1. What facilitation means here
The platform teaches. You do not lecture the content; every unit already contains the explanation, the demo, the quizzes, and the challenge. Your job is the five things a platform cannot do:
- Frame: open each session with why today's unit matters to this room.
- Run the room: manage devices, accounts, pace, and the live demos.
- Host the discussion: the units end in questions; you make them collide with real opinions.
- Verify the artifacts: every unit produces something; you check it exists and the student can defend it.
- Safeguard: privacy, age-appropriateness, and the emotional moments AI topics sometimes produce.
If you remember one sentence: never re-explain what the unit already explains. When a student is confused, send them back to the exact section, then discuss what they found. The course is designed to be the authority so you can be the coach.
2. The cohort at a glance
Audience: non-technical adults and teens 14-18, mixed comfort levels. Pace: 12 units over 8 weeks (default map below), or one unit per week for a 12-week school term. Session length: 90 minutes recommended, 60 possible if challenges go home.
Default week map:
| Week | Units | Session focus |
|---|---|---|
| 1 | 01 + 02 | Patterns and learning-from-examples; Quick Draw and Teachable Machine live |
| 2 | 03 + 04 | Bias and hallucination: the two big discussion sessions |
| 3 | 05 + 06 | Economics and language-as-interface |
| 4 | 07 | Context engineering, full session, the skills hinge of the course |
| 5 | 08 + 09 | Evaluation and agents |
| 6 | 10 | Build session: everyone ships something |
| 7 | 11 + capstone kickoff | Toolkit one-pagers; capstone projects chosen |
| 8 | 12 + showcase | Origination discussion; capstone presentations |
Session template (90 min): artifact showcase from last unit (10), frame today (5), students run the unit's Core Concept and Live Demo on their own devices while you circulate (35), group discussion (20), challenge launch and work time (15), close with one-line takeaways (5).
3. Before week one
- Every student has a working account on at least one assistant (Claude, ChatGPT, or Gemini). Most services require age 13+; check your region's rules and your organization's policy for under-18s.
- School-managed Google accounts often block Gemini and related tools by default; test with a real student account beforehand and have one fallback assistant everyone can access.
- Devices: browsers on anything work for most units; Unit 02 needs webcams, Unit 10 strongly prefers laptops.
- Ground rules poster, three lines: "Never paste private information (yours or anyone's) into an AI in this room. AI output gets verified before it gets repeated. We disclose AI help on anything that leaves this room."
- Pre-run every live demo the same day (see section 4).
- If this is a study cohort, complete Appendix A before any student arrives.
4. Running sessions with nondeterministic AI
This course's demos use live AI, which means outputs vary by day, model, and student. This is a feature. Handle it with three rules:
Pre-run same-day. Before each session, run the unit's demos yourself on the room's default tool. You are checking for changed interfaces, rate limits, and blocked sites, not memorizing outputs.
Variance is the lesson. When two students get different answers to the same prompt, stop and name it: this is next-token prediction, no database, exactly what Unit 04 teaches. The course's biggest ideas will volunteer themselves as live surprises; take them.
When a demo "fails," it usually succeeded. The strawberry letter-count in Unit 04 is a patched example: modern models often get it right, and the unit teaches it as history. If a model aces something the course says AI struggles with, that is Unit 11's leapfrogging lesson happening in the room. Script for the moment: "The course told you this field moves under its own textbooks. You just watched it."
Free-tier caps: a room of twenty hitting one tool simultaneously will hit limits. Spread students across two or three assistants, stagger heavy demos, and treat a cap as a teachable Unit 05 moment (every query costs someone money).
5. Unit-by-unit run notes
Unit 01, Pattern Machines. Produce: the Pattern Detective log. Run: Quick Draw is the energy spike; give it ten loud minutes. Discussion catalyst: "What has an algorithm gotten eerily right about you, and how?" Sticking point: students say the AI "understands" their doodles; redirect to pattern density, and plant the flag for Unit 04. Check: can the student explain one failure from their five inputs using training-data density?
Unit 02, Nobody Programmed This. Produce: a trained-and-broken Teachable Machine model. Safety note: webcams. Offer object classification (pen versus no pen) so no one must put their face on screen; check your organization's recording policy. Run: the break-it step matters more than the train-it step. Discussion: Move 37, then "should we use systems nobody can fully explain, and where?" Check: student names the exact input that broke their model and why.
Unit 03, Where the Knowledge Comes From. The heaviest discussion unit; protect time for it. Run: assign pairs different wedding cultures and restaurant cities for the demo, then compare richness of AI answers across the room; the gradient becomes visible and personal. This lands differently for students whose cultures are underrepresented; let them be the experts, and keep the tone at "the data is the story," never "the tool is against you." Check: the Bias Audit comparison paragraph exists and cites data density, not intent.
Unit 04, It Doesn't Know Anything. Produce: the Hallucination Hunter report. Run: the fake-citations demo is the session's spine; Mata v. Avianca does the emotional work for you. Sticking point: students over-rotate to "AI is useless"; the correction is the unit's own line: draft generator, not source of truth. Check: at least one caught fabrication with a real source correcting it.
Unit 05, This Stuff Isn't Free. Produce: the AI Budget Planner. Run: pricing pages change; if numbers on screen differ from the unit's examples, that is the point (say so). Discussion: "If frontier AI costs $200 a month, who gets left behind?" Check: the student can defend one model-per-task choice with arithmetic.
