Learning Journeys Early Years Planned Junior Open now Maker Open now Explorer Planned Creator Planned Researcher Planned
Get Started
  1. Early Years, journey 1
  2. Junior, journey 2
  3. Maker, journey 3
  4. Explorer, journey 4
  5. Creator
  6. Researcher, journey 6
Journey 5 of 6 · Creator

How a computer reads, and how it writes.

Creator is planned. It will focus on language AI: turning words into numbers, sorting sentences by meaning, writing one word at a time, and building chatbots. Three language brains already run inside Maker.

PlannedNo launch date yetThree language brains inside Maker
Maker's Word Brain window with Other ways a computer reads a sentence open: tabs for bag of words, weighing words, word embeddings, reading in order and attention, and an animation of a recurrent network reading the robot is not happy one word at a time with a memory.
Inside Maker today Real screen, Maker. Five ways a computer reads a sentence, animated.
Who it is for
Learners who read and type sentences comfortably and want to know how chatbots work.
What learners build
In Maker today: a chatbot that answers out loud, a story writer, and a map of word meaning.
What you need first
Typing sentences. Junior or Maker helps, but is not required.
Available now?
No. Creator is planned. Its three language brains run inside Maker now.
Who it is for

For learners curious about words and chatbots

Pictures and sounds come first in Junior. Creator is where the data is language, the kind of AI most people now meet every day.

What makes this stage different

Language looks like meaning, but a computer only sees numbers. Creator shows exactly how sentences become numbers, how a model picks an answer or a next word, and why a fluent answer can still be wrong.

Planned The aims below are for the Creator journey. The next section shows the language brains already inside Maker.

  1. Turns words into numbers

    Explains two or more ways a computer represents a sentence, and what each one loses.

  2. Builds a language model

    Teaches a sentence sorter and a next-word writer from their own text.

  3. Tests on new sentences

    Checks a model with sentences it never saw, and finds which words decided.

  4. Separates fluent from true

    Explains why a model that writes well can still be wrong.

See what exists today

The language brains inside Maker

Real screens from Maker, which is open now.

Word Brain X-Ray in 3D: the sentence hello robot i am so happy drops into word bins, the neurons light up, and three pillars show greeting, goodbye and happy.
Real screen. Words drop into bins, neurons fire, each idea gets a pillar.
Inside Maker today

Sorts sentences into ideas

A bag-of-words classifier the learner teaches with their own sentences. Its X-Ray shows each word’s push on the answer.

How it works

Input, processing, output, evaluation

Every language model in Creator will be explained in these four parts. The Maker brains already follow them.

  1. Input

    Sentences the learner types, or writing they paste in.

  2. Processing

    Words become numbers: counted in a bag, counted by which word follows, or placed by the company they keep.

  3. Output

    An idea and how sure it is, a next word, or a word family.

  4. Evaluation

    New sentences it never saw, and which words decided each answer.

Signature project

In Maker today: My Own Chatbot

Type Examples in Maker: two ideas, greeting with four sentences and joke with two of three, ready-made example sets, tap-to-add suggestions, and a tip that ten different sentences teach more than one sentence typed ten times.
Real screen, Maker. Teaching a chatbot what people say, one idea at a time.
Inside Maker today

One of Maker’s starter projects. Learners teach ways of saying things, and the robot answers out loud.

What the learner builds
A chatbot with two or more ideas, such as greeting and joke, each taught with at least three sentences.
What the AI contributes
A Word Brain that picks the idea a new sentence fits best, and says how sure it is.
What they must understand
It matches words, not meaning. “Not happy” and “happy” share a word, so it can confuse them.
How it is tested
By typing sentences it never saw, and reading which words decided the answer.
What learners understand

Ideas Creator is designed to teach

  1. Words become numbers

    Every language model starts by turning text into numbers. How it does that decides what it can notice.

  2. Order can be lost

    A bag of words ignores order, so “not happy” and “happy” look alike to it.

  3. Writing is choosing likely words

    A next-word model picks what usually came next, with no idea what the words mean.

  1. Meaning from company

    Words used in similar places end up close together on a map of meaning.

  2. Fluent is not the same as true

    Today’s chatbots are far bigger, but they still choose likely words, not checked facts.

  3. Varied examples matter

    Ten different sentences teach more than one sentence typed ten times.

Test and improve

Test with sentences it never saw

The same rule as every other journey: a model is judged on new examples.

  1. Write new sentences

    Type ways of saying each idea that were not in the examples, and see what it picks.

  2. Read which words decided

    The Word Brain shows each word’s push. A surprising push points to a missing example.

  3. Add varied examples

    Different words for the same idea, the way real people talk. Then test again.

  4. Compare methods

    “Other ways” shows how weighing words, word embeddings, reading in order and attention would change the answer.

In the real world

Language AI learners already use

Next-word suggestions

Phone keyboards suggest the next word by what usually follows, the idea behind the Story Brain.

Sorting messages

Email and support systems sort messages into kinds, as the Word Brain sorts sentences into ideas.

Search that understands

Search finds pages with similar words, not only the same ones: the idea behind the Word Galaxy.

Curriculum and progression

Where Creator sits on the path

Creator follows Explorer. A learner who knows what training and testing are will get the most from it.

  1. Before

    Explorer Planned

    Inside the network: layers, settings and evaluation.

  2. This journey

    Creator Planned

    undefined

  3. Next

    Researcher Planned

    Evidence, limits and responsible AI.

For educators and parents

Available today, and planned

What a teacher can use now inside Maker, and what Creator will add.

Inside Maker today

Word Brain
Sorts sentences into ideas; example sets such as hello or a joke, happy or sad, and robot commands.
Story Brain
Writes one word at a time from the learner’s writing, and shows every choice.
Word Galaxy
A map of word meaning, word families, and the “king minus man plus woman” idea.
Speaking and listening
Say makes the laptop speak; a Voice block listens for a spoken keyword.

Planned

Lessons built around language
A path through the four parts: input, processing, output, evaluation.
Chatbots with a character
Giving a bot a personality and example conversations.
Simple agents
Chaining steps so a program can plan and act. Not built yet.
Availability and access

Planned, with no launch date yet

Leave your email and we will write once, when Creator opens. The language brains are in Maker now.

Planned

Get told when Creator opens

You are on the list. We will email you when Creator opens.

One email when it opens. We keep only your address and which journey you asked about.

Start inside Maker now

Open the My Own Chatbot, Story Machine or Word Galaxy starter in Maker. Maker is open and free during early access.

See Maker

Questions

Questions about Creator

When will Creator open?
There is no launch date yet. Leave your email on this page and we will write once, when it opens.
Is this like ChatGPT?
It teaches the same basic idea at a small size: a model that turns words into numbers and picks likely answers. The models in Maker are tiny, trained on the learner’s own sentences, and fully visible.
Can learners build a chatbot now?
Yes, inside Maker: the My Own Chatbot starter teaches ideas with typed sentences, and the robot answers out loud.
Will Creator include AI agents?
Simple agents are planned for Creator, but are not built. We will not describe them as available until they are.
Where does the typed text go?
The sentences and writing a learner types stay in the browser on their device.

Words are data too.

Creator is planned. The language brains are inside Maker now.