{"course":{"slug":"ai-for-teachers","title":"AI for Teachers","url":"https://addaly.com/learn/ai-for-teachers","level":"beginner","track":"work"},"lesson":{"slug":"where-it-is-reliable-and-where-it-is-not","title":"Where it is reliable, and where it is not","url":"https://addaly.com/learn/ai-for-teachers/where-it-is-reliable-and-where-it-is-not","position":3,"of":57,"minutes":9,"answer":"Reliability tracks how many times a fact has been written down by people, so the model is sound on photosynthesis and unsound on the founding date of your district.","keyIdea":"Reliability tracks how many times a fact has been written down by people, so the model is sound on photosynthesis and unsound on the founding date of your district, and you can predict which before you ask."},"body":{"format":"markdown","text":"## One question to ask before you ask\n\nThe first lesson gave you the mechanism: the model guesses the next word from what people have written. That gives you a tool for predicting its reliability before you type a single prompt.\n\nAsk yourself: **how many times has this exact thing been written down?**\n\nThe boiling point of water at sea level has been written down millions of times, in every language, in every textbook, and it is always the same. The model has seen it so often that its guess is effectively a fact. The population of Kisumu at the 2019 census has been written down a few hundred times, inconsistently, in documents that also contain other numbers. The model's guess is a plausible-looking number in the right range. Sometimes it is right. You cannot tell from the answer which time this is.\n\nThat is the whole rule. Everything below is that rule applied.\n\n## Reliable ground\n\nThese are the areas where the model's answers are nearly always sound, because the writing it learned from is vast and agrees with itself:\n\n- **Core school science and maths.** Newton's laws, photosynthesis, the water cycle, quadratic equations, the periodic table. Explanations of these are excellent and the errors are rare.\n- **Grammar and writing in major languages.** English, Spanish, French, Hindi, Mandarin, Arabic, Portuguese. It has read more of these than any human.\n- **Well-known history.** Dates and causes of major events. The more written about, the safer.\n- **Explaining a concept several ways.** This is generation, not recall, and it is what the mechanism is best at.\n\n## Unreliable ground\n\nAnd here the answers look identical but should be treated as drafts to check:\n\n- **Local specifics.** The founding date of your school, the name of your district's first commissioner, which crops grow in your division. Written down rarely; often wrong.\n- **Precise numbers.** Populations, distances, percentages from a specific report. It produces a number in the right range because the right range is what appears in text. The exact figure is a guess.\n- **Citations.** Book titles, page numbers, journal articles, quotations attributed to a named person. It has learned what citations look like and will generate one that looks perfect and does not exist. Every citation gets checked. No exceptions.\n- **Recent events.** Anything after its training cutoff, which is typically several months to a year before the date you are using it. Unless it searched the web (the next lesson) it is guessing from older text.\n- **Small languages and regional literature.** A poet who wrote in Bhojpuri or Tigrinya has been written about far less than Shakespeare, and the model will confidently give you a biography assembled from fragments.\n- **Anything about your specific syllabus.** Which chapters are in the Class 9 board paper this year, what the WAEC marking scheme allocates for a question. It has seen some of this and much of it is out of date.\n\n## The trap in the middle\n\nThe dangerous cases are the ones that sit between the two lists: a well-known topic with a specific detail inside it.\n\nAsk about the French Revolution and the overview will be sound. Ask what Robespierre said on a particular date and you have crossed into a rarely-written-down specific, and the model will supply a quotation that sounds exactly right. The topic was reliable; the detail is not. You have to notice the moment the question changes shape.\n\nA useful habit: when reading an answer, underline every proper noun, number, date and quotation. Those are the specifics. Everything underlined is on the second list until you have checked it. Everything else is probably fine.\n\n## Why confidence tells you nothing\n\nThe tone does not change between the lists. The mechanism produces the most likely next word whether the underlying knowledge is a million textbooks or three blog posts, and the most likely next word after \"The population of Kisumu is\" is a number, stated plainly. There is no internal signal of uncertainty reaching the sentence. Some tools now add hedges like \"approximately\", but these are trained habits, not measurements.\n\nSo you cannot use the answer to judge the answer. You use the question. That is why the habit above works and why \"it sounded sure\" is never a reason.\n\n## Practice for your subject\n\nWrite down five questions from your subject: two you expect to be on reliable ground, two on unreliable ground, one in the middle. Ask them. Check the answers against a source you trust. Most teachers find their predictions are right four times in five; and the one miss teaches them where their subject's specifics hide.\n\n> Figure: The reliability of an answer is set by the question. A two by two grid, rows Topic written about a great deal and Topic written about rarely, columns You ask for the overview and You ask for a specific detail: Topic written about a great deal and You ask for the overview: Sound. Causes of the French Revolution.; Topic written about a great deal and You ask for a specific detail: Confident and unchecked. What Robespierre said on a named date.; Topic written about rarely and You ask for the overview: Thin but usually harmless. Farming in your division.; Topic written about rarely and You ask for a specific detail: A guess in the right range. Your district's 2019 census figure.. The dangerous cell is the top right: a well-known topic with a rarely-written-down detail inside it. The tone of the answer is identical in all four. Underline every proper noun, number, date and quotation, and you have marked the right-hand column."