Worked example · Make something from it
From scattered notes to a first draft in your own words.
An article, a video, a talk: something you're making from what you learn. Give it a project of its own, started as the question it answers, and the notes for it gather on its page as you write them down — moved in by you, or filed by Graphite once you turn filing on. Find what's missing and answer it in your own words, and when you can explain it, have the Guide arrange a first draft from your own sentences, with the gaps marked. Then write the rest, and publish it or set it out as a script for a video. Graphite never changes your words unless you accept a change it suggests, and never adds a sentence of its own to your draft.
- Write it down
- Work it out
- Make something of it
Step 1 of 3
Write it down.
Give the article a project of its own, started as the question it answers, and drop everything into the Inbox as it comes — a line, a link, a paragraph you pasted. Move each note to the project when you know it belongs, or, with File my Inbox on, Graphite files it there.
A project of its own.
A project is something you're making from what you learn — an article, a video, a talk. It starts as a question and ends when you've made it. Make one from the + beside Folders in the sidebar, in a folder if you like — the welcome notebook keeps Why does attention matter in Machine learning — and write on its page what it's for: who it's for, and what done looks like.
A project's page is a note like any other, so its words are yours, and it gathers the rest as you go: the notes in it, what's missing from them, and a line of what Graphite notices — how many notes, how many questions are open, when you last taught it back, when a draft was started. Never a score, and never a status you fill in.
Why does attention matter
ProjectMachine learning4 notes · 2 questions open · taught back Oct 25 · draft started Oct 25
An article for people who use chatbots every day but have never read the paper: what attention does, in plain words, and why it makes long inputs expensive. It's done when it's published.
What's missing
- How does the model know which word to look at hardest?
- Does the model understand the sentence, or does it only match words?
- What did models do before attention?
Notes in this project
A thought, a question, a linkAdd- Attention reads the whole sentence at onceToday
- Draft — Why does attention matterToday
- Teaching — Why does attention matterToday
- Why long inputs cost so muchToday
Everything into the Inbox, as it comes.
Press ⇧⌘N, or ⌥⇧⌘N in a browser, and a note opens at the top of the Inbox, to name and write in, with no key and nothing sent; New note on the Graphite button's card makes one too, in the project whose page is in front, or else in the Inbox. Paste a paragraph, or a link on its own, anywhere outside a note's text, and it's a note the same way. Nothing asks what it is.
Every note in no project or folder is in the Inbox, one ruled line each. Arrange it as you go — drag a row, or press ⌘↑ and ⌘↓ — open a note in place, as a card in the list, to read, rename and edit it without leaving, and drag one onto a folder or a project in the sidebar to file it there.
Inbox
6 notes · drag a note onto a folder to file it
New note⌥⇧⌘N- Ask Sam to read the draftToday
- Similar meanings sit close together#transformers #machine-learningOct 7
- Attention is a weighted average#transformers #machine-learningOct 6
- How does a model know when to stop writingOct 5
- Welcome to GraphiteOct 4
Everything that changed is in theLogbook
Moved in, or filed for you.
When you know where a note goes, move it: drag it onto the project in the sidebar, or choose Move to… on its row or its card (⇧⌘M, or ⌥⇧⌘M in a browser), type a few letters of the project and press Enter. Select several with ⌘-click or ⇧-click, and Move to… or a drag moves them together. A note leaves the Inbox at once, and each move is a line in the Logbook, with Undo.
Or turn on File my Inbox in Settings → AI, and Graphite files the Inbox for you: it sends each note's title and first lines, with the names of your projects and folders, to your AI provider, and puts each in the project it clearly belongs to, or else the folder. A note none fits waits, and one about something new gets a chip on its row suggesting a project, which waits for your click. Each act is a line in the Logbook, with Undo.
Logbook
What you and Graphite changed, newest first. Undo puts it back as it was. Kept for 90 days.
Today
Filed “Everyone gets the tired cat at once” in “Why does attention matter”. It’s about how the article opens.
6:14 PMUndoRead a receipt: Corner Café, 14.50 USD on 3 Oct, filed in Receipts/2026.
