A mental model for AI assistants and a verification habit

Week 11 — opening Module D: AI use becomes the substance of the work this week

A short conceptual reading on AI assistants and the course’s verification habit, opening Module D — Generative AI literacy (Weeks 11–12). The companion hands-on walkthrough is Lab 8 — Drafting and verifying technical prose with an AI assistant. The exact prompts, rubrics, deadlines, submission details, extension procedures, and grading mechanics for the Week 11 work live in the Assignments/LMS space.

For ten weeks, every assignment in this course has carried a three-line AI Use Note — Tool / Purpose / Verification — appended to whatever else you were submitting. The Verification line has been the load-bearing line all along. This week, the AI use itself is the substance of the work, and the Verification line is the substance of the report.

You stay in the same VS Code + R + Quarto + TinyTeX stack you have used since Week 1. There is no new editor, no new render engine, and no new portfolio convention. There is also no required AI assistant — ChatGPT, Claude, Copilot, Gemini, or any similar tool is fine, and a free-tier account is sufficient. What changes this week is that you are auditing an AI interaction rather than appending a disclosure to someone else’s substance.

Module D opens

Module D — Generative AI literacy — runs through Weeks 11 and 12.

  • Week 11 (this week). AI module I: debugging and verification. The focus is the mental model for what an AI assistant is, the verification habit the course has been building toward all term, and a small AI debugging audit — you use the assistant on a small sample, then verify what it said.
  • Week 12 (next week). AI module II: drafting and critique. The focus extends the same lab into drafting technical prose with the assistant’s help, then critiquing what it produced.

Both weeks’ assignments are not droppable under the weekly best-9-of-11 rule. From the syllabus: “The AI module assignments in Weeks 11–12 are part of the Generative AI Literacy category and are not droppable.” They are the AI module’s parallel to the LaTeX Project (Module A) and the R Project (Module B).

The Week 13 Portfolio/workflow conference picks up your Week 11 and Week 12 work along with your AI Use Notes from earlier weeks. The conference reviews your portfolio as a whole — including your evolving verification habit — before the final assembly weeks.

What an AI assistant is, in plain terms

A generative AI assistant is a pattern matcher that produces fluent text. It is:

  • not a database,
  • not a search engine,
  • not a fact checker,
  • not a reasoner in the sense of a person who has read the source and thought about it.

Plausible-sounding output is the default. Correctness is not. This is the central thing to internalize about how the assistant fits into your work.

The course’s public AI use guidelines already lays out what the assistant is good at (reformatting, surfacing packages, explaining unfamiliar code, first-draft prose, boilerplate) and what it is bad at (citations, LaTeX packages, recent software, math correctness, plot interpretation, subtle code rewrites). The note you are reading now does not duplicate that list. Read that page if you have not already, and treat it as the operational reference for Week 11.

For background on where the course’s AI position comes from, see the AI reading spine — UNESCO frameworks, peer-reviewed studies on generative AI in education, and the published evidence behind the course’s “no automated detectors” position. The reading is optional; the Week 11 work does not require citing the literature.

Verification, not outsourcing

The most common misuse of an AI assistant — in this course and outside it — is treating its output as authoritative.

A working technical professional uses an assistant the way you might use a fast intern: helpful, fluent, and wrong often enough that you have to check. You wouldn’t put an intern’s work into production without reading it. You wouldn’t list the intern’s name in a bibliography. You wouldn’t trust the intern’s citations without checking. The same posture applies to a generative AI assistant.

Practically, in this course, that posture comes down to a short list of habits:

  • Anything the assistant says that matters — a citation, a package name, a code behavior, a math step, a definition, a claim about a paper — has to be verified against the actual artifact. The rendered PDF. The R console output. The CTAN or CRAN page. The cited paper itself.
  • The Verification line of the AI Use Note is the discipline that makes this concrete. It is the third of three labeled lines (Tool / Purpose / Verification), and in Week 11 it is a short paragraph (still labeled Verification, still the third line — there is no fourth line) that summarizes what you actually checked.
  • The verb-target-outcome shape — I ran the code; the output was X. I clicked through to the DOI; the page did not resolve. I searched for the package on CTAN; no result. Each verification operation is a check you performed on evidence you can point at.

The Week 11 work

The Week 11 work is an AI debugging audit. You use an AI assistant on a small, fixed technical sample, and you write a short Quarto-to-PDF report — a few pages — on what the assistant said and what you verified.

