GenAI Field Guide

Getting started

Documents, images, and video

Today's chat tools can read PDFs, slide decks, scanned forms, photos, and video directly. That makes them excellent at "what does this say?" and dangerous at "what is the exact figure on page 214?"

≈2 minLandscape facts checked — re-verify before relying on them.
How often will you need it? Once Attach itto the chat Daily Put it in aproject A wholecorpus Retrieval(RAG); askyour datashop
Once, daily, or a whole corpus: attach, project, or retrieval.

Three ways to give it a document

Situation Do this Why
The document fits and you need it once Attach it to the chat Simplest, and the model sees all of it
You use the same references every day Put them in a project Persists across chats; the harness re-sends or retrieves them for you
The corpus is large, changes often, or many people query it Ask your data/AI shop about RAG Retrieval pulls in only the relevant chunks; citations you can check
Put key facts first or last Start Middle End Lower curve = less reliable recall
The "lost in the middle" effect (Liu et al., 2024): material at the start and end of a long document is used more reliably than material buried in it.

What it is good and bad at

Good at Verify, or hand to a tool
"Summarize this." "What does section 4 require?" "Compare these two versions." Exact numbers pulled from long documents, especially mid-document
"Find every place this deck mentions manning." Counts, sums, and dates: a language engine, not a calculator