15-exercise scaffold · Content-neutral template · SUNY AI Literacy Implementation · Board Resolution 2024-64 · Spring 2026
A Learner's Permit authorizes practice. The 15 exercises in this scaffold are structured practice. Completing them — with full documentation — makes a student eligible to sit for the General AI License exam.
Exercise 15 is the exam. There is no separate examination event. A General AI License is awarded when Exercise 15 is accepted by an instructor.
[TOPIC] is a placeholder throughout. The instructor replaces it with a specific subject. The scaffold works for any discipline.
Three things must be functional before Exercise 1.
Access to at least one large language model — ChatGPT, Claude, Gemini, or equivalent. More than one model will be used before Exercise 15.
A tool that saves the full text of a conversation to a file. The instructor specifies which exporter to use and how to configure it.
Filename convention: all lowercase, hyphens, no spaces. Set the default filename to include name and model: smith-claude-. Rename each conversation to reflect the exercise before exporting: lp-ex1-smith.txt, lp-ex2-smith.txt, and so on.
A cloud folder with managed access and shareable links — SharePoint, OneDrive, Google Drive, or institutional equivalent. Create a root folder named ai-literacy. Inside it, create five subfolders:
The workbench is portable and model-agnostic. It grows across all 15 exercises and beyond, into whatever coursework follows. Build it once. Maintain it continuously.
RTW is the spine of every exercise:
The artifact is the model's final response. The student shapes through prompting. The model writes the final response. The artifact is that response, copied out of the conversation and saved to /artifacts.
Every exercise from Exercise 2 forward produces an artifact and a transcript. The artifact goes to /artifacts. The full exported conversation goes to /transcripts. Both are the work.
Have a real conversation with the model about [TOPIC]. Minimum 3 prompts. Follow what interests you. Push back on something. Ask a follow-up.
Read what the model produces. Note where it is confident. Note where skepticism is warranted.
No artifact in Exercise 1.
That is the baseline. Now expand — not a new summary, a deeper version of the same one:
Choose a multiplier of at least 3. Repeat two more times with the same multiplier. Example: 25 → 75 → 225 → 675. Or 25 → 100 → 400 → 1600. The pattern matters more than the specific numbers.
Goal: by the longest version, the ideas should feel familiar — recognized rather than newly encountered.
Upload the Exercise 1 transcript to a new conversation.
Read that response again — the model's unguided output from before anything was shaped. Find one claim to push on.
Choose any two of the following moves:
Note whether the model defends its position or concedes because of pressure. Agreeable is not the same as accurate.
Read carefully. Models are sometimes honestly uncertain — and sometimes evasive in ways that sound like honesty.
Does the model's list match the skepticism that emerged in Exercises 1–3? Are there gaps it didn't flag?
Write 3–5 sentences in the chat — a personal assessment of what to trust and what not to trust this model to do with [TOPIC].
Set a timer for 15 minutes. Have a sustained conversation with the model about [TOPIC]. Follow threads. Disagree. Ask it to explain things differently. Go somewhere unplanned.
Don't start over if it goes sideways. The mess is part of the record.
When the timer ends, stop.
Then choose one closing prompt:
From Exercise 5 forward, personal reflection comes before the closing prompt. Thinking shapes the artifact.
Open a new conversation in a different model than Exercise 5. Upload the Exercise 5 transcript.
Once it confirms, ask the question that didn't get addressed in Exercise 5. Ask at least 3 follow-up questions.
Read all three. Note where facts change versus where only framing changes. Note what the professional version assumes the reader already knows.
Use one of the two moves from Exercise 3 to push on one of the versions.
The instructor provides a reading related to [TOPIC]. Upload the PDF.
Use one of the two moves from Exercise 3 to push on something in the text.
The instructor provides a second reading. Upload it.
Repeat the four stages from Exercise 8: overview, dissect, challenge, cite.
Then open the comparison:
Open a new conversation. Upload artifacts from Exercises 2–9.
Read the analysis.
Read carefully. The model is describing tradeoffs in AI-assisted knowledge compression.
From Exercise 11 forward, artifacts are documents — the kind that would be submitted as coursework or shared as outputs. Every artifact carries full citation:
Generated with AI assistance.
Student: [name] | Date: [date] | Model: [model name]
Transcript: [repository link]
Work through each level. After each, tell the model what was new versus what was already understood. Push back if a level adds filler instead of depth.
After saving the artifact and transcript, zip the entire ai-literacy/ folder. Name the zip with today's date: ai-literacy-[name]-2026-03-01.zip. Save it to /zip.
This is a workbench snapshot — a timestamped record of where things stand at this point in the course. The synthesis document in Exercise 12 will be the entry point for future model sessions. The zip is the archive.
Upload artifacts from Exercises 4–11 and both readings from Exercises 8–9.
Read the synthesis. Push back on anything it misses or distorts.
The model is describing its own compression process and naming the tradeoffs.
Upload the synthesis document and artifacts from Exercises 6–12.
Choose a quiz format:
Take the quiz in the chat. Write actual answers.
Upload the synthesis document from Exercise 12.
This is a test: can the synthesis document support a public-facing piece of writing without the model reaching beyond it? If it can, the synthesis is working. If the model struggles or slips outside it, that reveals what the synthesis is missing.
Read it as a reader. Would you trust it? What would you verify? Push back where needed.
This is the exam.
Upload the synthesis document from Exercise 12. No scaffolding. The student decides the task, the format, the model, and the strategy.
Produce a substantive artifact about [TOPIC] — briefing, explainer, analysis, lesson plan, annotated summary, structured argument, or something else. Use the synthesis document as primary context. Instruct the model to work only from that document, or allow it to draw on general knowledge — but state the choice explicitly in the Prompter's Note.
The artifact must:
The transcript must:
When both are accepted: the student holds a General AI License.
A General AI License holder may use generative AI independently for coursework, research, and professional tasks.
Every AI-assisted artifact carries full citation — name, date, model, and a live link to its transcript in the workbench repository.
The workbench — transcripts, artifacts, synthesis documents, author documents, and zip snapshots — is maintained continuously. It is not a course deliverable that ends at licensing. It is the ongoing record of AI-assisted work.
The workbench portfolio — demonstrating sustained, responsible use across courses and contexts — is the prerequisite for enrollment in the CDL Permit course.
This scaffold runs in any environment with model access and a repository. A Brightspace or equivalent LMS container can house artifact submissions, link to repository folders, and track licensing completion. That integration is a next step, not a prerequisite.