You pick the question. You run the test.
The claim and the evidence live on the same page.
Two posters. Both make a claim. Only one lets you verify it.
One shows the actual output. The other describes the output. A description is not evidence — it is a claim about evidence.
Last lesson you gave the same prompt to different models and logged every output. That is your evidence. Today you make it checkable.
Add one more run — same prompt, one more model. That is the only new generating today.
Write your question in one sentence. Not "is AI biased" — something your two outputs can actually answer.
What you tested, in one sentence.
What changed, what was held, how many runs.
The actual outputs. Screenshots or verbatim — side by side.
The full four-part stem. Fourth clause included.
The fourth clause of your claim, written out — what this does not establish.
What you decided vs. what the model produced.
Zones 3 and 4 go next to each other. A reader must be able to check the claim against the evidence without turning anything over. If it is not on the page, it was not submitted.
Blank layout and two worked examples: poster-templates.html
After seeing your partner's — what is the one thing you would change about yours?
Write it down. This is the part that sticks.
| The question is testable | Small, specific, answerable with what you had today |
| One variable | You changed one thing and can say what you held constant |
| Evidence is real | Actual outputs — not described, not retyped from memory |
| The claim is bounded | All four clauses, and the fourth one is specific |
| Limits are honest | Two or more, about your test — not about AI in general |
| Authorship is named | Clear what you decided, clear what the model produced |
Not on this list: whether your result was interesting. A careful test with a boring result is a complete success. A dramatic finding you cannot support is not.
Image or PDF.
Lastname-U1D5
Photo of your notes on their poster, and the one thing you would change about yours.
Lastname-U1D5
Thinking: copy your circled answers and your claim into the Classroom response box. Evidence: attach your poster, plus your partner-review notes.
Open the attachment to check it is readable, then turn in.
Reply rules a person
wrote. Felt like
understanding. Wasn't.
Counted continuations
plus one rule for
choosing.
Prediction from learned
parameters. Fluent, sometimes
right, and often confident
whether or not it should be.
Fluency is not accuracy — and the meaning you feel in a machine's output is usually the meaning you brought to it.
Everything so far has been text.
Next: systems that generate pictures.
The same questions return — in a form you can see.
First question: what does an image generator produce when you leave something important unspecified? You will find that "unspecified" is never actually unspecified.