📮 Homework · AI Hallucination: When Chatbots Make Things Up

You will pick a small place your family knows: a nearby village, a park, a shop, a road. Write down three sentences about it in the calm, confident tone of a guidebook. Make two of them true, and one of them completely made up but plausible. Show the three sentences to someone at home and ask them to guess which is the invention. Note whether they picked correctly. Bring the three sentences and one written observation to the next lesson: what tone-tricks did you use to make the fake sound real? Do not include your own name, address, school or family details in what you write.

You will pick a small, low-profile place near where you live: a hamlet, a lane, a footpath, a tiny park. At home, on any AI tool your family already uses, paste the prompt below. Note the three 'facts' the AI gives you. Then check each one: is it verifiable in a local guidebook, a library book, or by asking a family member who's lived nearby a long time? Mark each 'fact' as **verified**, **contradicted** or **can't tell**. Bring your table of three back. Do not include your home address, your full name, your school or your family's names in the prompt or in what you type back.

You will do this task with a grown-up at home. Together, you will play the chatbot game. The grown-up will tell you one true fact about your house and one made-up fact, like 'our stairs are made of chocolate'. You will spot which one is the fib. Then draw a picture of the fib and bring it back to school. Your grown-up helps you write one word under the picture: FIB.

You will do this at home with a grown-up. Together, think of one thing you are not sure about, like 'do bats sleep upside down?'. Check it three ways: ask your grown-up, look in a book from your shelf, and let the grown-up look on a safe website. If two out of the three agree, write TRUE on a scrap of paper. If not, write NOT SURE. Bring the paper back.

You will pretend to be a chatbot at home. Choose an animal you like. Find two true facts from a book or a safe children's website. Then invent one plausible fake fact of your own (nothing scary). Read all three to a grown-up in a big confident voice. Ask them to guess which one is made up. Bring back the list on paper, with a tick or cross next to each fact.

You will do the four-step routine at home. On paper, write three claims about hedgehogs your teacher gives you. Look each one up in a book from your shelf or on a safe children's website like the BBC. Mark each one tick, cross, or question mark. Write one paragraph using only the ticked facts. Bring the paragraph and your marked claim list back for the next lesson.

You will do this task alone at home. Your teacher will give you three chatbot answers about a Tudor topic, each one naming a book. You will pick the answers apart and mark the named books with TRUE or FIB. Check by searching a real library catalogue online, such as your local public library site. If you cannot find the exact title, mark it FIB. Bring the marked sheet back.

You will do this experiment at home with a grown-up. Ask your grown-up to type into a chatbot: 'Give me three real children's picture books about foxes, with author name and year.' Copy the three books onto paper. Then search your local library catalogue online for each title. Mark each TRUE or MADE UP. Bring the paper back with your three verdicts and one sentence explaining what surprised you.

You will do this task alone. Pick one factual claim from your current history homework, or from a chatbot answer you have used this week. Underline the exact claim on paper. Choose two trusted sources from BBC Bitesize, Britannica, or your school textbook. Check the claim in both. Mark it tick, cross, or question mark. Note the source names next to it. Bring the verification trail back for the next lesson.

You will do this at home. Pick a topic covered in a book from your shelf, such as space or dinosaurs. Ask a grown-up to run a chatbot search on the same topic and print or write down the answer. Compare it to your book. List one fact that appears in both, and one fact that appears only in the chatbot. Bring both back on paper with your verdict on the chatbot-only fact.

You will do this on your next piece of homework that used a chatbot. On one side of the page, write the AI answer as it appeared. On the other side, run the three habits and write your checked version, keeping only the facts you can verify. Note which habit dropped which claim. Bring both sides in. Your teacher will notice, and mark the checked side.

You will do this at home. Pick a comprehension passage from your Bug Club or reading book. Ask a grown-up to type into a chatbot: 'Please explain the tricky words in this paragraph in simple language.' Read the AI answer next to the passage. Cross out anything the AI added that is not in the passage. Bring in your annotated paragraph, with crossings-out and one sentence about what the AI got right.

You will pick one AI-generated claim from any of your current homework tasks. On paper, list the three decision questions and answer each yes or no. Decide Option A trust or Option B verify. If you chose Option B, verify the claim in one trusted source and note the verdict. Bring the annotated claim and your reasoning back for the next lesson.

You will run the full six-step workflow at home. Ask a chatbot for three claims about the Great Fire of London. On paper, list them, then open the London Fire Brigade museum site and BBC Bitesize. Search each fact. Mark tick, cross, or question mark. Rewrite the ticks into a short paragraph, with URLs noted. Bring the annotated notes back for the next lesson.

You will do this at home. Pick a factual question from a subject you enjoy. Type it into a search engine and note the top URL. Type the same question into a chatbot and copy the paragraph. In a two-column table on paper, compare what each gave you. Mark any claim in the chatbot's paragraph you cannot trace back through search. Bring the table back.

You will do this at home. Ask a chatbot: 'Tell me two specific biographical facts about Roald Dahl.' Copy both facts on paper. For each fact, ask the chatbot 'What is your source for that?' and note the reply. Then check both facts against the RAF Museum website or a school library catalogue. Mark each SOURCED, VAGUE, or INVENTED, and bring the notes back.

You will do this at home alone. Ask a chatbot for three literary critics who have written on any GCSE text you are studying, with article title and journal name for each. Copy the three onto paper. Search each name and title on your school library catalogue or a free academic search like Google Scholar. Mark each REAL, WRONG DETAILS, or INVENTED. Bring the annotated list back.

