Teaching sequence

Work through this resource material in the following sequence:

10 minutes

Tune In: When do we interact with AI?

25 minutes

Find Out: Generative AI

20 minutes

Take Action: Data privacy and GenAI

5 minutes

Reflection

Learning intentions

Students will:

explore large language models (LLMs) and how they function.

Success criteria

Students can:

explain how LLMs process inputs and create outputs
describe data safety, risks and ethical considerations associated with LLMs.

While this lesson is signposted as 60 minutes, it would be most effective to teach it over several lessons.

Tune In: When do we interact with AI?

Step 1

Start the lesson by asking students to write down a definition of ‘artificial intelligence’ (AI) in their own words.

Invite students to turn and talk with a partner about their definitions. Ask them to note down any questions or points of disagreement/uncertainty that arise from sharing their definitions.

Open the discussion with the class about their AI definitions, encouraging students to share their responses. As students contribute, create a ‘to-be-checked’ list on the board of all the uncertainties and questions that are raised. See if you can write a class definition of ‘artificial intelligence’ that you can all agree on (this might not be possible yet!).

Suggested explanation: Today, when many people think of AI, they only think about tools like ChatGPT or other chatbot systems that can hold a conversation or write text. While these are familiar examples of AI, they represent only 1 type of AI, known as a large language model (LLM). But, AI is actually a much broader area of computer science that focuses on systems capable of performing tasks that would normally require human intelligence. These tasks can include things like understanding language, recognising patterns, making decisions, and solving problems. Modern AI systems can 'learn' from data and improve their performance over time without being explicitly programmed for every task.

Learning extension: An alternative approach to developing a definition of AI and collective understanding is to start by listing all the attributes students think define ‘intelligence’, e.g. be creative, solve problems, use and understand language, apply knowledge/skills to different contexts, feel and express emotions, be imaginative, have empathy, etc.

Step 2

Ask students to create a list of all the times in a day when they interact with AI.

Study and school work (Do they use AI for research or generating ideas? Do they use AI as a tutor or for spelling/grammar checks? Do their teachers use AI? Does the school use AI for administrative tasks?)
Purchases and shopping (Does their bank use AI? Self check-outs? Online shopping? Do they use password management systems?)
Relaxation and leisure time (Is AI used in creating the media they consume? Do they use AI to help with personal projects? Where do the recommendations they get come from?)
Community participation (Do any of their sporting or social clubs use AI? Do any community or government systems use AI?)

Record this list on the board or ask students to add their responses to an online class board.

Step 3

Use students’ examples of all the ways they interact with AI, along with the class definition, to discuss the impacts of AI on our day-to-day lives. Encourage them to consider the positive and negative effects, and why.

Find Out: Generative AI

Step 1

Explain that the class will now dive deeper into understanding what AI is and isn’t, as well as the different types there are. Invite students to share any prior knowledge they have of different types of AI (e.g. Generative AI, machine learning, LLMs, machine vision, etc).

Display the What is AI? Visual Explainer, slides 1–9, to explicitly teach the differences between rule-based systems, predictive AI systems and generative AI systems, and to introduce machine learning and large language models (LLMs). Use the presentation notes provided in the Teacher Content Information to support your explanations of this information. As you move through the Visual Explainer, allow students time to think-pair-share when you check for understanding or when you ask them to draw on their prior knowledge. Consider using thinking routines to further scaffold learning.

Learning extension: At this point, you could return to the list of possible interactions with artificial intelligence from the Tune In activity and re-evaluate how often and when students actually interact with AI on a daily basis.

Highlight that just because something is ‘smart’ tech like a smartphone, smart TV, or smart board, doesn’t mean the system is AI. Also, not all AI generates content, and many AI systems are used for prediction and classification.