The audit has the same shape regardless of which AI assistant you use:

  1. Ask the assistant a scoped, specific question about the sample.
  2. Record the assistant’s response, in summary, in your own words.
  3. Verify the assistant’s load-bearing claims against stable external evidence.
  4. Identify what the assistant got right, got wrong, fabricated, or omitted.
  5. Correct what needs correcting.
  6. Disclose with the three-line AI Use Note.

Lab 8 walks this workflow on a small generic illustrative sample (a Quarto rendering issue), so you have rehearsed the ask → record → verify → correct → disclose sequence on something low-stakes before the Week 11 work. The Lab 8 sample is different from your Week 11 sample. The lab walks the workflow; the Week 11 work applies it.

The exact prompt, the sample, the rubric, the deadlines, the submission details, the extension procedures, and the grading mechanics live in the course LMS.

Tool-agnosticism

The course does not require any specific AI assistant. ChatGPT, Claude, Copilot, Gemini, Cursor, Codeium, or a campus-licensed assistant are all fine. No paid tier is required. A free-tier account on ChatGPT or Claude is sufficient for the Week 11 work.

You name whichever assistant you used on the Tool line of the AI Use Note. The Verification habit is the same regardless of which assistant produced the response.

If a free-tier assistant runs out of usage mid-conversation, that is itself worth noting — summarize what you have, name the usage cap on the Tool line, and continue the verification yourself. The verification chain is what the work is graded on.

The “no detectors” position

This course does not use automated AI-detection tools to flag student work. Detector tools are unreliable in general, and there is peer-reviewed evidence that they are biased against non-native English writers (see the AI reading spine for the cited study).

What matters in this course instead — and especially in Week 11 — is:

  • a clear AI Use Note,
  • work you can explain if asked,
  • visible verification.

If you used AI productively and verified the output, you have nothing to hide. The course is built so you don’t need to.

Privacy carryforward

The AI use guidelines’ privacy section applies in Week 11 unchanged: do not paste other students’ work, non-public datasets, identifiable personal data, or LMS-only course content into an AI tool. The Week 11 sample is deliberately a small public-style artifact so the audit can proceed without privacy risk.

Common patterns to expect

When you do the Week 11 work, watch for these:

  • The assistant gives a fluent, plausible-sounding response with invented details. Most common pattern. The verification chain becomes a list of each invented detail checked against external evidence.
  • The assistant refuses to comment. A legitimate observation. Note the refusal, then verify the sample against external evidence yourself.
  • The assistant gives a long vague response. Quote one short load-bearing claim from it and audit that. Do not try to verify every sentence — the chain should focus on claims that matter.
  • The free-tier assistant runs out before the conversation finishes. Document this and continue verifying yourself.

In all of these, the audit findings are grounded in the verification chain, not in the assistant’s response volume.

Finishing well

Before you submit, render the document twice in a row and confirm the document is the same in both renders. This catches the same kinds of stability issues you saw in Weeks 9 and 10 (set.seed() placement; non-deterministic steps), now applied to a prose-heavy audit document.

A small audit-week debugging-hint list:

  • The render fails. Comment out a chunk if you have one; isolate the error; bring it to a scheduled studio meeting in the MAC.
  • The PDF is suddenly enormous. Probably a pasted long transcript. Summarize instead of pasting.
  • The verification chain reads as generic. Each entry should be verb + target + outcome. “Verified the DOI” is not an entry; “Clicked through to doi.org/…; received ‘DOI not found’” is.
  • The audit findings sound like opinions. Ground them in the verification chain. “The assistant fabricated the journal name” is a finding only if a verification entry shows the journal does not exist.
  • The AI Use Note Verification line is one line. In Week 11 it is a short paragraph that summarizes the verification chain.

Looking ahead

Module D continues next week with Week 12 — AI module II: drafting and critique. Lab 8 spans both Week 11 and Week 12: this week you use it as audit prep; next week you return to it for drafting-and-critique prep.

The Week 13 Portfolio/workflow conference picks up your Week 11 and Week 12 work as part of the portfolio review, together with your AI Use Notes from earlier weeks. Then Weeks 14–15 are portfolio assembly and final polish.

The exact Week 11 prompt and submission details live in the course LMS. Bring the rest yourself.