Unit 06, Words Are the New Code. Produce: the Communication Ladder screenshots. Run: fast, fun, four rounds of the same prompt improving; do one round as a whole room on the projector. Discussion: the schools question in the unit's socratic goes off in real classrooms; let it. Check: the worst-to-best gap is visible and the student can name what each layer added.
Unit 07, Context Is Everything. The skills hinge; give it a full session. Produce: the five-layer Context Workout. Run: have two volunteers do the cover-letter layering live with their real situations. Sticking point: students hunt for magic phrases; keep hammering substance over syntax. Check: layer 5 output is dramatically better than layer 1, and the student names which layer mattered most.
Unit 08, Be the Judge, Not the Audience. Produce: the Fact-Check Report on a topic the student knows deeply. Run: the deliberate-error demo works brilliantly as a pair exercise (one student's AI plants the error, the partner's AI hunts it). Discussion: the 91% accuracy at scale math; do it on the whiteboard. Check: the accuracy percentage is computed, and the explanation uses next-token prediction correctly.
Unit 09, AI That Does Things. Produce: the Agent Workflow design and post-mortem. Run: the "Book the thing for next Thursday" ambiguity test is the best sixty seconds of the unit; run it on three different tools simultaneously. Discussion: reversibility as the criterion for delegation; make students rank their own five tasks. Check: the human/AI split has reasons, not vibes.
Unit 10, Build Something Real. The high-energy session; protect it from lectures entirely. Produce: a live, shareable link. Run: laptops matter today; Claude Artifacts first (no new account needed), then Replit for those who can register (school emails sometimes cannot; pair those students). Expect the room to desync wildly; the session template collapses into open studio time with you circulating. The iteration requirement (three rounds minimum) is the pedagogy; students who accept the first output get sent back with "describe what's wrong to the AI, not to me." Check: the link opens on your device, and the student narrates one thing the AI got wrong and how they steered it.
Unit 11, Your AI Toolkit. Produce: the one-page stack. Run: light session, good for catching up stragglers; do the 15-minute tool test live on whatever tool is trending that week. Capstone kickoff happens here: every student leaves with a project chosen and a first prompt written. Check: another student can read the one-pager and understand the workflow.
Unit 12, What's Yours. Produce: the capstone, presented. Run: the origination demo (your sentences versus AI's versus the collaboration) is the course's emotional payoff; do it as a silent-writing exercise before anyone opens a tool. Presentations: three minutes each, and the mandatory question for every presenter is the unit's own: "What's yours in this?" Discussion: keep the power-concentration section grounded in their experience ("what did the AI not know about your world?" links back to Unit 03). Check: the one-page brief answers what AI did, what the student did, and one thing learned.
6. Tricky moments
"The AI said something different from the course." Correct response: "Good, now you have two claims. Which one can you verify?" Then do it. The course wins these on facts and openly expects to lose them on product details; the freshness stamps exist for this.
A student gets a disturbing, biased, or wildly wrong output. Screenshot it (it is now teaching material), name which unit explains it (usually 03 or 04), and report it through the tool's feedback button as a live lesson in how these systems improve.
"Isn't this course teaching cheating?" The program's position, stated in one breath: it teaches disclosure and the defend-it test; work you cannot defend without the AI in the room is work you did not do, and every artifact here gets defended aloud. Invite the asker to any showcase.
A student's reflection reads AI-written. Do not accuse. Ask them to walk you through it verbally. The defend-it test resolves this without a confrontation, in either direction.
A student is anxious about AI and jobs or the future. Legitimate, and the course addresses it head-on in Units 09 and 12; do not dismiss with optimism. The honest frame: this program exists to move people from the audience to the operators' side of the change.
Webcam or account refusal. Always fine. Every unit has a path that works watching a partner's screen plus doing the written artifact; participation is never gated on an account or a camera.
7. Appendix A: study cohorts (implementation fidelity)
Skip this entirely unless your cohort is part of a registered study. If it is, this appendix is the protocol; deviations belong in your session log.
- Consent and enrollment: students join via the study's invite code so the platform tags cohort and arm; consent screens (and guardian consent where required by the ethics approval) complete before any unit is opened. You never handle consent paperwork ad hoc; the platform records it.
- Instruments are silent zones: the baseline and endline questionnaires happen on the platform, individually, without discussion, hints, teaching, or "just think about Unit 4." Your only sanctioned sentence: "It's short, it's not graded, and skipping is allowed." Never read items aloud to the room, never debrief items afterward, before endline completion.
- Do not convert the course into assessment: quizzes stay ungraded, no leaderboards, no score talk. Grading the quizzes contaminates the learning measures.
- Consistency across facilitators: run the session template as written, keep the week map, and log per session: date, units covered, attendance count, any deviation and why, any tool outage. Five lines in a running document is enough, and it becomes the fidelity section of the paper.
- Equal-treatment rule for arms: if the study has two arms, you deliver identical enthusiasm, time, and help to both; the difference between arms lives in the materials, never in you.
- Incidents: any student distress, privacy slip, or consent question pauses the activity for that student and goes to the study lead the same day.
Methodological aside for the study lead: the AILIT-S construct maps most directly onto 101's content; 102-only cohorts may show ceiling effects at baseline. Plan measures accordingly.