TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 18—21 · POSITIONAL · METHOD
Set the rubric before you generate
A rubric written in advance fixes the standard before the polish can lower it.
When to use
Before any generation whose result you will have to choose between, defend, or show to someone else. If you are about to make more than one image, write the rubric first.
The method
Three short lists, written before you press generate. Must do: what the image has to achieve. "Show the court in afternoon shade," "read as load-bearing brick." Must hold: what cannot change. "The plan we drew," "a 3-metre floor-to-floor." Must avoid: what you will reject on sight. "Glass curtain walls," "anything the section cannot support." Then generate, and hold each output to the three lists in order. An image that fails must hold is out, however good it looks.
Watch for this
Rewriting the rubric after the batch arrives, to let a beautiful wrong image in. That is the model's fluency deciding for you, one step removed. If the rubric was wrong, change it before the next run, and write down why.
The Lab's note
Criteria before generation is the Lab's method, and it is the twin of the criteria layer in the reasoning brief elsewhere on this map. It is not the only way to keep judgment honest. It is the one that turns "do I like it?" into a check you can defend to someone who does not share your taste.
Try it
Write a short rubric for the image you want: three lists, a line or two each. Generate. Hold the outputs to it. Then find the case where the rubric throws out an image your eye wanted to keep, and notice how that feels. That is the method working.
Prove it
Show one case where the rubric you wrote first rejected a good-looking output, and say what the rubric saw that your eye did not.
How it works
Without criteria, you judge an image by whatever stands out most, and in generated images what stands out most is polish. The model is good at polish, and polish tells you nothing about fit. A rubric written in advance moves the judgment to properties you chose for reasons, and it fixes the standard before the polish can lower it. The order matters. Must hold comes before must do, because an image that breaks a fixed constraint is not a candidate, however well it does everything else.
What this idea builds on
What this idea opens up
- Nothing yet names this as a foundation.
Sources
- collaborative-red-pen
- iiid-saurashtra_ai-fundamentals
- Your Prompt Is Your Brief
- The Seduction of the Machine
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.