/ the spiral curriculum

Ten ideas, learned four times over

The IJAIO syllabus teaches generative AI the way children actually learn: in a spiral. The same ten fundamentals run through every band, from Grade 1 to Grade 12. What changes is never the ideas, only the depth at which a learner holds them.

Age-appropriate by design

Every idea arrives in a shape the learner can hold. At six, next-token prediction is a "what comes next?" guessing game; at sixteen it is a probability distribution over a vocabulary, shaped by sampling and temperature. No band demands maturity that hasn't arrived, and none talks down to maturity that has.

Consolidation, not repetition

Each return to an idea reactivates what was learned before, then stretches it. Guessing games become n-gram models; n-gram models become transformers. Nothing is learned once and forgotten: every year reinforces the last, and by Upper Secondary a learner isn't meeting these ideas for the first time, but mastering them for the fourth.

The ten fundamentals Text as numbers Next-token prediction Embeddings Neural networks Generation Prompting Hallucination & truth AI agents Data & learning Ethics & safety

Read down the page to follow one learner's pathway: the same ten cards, four times, each band in its own voice. Every card below is anchored in that band's Pillar learning outcomes.

Lower Primary · Grades 1–3
/ generative-ai first principles

How big ideas look at this age

Text as numbers
Computers see numbers, not letters; encode a word with a code sheet.
Next-token prediction
“What comes next?” pattern games.
Meaning as vectors
Group things by shared features.
Neural networks
Follow simple rules by hand.
Generation vs retrieval
AI can make NEW things vs find existing ones.
Prompting
Ask clearly and kindly.
Hallucination & truth
AI can be wrong, check with an adult.
AI agents
Sense → Think → Act helper robot.
Data & learning
AI learns from many examples we give it.
Ethics & safety
Privacy, honesty, kindness.
Lower Primary syllabus →
Upper Primary · Grades 4–6
/ generative-ai first principles

How big ideas grow at this age

Text as numbers
Hand-tokenise sentences; one word is not always one token.
Next-token prediction
“Guess the next word”; AI picks a likely next word.
Meaning as vectors
Feature grids show how similar words sit near each other.
Neural networks
A neuron: inputs → weighted sum → fires.
Generation vs retrieval
Generative vs search; image generators.
Prompting
Prompt = role + instruction + example; then iterate.
Hallucination & truth
AI hallucinates; cross-check before trusting.
AI agents
An agent pursues a goal through the loop.
Data & learning
Data in → pattern out; learning from data, not fixed rules.
Ethics & safety
Bias from data; fairness; citing AI help.
Upper Primary syllabus →
Lower Secondary · Grades 7–8
// generative-ai first principles

Each fundamental, made concrete

Tokenisation
Use a tokeniser; explain letter-miscounts; tokens → IDs.
Next-token prediction
Bigram/trigram Markov be-the-LLM by hand.
Embeddings
Inspect an embedding space; analogy arrows.
Neural networks
Human neural-network role-play; perceptron by hand.
Diffusion
Diffusion as denoising static into a picture.
Prompting
Prompt frameworks with few-shot examples.
Hallucination & truth
Why it sounds confident; verify sources.
AI agents
Sense → Think → Act with tool use.
Data & learning
Split data to train and test; quality and balance matter.
Ethics & safety
Dataset bias; deepfakes; human oversight.
Lower Secondary syllabus →
Upper Secondary · Grades 9–12
// generative-ai first principles

Each fundamental, at full depth

Tokenisation
Sub-word tokenisation; token IDs feed the embedding layer.
Next-token prediction
Probability over a vocabulary; sampling and temperature.
Embeddings
Vector space, cosine similarity, and bias baked into embeddings.
Neural networks
Forward pass, activation, loss, gradient descent, backprop intuition.
Diffusion & GANs
Forward/reverse diffusion, GANs, and prompt conditioning.
Prompting
Constructed few-shot prompting and system-prompt design.
Hallucination & truth
Hallucination as a next-token consequence; RAG and RLHF limits.
AI agents
Tool-calling loops, planning, observe-and-re-plan.
Data & learning
Overfitting, precision/recall, and curating a dataset.
Ethics & safety
Fairness metrics, governance, accountability, and transparency.
Upper Secondary syllabus →

Same ten ideas. Four depths.

← Back to the IJAIO overview