},"check":{"question":"A geography teacher asks a chatbot two questions: the boiling point of water at sea level, and the population of her town at the last census. She gets confident answers to both. Which is the honest assessment?","options":[{"text":"Both are factual questions with definite answers, so both are equally likely to be right and equally worth trusting.","correct":false,"tempting":"Treats correctness as a property of the question type rather than of how often the answer appears in training text; a rare local figure has been written down a handful of times, a physical constant millions."},{"text":"Neither answer can be trusted, because the model does not know anything and every answer it gives is a guess of the same quality.","correct":false,"tempting":"Overshoots: a guess informed by millions of consistent examples is reliable in practice, and refusing to distinguish leaves the teacher unable to use the tool at all."},{"text":"The boiling point is almost certainly right; the population may be a plausible invention and needs checking against the census.","correct":true},{"text":"The population is the more reliable of the two because it is a specific number, and the model handles numbers better than it handles science explanations.","correct":false,"tempting":"Inverts the mechanism; specific numbers with few occurrences in text are among the least reliable outputs, and the model has no arithmetic or lookup behind them."}]},"prev":{"slug":"the-free-tools-you-actually-have","title":"The free tools you actually have","url":"https://addaly.com/learn/ai-for-teachers/the-free-tools-you-actually-have"},"next":{"slug":"why-it-cannot-look-it-up","title":"Why it cannot look it up","url":"https://addaly.com/learn/ai-for-teachers/why-it-cannot-look-it-up"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Where it is reliable, and where it is not', in AI for Teachers. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-teachers/where-it-is-reliable-and-where-it-is-not (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Where it is reliable, and where it is not","courseTitle":"AI for Teachers","module":{"index":1,"slug":"understanding-the-machine","title":"Understanding the machine you are about to use"},"answer":null,"keyIdea":"Reliability tracks how many times a fact has been written down by people, so the model is sound on photosynthesis and unsound on the founding date of your district, and you can predict which before you ask.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"cols":["You ask for the overview","You ask for a specific detail"],"rows":["Topic written about a great deal","Topic written about rarely"],"type":"matrix","cells":[["Sound. Causes of the French Revolution.","Confident and unchecked. What Robespierre said on a named date."],["Thin but usually harmless. Farming in your division.","A guess in the right range. Your district’s 2019 census figure."]],"title":"The reliability of an answer is set by the question","caption":"The dangerous cell is the top right: a well-known topic with a rarely-written-down detail inside it. The tone of the answer is identical in all four. Underline every proper noun, number, date and quotation, and you have marked the right-hand column.","cellLabels":[["reliable ground","the trap in the middle"],["check it","unreliable ground"]]},"alt":"Figure: The reliability of an answer is set by the question. A two by two grid, rows Topic written about a great deal and Topic written about rarely, columns You ask for the overview and You ask for a specific detail: Topic written about a great deal and You ask for the overview: Sound. Causes of the French Revolution.; Topic written about a great deal and You ask for a specific detail: Confident and unchecked. What Robespierre said on a named date.; Topic written about rarely and You ask for the overview: Thin but usually harmless. Farming in your division.; Topic written about rarely and You ask for a specific detail: A guess in the right range. Your district’s 2019 census figure.. The dangerous cell is the top right: a well-known topic with a rarely-written-down detail inside it. The tone of the answer is identical in all four. Underline every proper noun, number, date and quotation, and you have marked the right-hand column."}},"blocks":[{"t":"h2","id":"one-question-to-ask-before-you-ask","runs":[{"text":"One question to ask before you ask"}]},{"t":"p","runs":[{"text":"The first lesson gave you the mechanism: the model guesses the next word from what people have written. That gives you a tool for predicting its reliability before you type a single prompt."}]},{"t":"p","runs":[{"text":"Ask yourself: "},{"text":"how many times has this exact thing been written down?","b":true}]},{"t":"p","runs":[{"text":"The boiling point of water at sea level has been written down millions of times, in every language, in every textbook, and it is always the same. The model has seen it so often that its guess is effectively a fact. The population of Kisumu at the 2019 census has been written down a few hundred times, inconsistently, in documents that also contain other numbers. The model’s guess is a plausible-looking number in the right range. Sometimes it is right. You cannot tell from the answer which time this is."}]},{"t":"p","runs":[{"text":"That is the whole rule. Everything below is that rule applied."}]},{"t":"h2","id":"reliable-ground","runs":[{"text":"Reliable ground"}]},{"t":"p","runs":[{"text":"These are the areas where the model’s answers are nearly always sound, because the writing it learned from is vast and agrees with itself:"}]},{"t":"ul","items":[[{"text":"Core school science and maths.","b":true},{"text":" Newton’s laws, photosynthesis, the water cycle, quadratic equations, the periodic table. Explanations of these are excellent and the errors are rare."