Study Guide Instructor view
When AI Gets It Wrong
Teaching notes for Lesson 8.3, with controls for presenting it.
Keep this page on your own screen. Share only the lesson window. The lesson window holds no teaching notes, so sharing it cannot show them.
Presenting
- Open the lesson in its own window and share that window, not this one.
- Use Next and Previous here, or the lesson’s own, to move through the lesson. Next reveals one thing at a time: the next idea, or the answer to the question on screen, then the next step. Previous takes the last reveal back. The line below says what is on screen and what Next will show.
- Reveal hints and answers when the class is ready, and reset a demonstration before using it again.
The controls need JavaScript and a browser that supports BroadcastChannel. Without them, open the lesson in presentation mode and use its own controls.
The two windows talk to each other only inside this browser, on this device. Nothing is sent to a server, and no account is needed. A second device (a phone or another tablet) cannot follow the lesson window yet.
Now on the lesson window
- On screen
- Next teaching point
Guidance for this point
Lesson plan
- Level
- Beginner
- Time
- About 55 minutes; the steps below add up to 55
Starting from a confident but unreliable homework answer, learners see why fluent text can be false, how the context window and knowledge cut-off limit answers, and how to check claims at their source.
Objectives
- Explain what a hallucination is and why it follows from how a language model writes
- Explain the context window and what happens when a conversation outgrows it
- Explain the knowledge cut-off, and decide whether a question is time-sensitive
- Check a statistic, quotation or reference at its original source
- Explain why grounding an answer in supplied sources reduces errors without removing them
- Name situations where checking AI output is essential
Before the lesson
- Prepare one real reference that exists and one plausible invented one, to show on the whiteboard in step 2 (label which is invented after the vote).
- Have a primary source ready to open live in step 5, such as a published letters collection or an official exam timetable.
Steps
Introduction Opening
On the lesson window now
On screen: the lesson’s title, module and number; Next adds its objectives and key terms, as the lesson page lists them; Next again begins the first step.
Objectives on screen
- Explain what a hallucination is and why it follows from how a language model writes
- Explain the context window and what happens when a conversation outgrows it
- Explain the knowledge cut-off, and decide whether a question is time-sensitive
- Check a statistic, quotation or reference at its original source
- Explain why grounding an answer in supplied sources reduces errors without removing them
- Name situations where checking AI output is essential
Purpose: Starting from a confident but unreliable homework answer, learners see why fluent text can be false, how the context window and knowledge cut-off limit answers, and how to check claims at their source.
1. An answer that sounds expert Think 5 min
On the lesson window now
Purpose: Raise the problem: confidence is not evidence.
Ask
- How could Amira tell which parts are true?
Responses to expect
- Look them up
- Ask the AI if it is sure
Misconceptions to challenge
- A learner may believe: “If it sounds confident, it is probably right.” Respond: Ask what in the reply would look different if a quotation were invented. Nothing would.
In the lesson window
- Think it through: Everything in the reply sounds expert. How could Amira tell which parts are true?
2. Why fluent text can be false Discover 8 min
On the lesson window now
Purpose: Connect hallucination to next-token prediction.
Ask
- What is the most likely text after “Author, Title (Year)”?
- Which kinds of claim are the riskiest?
Responses to expect
- Something shaped like a reference
- Statistics, references, quotations
Misconceptions to challenge
- A learner may believe: “Hallucinations are rare bugs that newer models have fixed.” Respond: They follow from how text is produced; newer models make fewer, none makes none.
On the board or with the pen
- Show a real and an invented reference side by side; let the class vote which is real before revealing.
In the lesson window
- Think it through: A model has read millions of references, all shaped like “Author, Title (Year), Publisher”. When asked for a book on early railways, what is the most likely next text it can produce?
3. What the model can see at once Experiment 9 min
On the lesson window now
Purpose: Make the context window concrete and show its failure without warning.
Ask
- Where is the first instruction now?
- What would you do differently?
