Best for
Unit 7 introductory lesson — before students use any AI tools. Sets the conceptual and ethical foundation for the rest of the unit.
Digital Technologies & AI Ethics • Unit 7 • Years 9–13 • Concepts
Introduction to Large Language Models
What is an LLM, and how does it actually work? Understanding the mechanics behind AI language tools is the first step toward using them responsibly — and evaluating them critically.
Free concepts handout, premium localisation path
Want this adapted for your school's specific AI tools or local context? Te Wānanga can localise it — adding examples from the digital tools your students actually use and connecting to community digital sovereignty initiatives.
- Add locally relevant examples of AI tools students encounter at home and school.
- Connect to your school's digital citizenship policy and acceptable use agreements.
- Integrate with your kura's approach to te reo Māori and mātauranga Māori in tech.
Kaiako planning snapshot
- Use length: 40–50 minutes. Concept sections (~15 min), paired discussion (~10 min), reflection questions (~20 min).
- Grouping: Read-aloud or silent read for concept sections; reflection questions individual. Brief whole-class discussion at the end.
- Prep: No tech required. If students have used ChatGPT or similar before, that prior experience is useful discussion material — ask them what they noticed about how it responded.
- Differentiation: Entry: complete questions 1 and 2 only. On-level: all four questions. Extension: research one real-world example of LLM bias or data underrepresentation and bring it to the next lesson.
- Neurodiversity support: The process flow (tokenization → processing → generation) is a visual sequence — draw it on the board for students who benefit from visual anchors. Allow verbal responses for reflection questions if preferred.
Resources already provided
- Conceptual explanation of LLM architecture (training data, parameters, prompting)
- Step-by-step process flow: tokenization → processing → generation
- Critical considerations — bias, environment, cultural sensitivity, privacy
- Source-bounded framing: te reo Māori as taonga and published Māori data-sovereignty principles
- Four structured reflection questions with response spaces
All content is provided. No supplementary materials needed to run this lesson.
Ngā Whāinga Akoranga / Learning Intentions
- We are learning to explain how a Large Language Model works — what it is trained on, how it generates output, and what its limitations are.
- We are learning to evaluate AI tools critically — recognising bias, data gaps, and cultural considerations.
- We are learning to use named sources when examining te reo Māori, Māori data sovereignty, and AI-system limitations.
Paearu Angitu / Success Criteria
- I can describe what an LLM is and explain the difference between training data, parameters, and prompting.
- I can give at least two reasons why LLM outputs should be evaluated critically.
- I can use a named source to explain one implication of uneven or undisclosed language coverage without claiming to define mātauranga Māori.
Why this matters in Aotearoa
The Waitangi Tribunal identifies te reo Māori as a taonga of te iwi Māori. Training data and language coverage differ between models and are often not fully disclosed, so te reo Māori capability must be checked rather than assumed. Te Mana Raraunga states that Māori have authority over Māori data and its collection, interpretation, use, and reuse. Understanding an LLM therefore includes asking what data it used, what it can verify, and who holds authority over any Māori language or knowledge involved.
He aha te LLM? / What is a Large Language Model?
A Large Language Model is a type of artificial intelligence trained on a massive amount of text data. Its core function is predicting the next most likely word in a sequence — which enables it to generate human-like text, answer questions, translate languages, and much more.
Training Data
LLMs are trained on vast datasets — internet text, books, articles. This is how they "learn" language patterns. But the data may not represent all languages, cultures, or perspectives equally. What voices are missing?
Parameters
Internal variables the model uses to make predictions. Models are measured by parameter count — from millions to hundreds of billions. More parameters does not automatically mean better or more culturally aware output.
Prompting
The input you give — your question or instruction. The clarity and specificity of your prompt significantly affects the quality and reliability of the output you receive.
Te ara haere / How it works — step by step
1
Text Input
You type a question or prompt
2
Tokenization
The model breaks your text into smaller pieces called tokens
3
Processing
The model uses its parameters to understand context and predict likely responses
4
Generation
It predicts and outputs the most statistically likely response — not necessarily the most accurate or culturally aware
Ngā āhuatanga hira / Critical considerations
Bias and fairness
Language and cultural coverage differs by model and may not be disclosed. Inspect the system’s documentation and test specific claims rather than assuming equal representation.
Environmental impact
Training and running large models requires enormous computational energy. This has a real environmental cost.
Cultural sensitivity
An LLM can generate fluent Māori-related text without holding cultural authority. Check claims against a named source and leave restricted or local knowledge out of the system.
Privacy and data
What happens to data you enter? Te Mana Raraunga states that Māori have authority over Māori data, including its collection, interpretation, use, and reuse.
Whakaaro hōhonu / Reflection questions
Answer in your own words. Use the spaces below — do not use an AI tool to answer questions about AI tools.
1. Choose one claim an LLM makes about a Māori topic. What can a named source verify, what remains uncertain, and how does undisclosed training data affect your confidence?
2. What are two potential benefits and two potential risks of using LLMs in education?
3. The Waitangi Tribunal identifies te reo Māori as a taonga of te iwi Māori. What source, permission, and human review would you require before using or publishing AI-generated text in te reo Māori?
4. What three questions would you ask before trusting information from an AI model?
Entry, on-level, and extension pathway
Entry
Complete questions 1 and 2. Draw or label the 4-step process flow. Discuss your answers with a partner before writing.
On-level
Complete all four reflection questions with specific examples where possible. Aim for 3–4 sentences per answer.
Extension
Research a real documented case of AI bias or indigenous language underrepresentation. Write a paragraph connecting it to the concepts here and bring it to the next lesson.
Aronga Mātauranga Māori
The authority boundary comes from named sources: the Waitangi Tribunal identifies te reo Māori as a taonga of te iwi Māori, and Te Mana Raraunga states that Māori have authority over Māori data. Ākonga use those statements to ask who supplied language data, who may authorise its use, what the model can verify, and what requires qualified human review. They are not asked to define mātauranga Māori from personal intuition.