Step 2

Explain to students that they will now play a game. Make sure each student has a piece of paper and a pen. They are going to write a sentence collectively. Every student writes the first word of their sentence at the top of the piece of paper. Once everyone has written a word, they pass their paper to the next student (on their right or left, depending on the class layout). This student will then write the next word in the sentence and pass the paper to the next student, continuing until the sentence is complete.

Outline the rules of the game:

You may only write one word.
You may modify the word that came before yours by extending it (e.g. run → running).
Your sentence must make sense.
You must use correct punctuation and grammar.
You can end the sentence at any time (as long as it makes sense).

Play the game until at least 5 sentences have been created and invite students to read their sentences aloud. Ask the class, ‘How did you choose which word to put next in the sentence?’ Call on a few volunteers to share their responses. Student answers might include:

I read the sentence that was already there and used that to figure out the next word.
I read a common phrase and just put the word that normally comes next.
I thought about what would make the sentence make sense. (prediction)
I chose a word that would make the sentence more interesting or funny.
I thought about what kind of sentence it was, like a question or a statement.
I tried to make sure the sentence flowed properly. (prediction)
I picked a word that would surprise people.

Highlight that in this game, students used prediction and context to choose the next word. It demonstrates a basic simulation of how an LLM generates an output. Explain that if students were the equivalent of an AI tool generating these sentences, the word choices (i.e. the output) are made because of ‘system-level prompts’ that are like instructions written by the platform and form the basis for setting AI’s behaviour, tone, or scope before any user input. Users usually don’t see or write system-level prompts unless they're building their own AI tool (which students will explore in more detail in Lessons 2 and 3).

At this point, it may be worthwhile to clearly distinguish between different types of prompts (commonly assumed to have only one meaning), emphasising that there are those that are ‘system-level prompts’ – the instructions and rules for how an AI behaves and generates output; and prompts – what you ask the AI tool to do.

These ideas will tie into the next section, where students explore in more detail how an LLM generates an output. Keep a note of the student's answers to how they chose which word to put next in the sentence-building activity, so you can refer back to them.

Step 3

Display the What is AI? Visual Explainer again and use slides 10–14 to explain in more depth how an LLM chooses the next word in the sentence. Explore how it

converts language into tokens
tokens into embedding vectors
compares embedding vectors for patterns and similarities
uses self-attention to understand context
and generates an output.

Use the presentation notes provided in the Teacher Content Information to support your explanations of this information.

When discussing creating the output, link the concepts of predicting and context back to how the students created their sentences in the game they just played.

Step 4

Remind students that the output of an AI tool is only as good as the training data and the instructions written by the developers. Remind them that to build an LLM, you need vast amounts of language (in the example on slide 9, Meta’s Llama GenAI was trained on 15 trillion tokens, or around 11 trillion words).

Ask students, ‘If you needed a very large amount of text to train an LLM, and you needed it to understand how all sorts of different users would write, and be able to respond appropriately, where could you get this text from?’

Elicit responses and discuss students’ ideas. Through this discussion, identify that the training data for an LLM needs:

to incorporate a range of writing styles from academic to professional to colloquial.
permission to use the text, meaning you must own the copyright, have permission from the author, or have the text in the public domain (such as Wikipedia).
new language samples to be added to the training data to keep it up to date (because language evolves).

Explain to students that this means, for most LLMs, much of the language in their training data comes from crawling the web and ‘scraping’ openly available data. This is then supplemented by other sources, including public-domain books and research papers.

Guide students to understand that this scraped data includes forums, blog posts, comments sections, opinion pieces and more. Inform students that this text is not read or fact-checked before being incorporated into the training data because the amount of text is vast (how long would it take you to read 73 million textbooks?).

Point out that for training data purposes, an AI tool doesn’t need to have accurate information; it just needs to have accurate examples of how people use language.

Step 5

Explain to students that the training data that goes into an LLM is just one part of the AI information ecosystem. This LLM sits within a GenAI system with many components, each of which adds more data to the AI information ecosystem.

