💭 in your head · nothing to type
You hold a notebook from the previous unit. Ask the question people usually postpone: how is it different from a regular AI chat at all?
Picture two interlocutors. The first has read everything on the internet and answers from memory — smooth, confident, no pauses. The second sits over a stack of papers you yourself brought and leafs through the stack before every answer. While answering, they point a finger at a paragraph.
Which of the two would you take along to check a decision that matters to you?
💭 in your head · nothing to type
You probably picked the second one — and not because he is smarter. He is checkable.
The first interlocutor may be right more often. But when he is wrong, you will not find out: his answer has no seam to pull on. The second one's mistake shows immediately — the finger pointed at a paragraph, and the paragraph says something else.
Hold that feeling: trust here is built not on the quality of answers but on the ability to check them.
A regular chat answers from "general knowledge" — whatever the model absorbed in training. The notebook works differently: before answering, it finds the relevant pieces in your sources and builds the answer on them. That is source-grounding.
Three consequences worth knowing:
1. A citation is not a decoration — it is a seam. The footnote number in a notebook answer leads to an exact spot in the source. A seam can be opened: click and see what is actually written there. In module 0 you checked the "back of the can" for materials — this is the same gesture, at the level of every single claim.
2. The notebook knows only what you gave it. Ask about something absent from the sources — the honest reply is "the sources do not cover this." Not a weakness, a feature: the boundaries of knowledge are visible; a from-memory chat has no visible boundaries.
3. For the notebook, a video is its transcript. It does not watch the clip — it reads the subtitles. Intonation, diagrams on a whiteboard, code on the screen — anything that never made it into text does not exist for it. A talk where the point is shown rather than said will shrink to its spoken lines.
If you took the context-and-memory module, you will recognize the mechanics: sources are cut into chunks, and the chunks relevant to your question are retrieved. They are placed into the model's context with the instruction "answer only from this." A notebook is managed context with mandatory citations — not a different kind of magic.