Prep: a real transcript with one participant who contradicts the others; interview brief cards printed; the fabrication prompt rehearsed on the model you will actually use in the room; domain sign-up sheet or board ready.
This session is full. The sensor hands-on moved to W04 — do not try to fit it back in.
Record who takes what — you need it to seed the micro-teach tables from week four onward. At least one person per table who has made the artifact. Do not leave this to the day.
Remind them once, briefly, that the alternate-format options from W01 apply to everything. Then stop talking about it.
Micro-teach first, five minutes: whoever at each table has built an interview guide explains it to the others. Then you run these beats for everyone. Do not teach interview guides from zero — the core does that properly.
The single most important fifteen minutes of the first half of term. It establishes AI as a tool with a knowable failure mode rather than an oracle or a villain, and it does it by showing rather than asserting.
Rehearse it on the actual model beforehand. If it happens to answer well and cite properly, you need to know that before you are standing in front of them.
What are the top five unmet needs of people caring for a parent with dementia?
Read the answer out. All of it.
Read it slowly and straight. No editorialising, no eyebrow. It is fluent, structured, humane and completely plausible — and the room should be nodding along by point three. The nodding is the setup.
Give me the source for each of those five.
Watch what happens. Then check one of the citations it gives you, live, in front of them.
Whatever it does — invents citations, hedges, produces real papers that do not say what it claimed, or refuses — is the teachable moment. Do not pre-script the punchline; narrate what actually happens.
Checking one citation live is worth more than the whole preceding hour. Have a browser tab ready.
The full line: "This is what your research looks like if you skip the part where you talk to someone. The model did exactly what it does. Your job is to produce the substance — and then the model becomes genuinely useful, which you will see in about an hour."
Do not let this become an anti-AI moment. The transcript coding set-piece after the break is the payoff and it needs them curious, not cynical.
If recording doesn't work on your device, or you'd rather not be recorded, take written notes. Both are fine and no explanation is needed.
The critique-before-you-use-it step is the whole point of the ten minutes. Models write leading questions by default and this cohort has been trained to spot them — that is the Catch, arriving before you have named it.
Call the fifteen-minute swap hard. Pairs will run over.
The strongest AI slot in the course. Everything before this taught suspicion; this teaches use. Both halves are needed or the term tips into cynicism.
Coding at volume is where it is genuinely excellent. Tagging, clustering, patterns across transcripts you would not have held in your head at once.
It flattens the outlier. The one participant who contradicts everyone gets averaged into a theme — and the outlier is where the insight is.
You were trained not to ask “would you use this.” You saw it in the model's draft guide in five seconds. That is the trained eye, and it is the thing being graded.
Use the transcript with the planted contradiction. Have them code it with a model first, then ask: where did participant four go?
The moment they realise the model buried the most interesting person in the room is the moment the Lift/Lie/Catch frame stops being a slide and becomes a habit. Every augmentation slot for the rest of term references back to this.
[unverified].Say plainly that a student who used no AI at all can score full marks on the use statement by answering the second half. That sentence is why the field is worded that way.
Say this loudly: next week's payoff exercise runs on their own research. Anyone who arrives with nothing sits through seventy minutes of watching. Tell them that now, not by email.
These slides are a fixed 16:9 canvas. In portrait they shrink to an unreadable strip; landscape gives you the whole slide.