Responses to expect
- It has fallen out of the window
- Repeat it; shorter pastes; new chat
Misconceptions to challenge
- A learner may believe: “The model chose to ignore my instruction.” Respond: It can no longer see it. Nothing is chosen.
On the board or with the pen
- Draw a window as a frame on the whiteboard and slide sticky-note “messages” through it as you step through.
In the lesson window
- Check your understanding: After a long chat, an assistant stops following a rule you gave at the start. What is the best thing to do?
4. A model frozen in time Predict 6 min
On the lesson window now
Purpose: Name the knowledge cut-off and the habit of asking “is this time-sensitive?”.
Ask
- Which questions change their answer from year to year?
Responses to expect
- Dates, prices, results, laws
Misconceptions to challenge
- A learner may believe: “A model knows today’s news.” Respond: Only if the app added a search; the model itself stops at its cut-off.
In the lesson window
- Check your understanding: Which question is most risky to answer from a model on its own, without a current source?
5. Checking a claim at its source Apply 8 min
On the lesson window now
Purpose: Practise checking at a primary source; reject checking with another AI.
Ask
- Where were these words first written?
- Why is a second chatbot not a check?
Responses to expect
- A letter, a book, an official page
- It can repeat the same error
Misconceptions to challenge
- A learner may believe: “If two AIs agree, it must be true.” Respond: Agreement is not evidence; both can be wrong in the same way.
On the board or with the pen
- Open a primary source live and annotate where the quotation does or does not appear.
In the lesson window
- Apply it: The assistant gave Amira a quotation from a famous engineer, with a year. What is the best way to check it?
6. Making errors less likely Explain 7 min
On the lesson window now
Purpose: Give habits that reduce errors, and RAG as a built-in version of grounding.
Ask
- What does pasting the real document change?
- Does an answer with links need checking?
- Is an answer from the school’s own policies, using RAG, always right? (the check question)
Responses to expect
- Errors become misreadings, easier to spot
- Yes: open the source
- Less likely to be wrong, not never wrong: it can fetch the wrong passage or misread the right one; “always right” is the misconception to address
Misconceptions to challenge
- A learner may believe: “An answer with sources is always right.” Respond: Open the link and find the claim in it.
In the lesson window
- Check your understanding: A school’s help assistant answers staff questions from the school’s own policy documents, using RAG. Which statement about its answers is accurate?
7. When checking is not optional Check 5 min
On the lesson window now
Purpose: Identify high-stakes contexts.
Ask
- Who would be harmed if this were wrong?
Responses to expect
- Health, safety, money, legal, people decisions, anything published
In the lesson window
- Check your understanding: Which of these most needs every AI-produced fact checked before anyone relies on it?
8. Bring it together Consolidate 7 min
On the lesson window now
Purpose: Retrieval through advice to Amira.
Ask
- Write three pieces of advice for Amira.
- A precise statistic with no source: keep it, ask again, or find it? (the closing task)
Responses to expect
- Check each quotation; check the score; confidence is not proof
- Find it at a primary source or remove it; “ask the assistant again” shows the belief that agreement is evidence
On the board or with the pen
- Collect three pieces of advice from the class on the whiteboard and number them.
In the lesson window
- Think it through: Without looking back, write three pieces of advice for Amira before she hands in her homework.
- Apply it: A friend’s AI-written essay says “72% of UK pupils walked to school in 1950”, with no source. The essay is due tomorrow. What should your friend do?
Lesson recap Closing
On the lesson window now
On screen: the lesson’s objectives again; Next adds the closing question, which asks the class to explain each one in their own words without looking back. It says nothing about what anyone has learnt: use the answers to decide what to revisit.
Points to listen for
- Hallucination follows from next-token prediction.
- Statistics, references and quotations are high risk.
- The context window and the cut-off limit what a model can use.
- Check at the primary source, never with another AI.
Consolidation
- Hallucination follows from next-token prediction.
- Statistics, references and quotations are high risk.
- The context window and the cut-off limit what a model can use.
- Check at the primary source, never with another AI.