}],[{"text":"Grammar and writing in major languages.","b":true},{"text":" English, Spanish, French, Hindi, Mandarin, Arabic, Portuguese. It has read more of these than any human."}],[{"text":"Well-known history.","b":true},{"text":" Dates and causes of major events. The more written about, the safer."}],[{"text":"Explaining a concept several ways.","b":true},{"text":" This is generation, not recall, and it is what the mechanism is best at."}]]},{"t":"h2","id":"unreliable-ground","runs":[{"text":"Unreliable ground"}]},{"t":"p","runs":[{"text":"And here the answers look identical but should be treated as drafts to check:"}]},{"t":"ul","items":[[{"text":"Local specifics.","b":true},{"text":" The founding date of your school, the name of your district’s first commissioner, which crops grow in your division. Written down rarely; often wrong."}],[{"text":"Precise numbers.","b":true},{"text":" Populations, distances, percentages from a specific report. It produces a number in the right range because the right range is what appears in text. The exact figure is a guess."}],[{"text":"Citations.","b":true},{"text":" Book titles, page numbers, journal articles, quotations attributed to a named person. It has learned what citations look like and will generate one that looks perfect and does not exist. Every citation gets checked. No exceptions."}],[{"text":"Recent events.","b":true},{"text":" Anything after its training cutoff, which is typically several months to a year before the date you are using it. Unless it searched the web (the next lesson) it is guessing from older text."}],[{"text":"Small languages and regional literature.","b":true},{"text":" A poet who wrote in Bhojpuri or Tigrinya has been written about far less than Shakespeare, and the model will confidently give you a biography assembled from fragments."}],[{"text":"Anything about your specific syllabus.","b":true},{"text":" Which chapters are in the Class 9 board paper this year, what the WAEC marking scheme allocates for a question. It has seen some of this and much of it is out of date."}]]},{"t":"h2","id":"the-trap-in-the-middle","runs":[{"text":"The trap in the middle"}]},{"t":"p","runs":[{"text":"The dangerous cases are the ones that sit between the two lists: a well-known topic with a specific detail inside it."}]},{"t":"p","runs":[{"text":"Ask about the French Revolution and the overview will be sound. Ask what Robespierre said on a particular date and you have crossed into a rarely-written-down specific, and the model will supply a quotation that sounds exactly right. The topic was reliable; the detail is not. You have to notice the moment the question changes shape."}]},{"t":"p","runs":[{"text":"A useful habit: when reading an answer, underline every proper noun, number, date and quotation. Those are the specifics. Everything underlined is on the second list until you have checked it. Everything else is probably fine."}]},{"t":"h2","id":"why-confidence-tells-you-nothing","runs":[{"text":"Why confidence tells you nothing"}]},{"t":"p","runs":[{"text":"The tone does not change between the lists. The mechanism produces the most likely next word whether the underlying knowledge is a million textbooks or three blog posts, and the most likely next word after “The population of Kisumu is” is a number, stated plainly. There is no internal signal of uncertainty reaching the sentence. Some tools now add hedges like “approximately”, but these are trained habits, not measurements."}]},{"t":"p","runs":[{"text":"So you cannot use the answer to judge the answer. You use the question. That is why the habit above works and why “it sounded sure” is never a reason."}]},{"t":"h2","id":"practice-for-your-subject","runs":[{"text":"Practice for your subject"}]},{"t":"p","runs":[{"text":"Write down five questions from your subject: two you expect to be on reliable ground, two on unreliable ground, one in the middle. Ask them. Check the answers against a source you trust. Most teachers find their predictions are right four times in five; and the one miss teaches them where their subject’s specifics hide."}]}],"check":{"question":[{"text":"A geography teacher asks a chatbot two questions: the boiling point of water at sea level, and the population of her town at the last census. She gets confident answers to both. Which is the honest assessment?"}],"options":[{"text":[{"text":"Both are factual questions with definite answers, so both are equally likely to be right and equally worth trusting."}],"correct":false,"tempting":[{"text":"Treats correctness as a property of the question type rather than of how often the answer appears in training text; a rare local figure has been written down a handful of times, a physical constant millions."}]},{"text":[{"text":"Neither answer can be trusted, because the model does not know anything and every answer it gives is a guess of the same quality."}],"correct":false,"tempting":[{"text":"Overshoots: a guess informed by millions of consistent examples is reliable in practice, and refusing to distinguish leaves the teacher unable to use the tool at all."}]},{"text":[{"text":"The boiling point is almost certainly right; the population may be a plausible invention and needs checking against the census."}],"correct":true,"tempting":null},{"text":[{"text":"The population is the more reliable of the two because it is a specific number, and the model handles numbers better than it handles science explanations."}],"correct":false,"tempting":[{"text":"Inverts the mechanism; specific numbers with few occurrences in text are among the least reliable outputs, and the model has no arithmetic or lookup behind them."}]}],"order":[0,3,2,1],"why":null},"prev":{"slug":"the-free-tools-you-actually-have","title":"The free tools you actually have","minutes":9},"next":{"slug":"why-it-cannot-look-it-up","title":"Why it cannot look it up","minutes":8}}}