Show students What is AI? Visual Explainer slide 15 to explore these different components. Then display slide 16 to discuss the different information that these components bring to the ecosystem, including:

user prompts
inputs
guardrails
Feedback loops
regulations.

Emphasise that a GenAI tool isn’t ‘looking up the truth’. Reiterate that an LLM generates text by predicting which words are most likely to come next. Just like in the previous game, GenAI doesn’t necessarily produce an output that is accurate and factual; instead, it creates an output that it predicts will make sense.

Step 6

Organise the class into small teams and run a quick-check quiz for understanding. Display slide 17 of the What is AI? Visual Explainer to run the quiz to ensure that students understand that an LLM generates text by predicting which words are most likely to come next, and that GenAI doesn’t necessarily produce an output that is accurate and factual, but rather, one that it predicts will make sense. AI doesn’t know what is true! That’s why it’s important to think critically and check information. Suggested explanations to answers are below (and are also listed in the Teacher Content Information):

1. What is a Large Language Model (LLM) mainly doing when it generates a response? A. Searching the internet for the correct answer B. Copying information from a textbook C. Predicting the most likely next words in a sentence D. Asking a human for help	Correct answer: C Predicting the most likely next words in a sentence.
2. Why might an AI tool give an answer that sounds correct but is actually wrong? A. It is broken B. It is trying to trick the user C. It predicts what sounds right D. It only uses old information	Correct answer: C It predicts what sounds right.
3. GenAI tools always give accurate and factual answers because they are trained on lots of data. True or false?	Correct answer: False GenAI tools do not always give accurate answers because they predict what sounds right, not what is true.
4. A student asks an AI tool a question and gets a confident, detailed answer. What should the student do next? A. Trust the answer straight away B. Check the information using other reliable sources C. Ignore the answer completely D. Share it with others without checking	Correct answer: B Check the information using other reliable sources.
5. In your own words, explain the difference between: • predicting what sounds right • knowing something is true	Example answer: • Predicting what sounds right means choosing words that fit together and make sense in a sentence. • Knowing something is true means the information is accurate and can be supported by evidence or reliable sources. (Answers will vary, but should clearly show this distinction.)

Take Action: GenAI and digital footprints

Step 1

Explain to students that now they understand how an LLM builds output, and what goes into the training data, ask ‘How useful and accurate do you think an LLM would be for the following tasks?’

Organise the class into groups of 3–4, and provide each group with a Student Worksheet. Instruct students to decide whether they would use an LLM for any of these tasks:

Proofreading an essay for spelling and grammar errors.	Providing advice on friendships.
Teaching you how to solve maths equations.	Answering questions about political parties.
Writing a geography essay discussing the livability of your local community.	Operating as a search engine.
Explaining a historical event.	Answering questions about events hosted at your school.
Providing medical advice.	Writing a short story for an English class.
Summarising a research paper.	Writing your resume for you.
Providing you with topic ideas for your history project.

Ask students to consider the information they would need to input, the information in the AI ecosystem, and how the LLM structures the output. Point out that they need to justify their answer in 2–3 sentences. Alternatively, allocate 2–3 tasks to each group to share with the class, rather than responding to each one.

Once groups have completed their worksheet, bring the class back together to discuss students' responses and reasoning.

This activity is a great opportunity for students to practise their critical thinking skills. For most of these tasks, there could be caveats provided (I wouldn’t use GenAI for… unless … / I would use GenAI for… but …), and students should be encouraged to explore them.

Learning support: Instead of having students work in groups, choose 2–3 of these tasks and work through them together as a class.

Step 2

Explain that we’ve learnt how LLMs are trained on data in the form of text. However, AI systems can collect a wide range of data. Choose one of the digital systems from the list below (e.g. a streaming platform) and ask students, ‘What type of data does this system collect?’ For this example, students may suggest:

which shows you watch
which shows you start and then leave without finishing
how long you spend watching
what genres of shows you prefer (mystery, documentary, rom-com, etc.)
which actors and actresses you prefer to watch
the text you type into the search bar
the languages you watch shows in
the style of show you prefer (cartoon, live action, anime, etc.)
any information from your user profile (name, age, system language, etc.)
your location.

