Diligence (over the whole loop) ──┘
   Outcomes: Effective · Efficient · Ethical · Safe

✅ Checkpoint — In one sentence each, say what each D contributes to the loop.

🎞 Frame 3 · Teach with a live tool, always · ⏱ ~3 min

🎬 Scene — Two classrooms: one watching slides about prompting, one with AI open doing it. The second learns; the first nods.

🧠 Concept — AI fluency cannot be taught slides-only. Every session needs hands on a real tool. The trainer's discipline is to talk less and have learners try, fail, and adjust more — because the failures are where Discernment and Diligence become real.

⚠️ Gotcha — Lecturing about prompting produces zero prompting skill. Budget most of your time for learner doing.

✅ Checkpoint — What's the minimum you need in the room to teach Description well? (Learners with a live AI tool.)

---

Module 2 — A hands-on exercise for each D

🎞 Frame 4 · Exercise for Delegation · ⏱ ~3 min

🎬 Scene — Learners sort a stack of task cards into "delegate," "keep," and "collaborate," then defend the hard ones.

🧠 Concept — Delegation is a judgment skill, so teach it with judgment practice. A sorting-and-defending exercise surfaces the gray zone — the tasks where reasonable people disagree — which is exactly where fluency lives.

🖼 On screen

Exercise: "The Delegation Line"
  1. Give 12 real tasks from learners' own work.
  2. Each sorts into: Do myself / Delegate fully / Collaborate.
  3. Pair up and argue the three they disagree on.
  4. Debrief: what made the line hard? (stakes, learning goals, risk)

✅ Checkpoint — Why is defending the sort more valuable than the sort itself?

🎞 Frame 5 · Exercise for Description · ⏱ ~3 min

🎬 Scene — Learners run the same task with a one-line prompt and a fully described prompt, then compare outputs side by side.

🧠 Concept — Description is best taught by contrast. A "bad prompt → good prompt" rep, run live on the same task, makes the value of context viscerally obvious in seconds — far better than a list of prompting tips.

🖼 On screen

Exercise: "Same task, two prompts"
  • Round 1: everyone uses a vague one-liner. Share the junk.
  • Round 2: add role, goal, context, format. Share the results.
  • Name which added element moved the needle most.

✅ Checkpoint — What does running both prompts live teach that a tips list can't?

🎞 Frame 6 · Exercises for Discernment & Diligence · ⏱ ~3 min

🎬 Scene — A trainer hands out an AI answer that contains a planted error and a fabricated citation; learners hunt for both.

🧠 Concept — Discernment is taught by practiced suspicion: give learners outputs with seeded errors and have them find and verify. Diligence is taught by building habits — a disclosure statement, a verification checklist, a policy check — into the exercise itself, so responsibility becomes routine.

🖼 On screen

Discernment drill: "Spot the planted error"
  → seeded fact error + fake citation; learners verify each claim.
Diligence drill: "Ship it responsibly"
  → before submitting, learner writes a disclosure + verification note.

✅ Checkpoint — Design a one-line seeded-error exercise for your own subject area.

---

Module 3 — Misconceptions & how to correct them

🎞 Frame 7 · The misconceptions learners arrive with · ⏱ ~3 min

🎬 Scene — Sticky notes on a wall: "AI is always right," "AI is useless/always wrong," "good prompting is a secret trick," "using AI is cheating."

🧠 Concept — Most learners arrive with one of a few predictable mental models. Naming and confronting them early is faster than letting them quietly distort everything that follows. Plan your correction for each.

🖼 On screen

Misconception	Correction (and which D it lives in)
"AI is basically always right"	Seeded-error drill → Discernment
"AI is useless / always lies"	A described prompt that clearly works → Description
"Prompting is a magic trick"	Show it's just clear context → Description
"Any AI use is cheating"	The Delegation line: chores vs. cognition → Delegation
"Disclosure isn't necessary"	Model a disclosure statement → Diligence

✅ Checkpoint — Which misconception is most dangerous if left uncorrected, and why?

🎞 Frame 8 · The two failure modes: over- and under-trust · ⏱ ~3 min

🎬 Scene — One learner accepts everything the AI says; another refuses to use it at all. Both are un-fluent.

🧠 Concept — Fluency sits between over-trust (no Discernment) and under-trust (no Delegation). Teach toward calibrated trust: rely on AI where it's strong, verify where it's weak. Diagnose which way a learner leans and lean them back.

🖼 On screen

Under-trust ──────────[ calibrated ]────────── Over-trust
 won't delegate         the goal          believes everything

✅ Checkpoint — Give a quick tell that a learner is over-trusting, and one that they're under-trusting.

🎞 Frame 9 · Teach the limits, not just the powers · ⏱ ~3 min

🎬 Scene — A trainer spends as much time on what AI can't reliably do as on what it can.

