TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · The Brief · AGE 17—19 · POSITIONAL · METHOD
Briefing for reading, not making
An analysis brief is judged by whether every answer can be traced into the material.
When to use
Whenever you hand a model material to read rather than a thing to make: a client email to summarise, two bye-law clauses to compare, a survey to pull numbers from, a transcript to extract decisions from.
The method
Five parts. The material: paste or attach it, and say what it is. The question: one job, stated plainly — extract, compare, summarise or classify; not all four. The form of the answer: a table, a list with page references, a paragraph of fixed length. The rule for gaps: "if it is not in the material, say so; do not fill from general knowledge." The check: ask for the exact words or location behind every claim, and open them. The gap rule asks the model not to fill; the check you make against the material is what catches it when it does. An analysis brief is judged by whether its answers can be traced back into the material. A making brief is judged by whether the result suits. That is the whole difference.
Watch for this
Letting the model mix what the document says with what it knows. Without the gap rule, a summary of your client's brief will quietly contain sentences the client never wrote — plausible ones, which are the hardest to catch.
The Lab's note
The five-part analysis brief is the Lab's scaffold, built because students arrived knowing only how to brief for images. It is not the only workable shape. It is the one where, when an answer fails, we can point to the part that was missing.
Try it
Take one real document — a competition brief, a municipal notice. Brief a model to extract every date, number and named obligation into a table, with the page or line each came from. Check five entries against the source. Count the ones that are wrong or invented.
Prove it
Write one analysis brief with all five parts labelled, run it on a document you know well, and mark which answers you could trace and which you could not.
How it works
A model reading a document is still predicting the next tokens. But now the document sits in its context window, and the likeliest continuation of "what does clause 4 say" is usually something close to clause 4. The brief's job is to keep it there: to make quoting the text more likely than drifting from it. This card is the way in to the whole beyond-image part of the map — codes, quantities, climate numbers, minutes — where the model reads more than it draws.
What this idea builds on
- A starting idea.
What this idea opens up
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.