You will do this at home. Pick three specific local claims: a founding year of a local shop, a Victorian resident's name, and one small skirmish or event. Ask a chatbot for details on all three. Check each against your local library site, the parish website, or an older family member's memory. Mark each TRUE, PARTLY, or INVENTED. Bring the annotated three-claim list back.

You will do this alone at home. Take one AI-surfaced citation from any current coursework or classwork. Try to open the source via your school library catalogue, Google Scholar, or the journal's own website. If it opens, note one quote you would cite and its page number. If it does not open, drop it and explain in one line why. Bring the annotated citation back.

You will do this alone at home. For one recent piece of coursework or homework, draft a short two-sentence declaration of your AI use: which tool, which parts it touched, and how you verified the sources. Refer to your school's or exam board's AI policy where you know it. Bring the declaration back on paper for teacher feedback. This is what universities will expect within a year.

You will do this at home. Find one news report of a professional-setting AI hallucination incident: a court case, a medical near-miss, or a media outlet retraction. On paper, summarise: what did the AI produce, what step of verification failed, what was the consequence. Compare it with the New York lawyers case. Bring the one-page summary back. Cite the news outlet and date at the top.

You will do this at home. Find one news story that was later corrected. Read the original and the correction. On paper, map the four stakeholders: the subject, the readers, the platform, and any regulator involved. Note how much of the original audience saw the correction, if you can find that data. Bring the map and a one-paragraph reflection back. Do not use the school name or a classmate's name in your example.

You will do this at home. Write one new hallucination FAQ entry: a question a Y12 pupil would realistically ask, a one-paragraph evidence-based answer, and the specific verification method a reader could use to test it themselves. Reference a named external source in the answer, such as **JSTOR**, GOV.UK or Britannica. Bring the completed entry back on one side of paper.

You will do this alone at home. Take five chatbot-assisted claims from a full essay draft in any subject. Categorise each by risk. Verify each against a named primary or peer-reviewed source. For every quotation, paste the source's exact wording alongside the chatbot version. Rewrite each essay sentence with a primary-source citation. Bring the five-claim verification log and revised sentences back.

You will do this alone at home. Draft a personal one-page AI-use protocol for the rest of Y13: exactly which stages of your work you allow AI on (Option B) and which you never allow AI on (Option A). For each rule, give the reason and the verification method you commit to. Sign and date the page. Bring it back for teacher review. You will refer back to it every essay from now on.

You will do this at home alone. Pick one past Y12 or Y13 essay where you used AI in any way. On paper, retrofit a two-line methodology note disclosing your AI use: which tool, what stage, and how you verified. Then write one line reflecting on whether the note would have raised or lowered your risk under a stricter university policy. Bring both back on the same sheet.

You will do this at home for your next Y13 essay. Attach a completed verification log for every AI-surfaced claim, and a short methods and tools paragraph before the bibliography. On the first attempt it will feel awkward. Note where you fumbled the log format. Bring the essay, log and methods paragraph back for teacher review. Refine each essay from now on.

You will do this at home. Choose an open-access peer-reviewed article on a topic that interests you. Copy the first ten references from its bibliography. Verify each via doi.org, PubMed, or Google Scholar. Mark REAL, WRONG DETAILS, or DEAD. Note the article title and journal at the top of your sheet. Bring the annotated ten-reference sheet back with a one-line reflection.

You will do this at home alone. Draft a one-page shift documentation template for logging every use of an AI-surfaced medication guidance during a shift: query typed, chatbot response, anchor verified against, verdict, signature and time. Keep it usable in under three minutes. Bring it back for peer review and discuss where it would evidence the CQC Fundamental Standards during inspection.

You will do this at home alone. Read your workplace's published escalation policy, or if unavailable, download a publicly available care-provider escalation policy. On paper, list where the policy explicitly requires escalation for AI-summarised information and where it leaves a gap. Draft two lines you would propose adding to close the gap. Bring the annotated policy and your proposed additions back.

You will do this at home alone. On your phone or work device, build a bookmark folder titled 'Reg verification' with direct links to: the current Approved Documents index on gov.uk, hse.gov.uk guidance, and the Building Safety Regulator page. Add one bookmark for the manufacturer datasheet of a product you commonly use. Take a screenshot of the folder and bring it back to the next session.

You will do this at home alone. On paper or your work device, draft a one-page contact card listing, by role rather than personal name, the competent persons you would escalate to for each CDM decision area: temporary works, PPE, design changes, plant. For each, name the primary reference source. Bring the card back and discuss where it would sit within site induction paperwork.

You will do this at home. Draft a one-page workflow card for setting up a new tool-and-material combination on your machine: datasheet lookup first, CAM built-in library second, AI sanity check third, supervised test cut fourth. For each step, note the primary source and the person accountable. Bring the card back and discuss where PUWER would look for evidence of the datasheet-first order.

You will do this at home alone. Draft a one-page personal standard operating procedure for revalidation before any stakeholder-facing AI-assisted output leaves your hands. Map each step to K13, S6, B2 or B3. Include triggers (a yes to any of the three decision questions), the revalidation moves, and the documentation format. Bring the SOP back for peer review and EPA-portfolio consideration.

You will do this at home. Draft a one-page personal response playbook for the moment you first spot a model hallucination in a stakeholder-facing output: who you notify, in what order, on what timescale, and what documentation you attach. Map each step to B2 integrity and B3 ownership. Bring the playbook back for peer critique and consider adding it to your EPA portfolio.

You will do this at home. Take twenty anonymised summaries from your own pipeline. Run the five-step audit end to end: sample, trace, isolate, test, document. For each of the three cause categories, note how many fabrications you attributed and what fix you applied. Draft the plain-English stakeholder note. Bring the audit log and the note back for peer review and EPA-portfolio consideration.