6:13 PMUndoMoved “Open the article with the tired cat” to “Why does attention matter”.
5:52 PMUndoFriday
Filed “Tokens are pieces of words, not words” in “Machine learning”. It’s about machine learning, and none of your projects is.
9:05 AMUndo
Step 2 of 3
Work it out.
On the project's page, find what's missing: the questions a reader would ask that your notes don't answer yet. Answer them in notes of your own, and teach it back with your notes closed to find what you can't explain yet.
Find what's missing.
On the project's page, choose What's missing in Actions, or in the Guide. It sends the project and its notes — up to twelve, in their order there — and the Guide says how many under its box. The Guide lists what stands between them and something you could publish: the questions a curious reader would ask that your notes don't answer yet, claims without support, and what's unclear — at most seven, each one sitting's work.
Add to note puts them under What's missing, labelled Asked by Graphite, with Undo. Answer each in a note of your own, in your own words, and tick it off where it's written. The project's page also lists the questions its notes ask, and a ticked one folds under Done when you leave.
Ask on the project's page.
Ask the Guide a question beside the project's page — where a reader would get lost, what to cut — and it answers from the project and its notes, citing them, then asks you one question back. On any one note, it sends that note and the notes it links to; select several notes in the Inbox or All notes and choose Ask Graphite, and what you ask goes to the Guide, which answers about them together. Asking writes nothing into a note: keep an answer with Add to note, or make it a note of its own.
GuideThis noteWhat's missingExplain simplyResearchTeach it backQuiz meCheck itLink relatedWhere would a reader get lost?
At the scores. Attention reads the whole sentence at once says “it” looks hardest at “cat”, but nothing in your notes says how a word decides where to look — and that's where you stalled in Teaching — Why does attention matter.
Why long inputs cost so much is clear on what long inputs cost, so the second half holds up.
What could you show a reader to make the scores concrete?
Add to noteNew note from thisAsk about this note…Sends this project and its 4 notes when you ask.
Teach it back before you draft.
Teach it back with your notes closed, to a curious student who asks why, and what you can't explain yet goes under What's missing too. Nothing makes you do it first; a draft from notes you can explain is a better draft.
Across 28 studies, preparing to teach, and then teaching, helped people learn — more when the teaching was interactive. Kobayashi, 2019
Teach·AttentionNote closedStudent
Why does every word have to look at every other one? Couldn’t it just look at the words next to it?
You
Because what “it” means can depend on a word much earlier. Attention lets every word look back, so the model remembers what came before. That’s also why long inputs cost so much more.
Afterwards, against your notes
Unsupported
Your note says the weights are computed fresh for each input. Attention isn’t memory.
Attention
Question
How much more, and why that much?
How much more do long inputs cost, and why that much?Find what it should link to.
Link related, in the Guide beside any note, suggests the notes it could link to, each with its reason, and adds none until you accept. A project holds what you're making; links hold what you know. A note is in one project at a time, and links to as many notes as it likes.
Link related·Attention is a weighted averageLink to your notes
Every abstraction hides a cost
Its example of attention hiding a quadratic cost behind one clean operation is this note's last point.
Linked under Connects toEmbeddings
The queries, keys and values a token's meaning is averaged from are projections of them.
AcceptDismiss
Connects to
Every abstraction hides a cost — Its example of attention hiding a quadratic cost behind one clean operation is this note's last point.
Step 3 of 3
Make something of it.
Ask for a draft, and the Guide arranges your own sentences in the order that explains it, marking each gap where it belongs; anything it can't find in your notes, it leaves out. Write the rest yourself, then publish it as a page or set it out as a script for a video.
A first draft, from your own sentences.
Choose A draft in Actions. The Guide arranges your own sentences, word for word, into paragraphs under short headings, in the order that explains it best, and marks each gap where it belongs with Missing and the question. It never adds a sentence of its own or rewords one of yours: one it can't find in your notes is taken out before you see it, and under the draft the app says how many it left out.