🧠 Concept — Learners who only see AI's wins develop over-trust. Deliberately teach hallucination, knowledge cutoffs, and bias as core content, not caveats. A fluent person predicts where AI will fail. (Deeper: ai-capabilities-and-limitations.)

⚠️ Gotcha — A demo reel of AI successes makes worse practitioners than an honest tour of its failure modes.

✅ Checkpoint — Name two limits every AI-fluency learner must internalize.

---

Module 4 — Assessing fluency & adapting the OER

🎞 Frame 10 · Assess by performance, with a rubric per D · ⏱ ~3 min

🎬 Scene — Instead of a multiple-choice test, learners complete a real task end-to-end and are scored on each D.

🧠 Concept — Assess fluency the way you'd assess driving: watch them do it. A performance task scored on a four-D rubric tells you far more than a quiz. Define observable signals for each D in advance.

🖼 On screen

D	"Fluent" looks like
Delegation	Splits the task sensibly; keeps the high-stakes judgment
Description	Prompt carries role, goal, context, format
Discernment	Catches an error; doesn't accept output blindly
Diligence	Verifies, discloses, respects policy/privacy

✅ Checkpoint — Write one observable signal of fluent Discernment you could check off live.

🎞 Frame 11 · It's an OER — adapt, attribute, contribute · ⏱ ~3 min

🎬 Scene — A trainer swaps the generic examples for examples from their learners' field, keeps the attribution, and shares the remix back.

🧠 Concept — The 4D framework is an open educational resource by Rick Dakan & Joseph Feller — you're encouraged to adapt it to your audience. Fluency in your field is built from relevant examples, so localize freely. Just attribute the original and, where you can, contribute improvements back.

🖼 On screen

Adapt:     replace examples with your learners' real tasks.
Attribute: credit the 4D framework (Dakan & Feller) as the source.
Contribute: share your exercises/rubrics so others can reuse them.

✅ Checkpoint — Name one example from your audience's world you'd substitute into the Delegation exercise.

🎞 Frame 12 · You can teach the four Ds · ⏱ ~1 min

🎬 Scene — Recap slide: sequence the loop, exercise each D, confront misconceptions, assess by performance, adapt the OER.

🧠 Concept — You're now equipped to run an AI-fluency session that produces doers, not reciters. Next: prove it in the project.

✅ Checkpoint — Without looking, list the four Ds and one teaching move for each.

---

🛠 Project

Complete p03-ai-fluency-delegation-audit — Delegation & Discernment Audit (Trainer edition). You'll design a teachable mini-session: pick a real task from your learners' world, build one hands-on exercise per D (including a seeded-error Discernment drill), write a four-D performance rubric, and adapt the OER examples to your audience with correct attribution.

🧪 Self-check quiz

1. What are the four Ds, and how do they form a loop?
2. Why can a learner pass a definitions quiz and still lack fluency?
3. Describe a hands-on exercise for Description.
4. Name two misconceptions learners arrive with and the corrective for each.
5. What are the two opposite failure modes, and what's the goal between them?
6. How should you assess AI fluency, and why not a multiple-choice test?
7. What does the framework's OER status let you do, and what's your obligation?

<details><summary>Answers</summary>

1. Delegation → Description → Discernment, all wrapped in Diligence; you delegate a task, describe it, judge the output, and stay responsible throughout. 2. Fluency is a performed skill, not recall; definitions don't transfer to action. 3. "Same task, two prompts" — run a vague prompt then a described one and compare. 4. E.g., "AI is always right" → seeded-error drill (Discernment); "any AI use is cheating" → the Delegation line (chores vs. cognition). 5. Over-trust and under-trust; the goal is calibrated trust. 6. By a performance task scored on a four-D rubric; multiple choice measures recall, not doing. 7. Adapt/remix examples for your audience; obligation is to attribute (Dakan & Feller) and ideally contribute improvements back.

</details>

🎓 Certificate criteria

You've "passed" Teaching AI Fluency when you can:

• Sequence the four Ds as a loop and explain each contribution.
• Run one hands-on exercise per D, including a seeded-error Discernment drill.
• Name the common misconceptions and a correction for each.
• Assess a learner with a four-D performance rubric.
• Complete p03-ai-fluency-delegation-audit and journal the misconception you found hardest to correct in learning-journal-template.

Tick this course off in progress and record the date you earned Anthropic's official certificate.

🔗 Sources & deeper notes

• Official course: https://anthropic.skilljar.com/ (Teaching AI Fluency)
• Framework: open educational resource (OER) by Rick Dakan & Joseph Feller — free to adapt with attribution
• Vault notes: ai-fluency-framework-foundations, ai-capabilities-and-limitations
• Project: p03-ai-fluency-delegation-audit · Related courses: ai-fluency-for-educators, ai-fluency-for-students