TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · The Brief · AGE 17—19 · FACTUAL · DURABLE

Negotiating the model's defaults

You cannot fill every slot, so you have to choose which defaults to fight.

The dilemma

Every slot you leave blank, the model fills with its default. The default is not an average of its data. It is a learned tendency, shaped by what the data contained most, by how the model was trained and tuned afterwards, by the system's own instructions and by how it samples — and it is tilted the way all of those are tilted. You cannot fill every slot; there are always more than you can name. So you have to choose which defaults to fight and which to accept.

The choices

Accept the default where it does not touch your argument. Fight it where it does, by naming what you mean before the model decides. Or fight everything, and write a brief so long that the model attends to none of it.

The consequence

Accept the wrong default and a decision you never made appears in your work looking like yours: the glass tower the model prefers, the Western porch, the extra trees. Fight every default and you lose what a generator offers, which is speed and surprise. The skill is predicting the default before you see it, so the fight happens in the brief and not in the re-roll.

The case

"A house in Dharwad" returns a house that has never seen Dharwad: a pitched tiled roof from somewhere else, a lawn, no compound wall, no jagali. Each is a default. The student decides: the roof matters — fight it, "flat RCC roof with a parapet"; the compound wall matters — fight it; the lawn does not matter for this study — accept it, or exclude it. Three lines in the brief, and the house moves from the model's default to the student's street.

Prove it

Before you run a minimal prompt, write down the defaults you predict it will impose. Run it. Score your predictions. Then write the brief that overrides the ones that matter.

Take it to crit

Can the student name the defaults their brief quietly accepted, say which ones they chose to fight, and show that the fight happened in the brief rather than in forty re-rolls?

How it works

A generative model returns a probable completion of what it was given. Where the brief is silent, "probable" follows the model's learned tendency, and that tendency over-represents the photographed, the published and the Western, because the data did and the later training did not correct for it. There is no villain here, and there is no "average" being looked up. The default is simply what the system leans toward when nothing pushes back, and it will be chosen for you unless you choose first. The exclusions card on this map is the tool; this card is the judgment about when to use it.

What this idea builds on

What this idea opens up

Sources

Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.

Age grows from 11 at the centre to 22 at the edge, and six sectors show the learning strands. Tab into the map and the arrow keys step from idea to idea, following the links where there is one. Enter opens the idea under the cursor, and E reads out its links and the reason recorded on each. Press slash for Search, question mark for the full key list, and Escape to leave. Open Ideas for the complete readable list, including what each idea builds on and what it opens up.

LOGIKA · RBDS AI LAB INDIA
ON-RAMP · AGE 11 · FIRST ENCOUNTERS, NOT GATES — IDEAS · — DEPENDENCIES
DONE
OPENS NEXT
SOLID — STANDS ON