Students given plain chatbot answers did better in practice and worse on the exam. A tutor with guardrails largely avoided the harm. Bastani et al., 2025
GuideThis noteWhat's missingExplain simplyResearchTeach it backQuiz meCheck itLink relatedDraft it
Reading the whole sentence
Before transformers, models read a sentence one word at a time and had to carry everything forward. Attention lets every word look at every other word at once. In "The cat sat on the mat because it was tired", the word "it" looks hardest at "cat". That's how the model works out what "it" means.
Missing: How does the model know which word to look at hardest?
Graphite left out one sentence that wasn't in your notes.
Save as a draftAsk about this note…Sends this project and its 4 notes when you ask.
Kept as a note of its own.
Save as a draft makes it a new note in the project, opening with a line that says Graphite arranged it from your notes — every sentence yours, what's missing marked — and the Logbook keeps a line for it, with Undo. Then write the rest yourself: the gaps are yours to fill. An outline works the same way, as a list, kept in the note or as a note of its own.
Draft — Why does attention matter
Arranged by Graphite from your notes: every sentence is yours; the headings and what's missing are Graphite's.
Reading the whole sentence
Before transformers, models read a sentence one word at a time and had to carry everything forward. Attention lets every word look at every other word at once. In "The cat sat on the mat because it was tired", the word "it" looks hardest at "cat". That's how the model works out what "it" means.
Missing: How does the model know which word to look at hardest?
What it costs
Every word is compared with every other word, so doubling the input makes four times the comparisons. That's why long documents are slow and expensive to run.
In the Logbook
Graphite arranged “Draft — Why does attention matter” from your notes.
6:12 PMUndoEdits you accept or reject.
Click the Graphite button, or press ⌘J: Format tidies the draft's headings, lists and spacing without changing a word, and Summarize writes a few plain sentences for its top that say what the draft says and nothing it doesn't, as Graphite's own block, titled Summary. Select some words and right-click: Fix the spelling, Shorten it and Summarize work on just those, or choose Ask Graphite and say what you want of them — say them plainer. Each is a suggestion: it shows in the note, change by change — words that would go struck through, words that would come, or be formatted anew, marked — and you accept or reject each change, or all of them at once.
Nothing is written until you've decided every one. What you accept is one line in the Logbook, with Undo, and the Graphite button's card lists it under its buttons, newest first, with when; reject them all and the note is as it was.
Draft — Why does attention matter
Arranged by Graphite from your notes: every sentence is yours; the headings and what's missing are Graphite's.
Reading the whole sentence
Formatting, change 1 of 2Before transformers, models read a sentence one word at a time and had to carry everything forward. Attention lets every word look at every other word at once. In "The cat sat on the mat because it was tired", the word "it" looks hardest at "cat". That's how the model works out what "it" means.
Accept RejectMissing: How does the model know which word to look at hardest?
What it costs
Formatting, change 2 of 2Every word is compared with every other word, so doubling the input makes four times the comparisons. That's why long documents are slow and expensive to run.
Accept RejectPublish it, or script it for a video.
Publish a page makes the finished note one web page, as an HTML file you own; Graphite doesn't host it. A script for a video sets it out as scenes — what to say and what to show — for you to record. Copy as Markdown is there for a blog or a newsletter.
Publish “Attention is a weighted average” as a web page
It’s a file you own. Put it on any website — Graphite doesn’t host it.
Check it firstCopy as MarkdownPublishAttention is a weighted average.htmlAttention is a weighted average
transformers · machine-learning
Every token's new meaning is a weighted average of all the tokens' value vectors, and the weights come from how well its query matches each key. The weights aren't learned: the matrices that make the queries, keys and values are.
Grew from Attention
Not yet
What it doesn’t do yet.
- It doesn't write the piece for you: the draft is your own sentences, arranged, with the gaps marked.
- No sync between devices: browser notes stay in that browser on that device.
- It doesn't share a project with anyone: a notebook is yours alone.
Start with the notes you already have. Read how to organize notes without maintaining a folder system for a small workflow you can try today.