TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Foundations · AGE 15—17 · FACTUAL · DURABLE
Finding versus inventing
A retrieved result has an address, and a generated answer often has none.
The idea
Retrieval brings back things that already exist, and each result has an address: you can go there and check. Generation makes a new output that may look exactly as trustworthy and have no address at all. A generated answer is not a lie; it can be correct, and it can be built from retrieved material. But the two feel the same in a text box, and one answer can be both at once. Underneath, they are different operations.
Why it matters
"Where did this come from?" has an answer for search and often none for generation. Professional work depends on that question.
See it in the studio
A search for IS-code ventilation requirements returns documents you can cite. A chatbot's answer about the same code may be a fluent blend of many codes. You cannot cite it, and you can only check it by opening the real code.
Watch for this
Tools that mix both modes in one box: retrieved facts and generated filler in a single smooth paragraph. You cannot see the joins.
Try it
Ask the same factual question to a search engine and a chatbot. For each claim in the chatbot's answer, try to find its address. Some claims will have none.
Prove it
Explain the difference between a result you can trace and an output that was generated, and why a generated answer can still be correct.
How it works
Modern systems increasingly join the two together: retrieve real documents first, then generate an answer from them. Done well, the answer carries addresses with it. Done badly, you cannot see the joins. A card higher up this map keeps apart the four places a system's knowledge can come from: its weights, the context you give it, what it retrieves, and what the product stores. The idea "bring your own facts" later on this map turns this repair into a habit.
What this idea builds on
What this idea opens up
- Nothing yet names this as a foundation.
Sources
- essential-ai-concepts
- DES.U
- ch07
- Pinecone, What is a Vector Database?
- Karpathy, Intro to LLMs
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.