Ask students: ‘Why does this system collect all of this data? What does it do with it?’ Student answers may include:

Predict which shows you may want to watch next.
Give preference to shows that are in your target language(s).
Show you certain advertisements based on your viewing habits or profile information.

Step 3

Explain that students will next explore a range of digital systems that collect different types of data, which is then used by an AI system. Direct students to return to their groups and provide them with the following list of digital systems. Ask each group to choose 2 AI systems and compare and contrast the data collected by each, using the space provided on page 3 of their Student Worksheet.

Fitness tracker (Fitbit, Apple watch, etc.)
Social media app (Instagram, Facebook, TikTok, etc.)
Streaming platform (Netflix, Disney+, Amazon Prime, Stan, etc.)
Digital school tools (NSW Educhat, Seesaw, Compass, Google Classroom, etc.)
Self-check-out (supermarket, retailers)
Sporting club registration (AFL, soccer, tennis, ballet, etc.)
Traffic monitoring cameras (phone usage, red light, speed detection)

Learning support: If students are struggling to think about the types of data that are collected by apps, suggest the following: locations, age and date of birth, device ID, browsing history, purchasing history, images and photographs, biometric data (fingerprints and face), contacts, school grades, financial information and health information.

Monitor groups’ responses in their Student Worksheet to ensure students are evaluating the appropriateness of the data collection by the systems they’ve chosen to analyse, i.e. What data seems essential vs ‘nice to have’? Who benefits most from collecting extra data? Which risks are most serious: privacy, profiling, errors?

Return together as a class and invite groups to share their responses. Then, ask students:

What was surprising to you, or something that you had never thought about before?
Which systems do you think pose the most risk?
Which systems do you think have (or need to have) the strongest safeguards?
Which data would pose the biggest risk if it were used incorrectly?
Who might be harmed if something goes wrong with the system’s data?

Learning extension: Ask groups to classify the systems they analysed using labels such as green = low risk, yellow = medium risk, and red = high risk.

Step 4

Based on the students’ new understanding of data collected by digital systems and how GenAI systems work, draft a set of rules to establish healthy, safe boundaries for the use of GenAI. Consider which information (data) is safe to provide to the GenAI model, and which information should not be provided or is risky.

Emphasise that digital systems don’t just ‘work’ – they collect data. Healthy tech use includes noticing what data is collected, who benefits, and what risks exist.

Step 5

As an exit ticket, ask students to answer the following questions:

Select all that apply: In your own words, an LLM makes answers by
- looking up facts
- predicting the likely next tokens/words
- following fixed rules.
Complete the sentence stems based on one everyday system we discussed.
- One essential data type it needs…
- Once ‘nice to have / extra data type it might collect…
Select all that apply: After today, I will be more careful about
- what I type into tools
- checking answers
- what data apps collect
- all of these.

Reflection

Ask students to rewrite their definition of artificial intelligence in their own words. This can be collected as part of an exit ticket and used to check for understanding. If there are misunderstandings or misconceptions that persist, take note of these and address them at the start of Lesson 2.

Learning support: You could present this to students as ‘I used to think AI was…, but now I think that AI is…’

Introduction to Large Language Models (LLMs)

Go to

Summary

Learning intentions

Success criteria

Lesson Guides and Printables

Lesson Details

Skills Required

Curriculum Mapping

Resources Required

UN Sustainable Development Goals

Additional Information

Related Professional Learning

In this lesson

Introduction to Large Language Models (LLMs)

Go to

Summary

Learning intentions

Success criteria

Lesson Guides and Printables

Lesson Details

Skills Required

Curriculum Mapping

Resources Required

UN Sustainable Development Goals

Additional Information

Related Professional Learning

In this lesson