What is the AI Skills and who takes it?

Practical artificial intelligence — how it works, how to use it well, and how not to be fooled by it, set by No board, no certificate, no exam. This is a skill, and the only test is whether you can use the thing.. School students from about Class 8, college students, teachers, and working people in Tamil Nadu and beyond who are being handed AI tools without being taught what they are. No coding, no maths beyond arithmetic, no prior computer science.

What is the AI Skills format?

There is no AI exam and this is not preparation for one. It is practice on a skill, in the same format as everything else here: a question, an answer, and an explanation of why. Eleven units, from what a model actually is through to Indian law. Start anywhere — the units do not depend on each other, except that 'How AI gets things wrong' is worth doing early. Written for three people at once: a student who is being told not to use AI, a teacher who is being told to use it, and someone whose job is changing under them. Nothing here depends on which model is best this month. The mechanisms are the same across every chatbot you will meet, and they are what stays true. Free, no login, no certificate. If you want a session run at your school, college or office, that is at /book.

Are these AI Skills practice questions free?

Yes. All 336 questions, with answers and explanations, are on one page with no sign-up, no login and no paywall. The page works with JavaScript switched off and once loaded it works offline.

Where do these questions come from?

They were written by language models and then cross-checked: every question was sent back to a different model with the answer removed, and only the ones where the second model independently chose the same answer are on this page. 0 of 336 have passed that check. Nobody who has sat the AI Skills has reviewed them yet, so treat them as practice, not as an official answer key.

AI Skills Practice Questions 2026 — Free, with Answers and Explanations

Read this before you trust an answer. These questions were written by language models, not cut from past papers. Every one was then sent back to a different model with the answer hidden, and only the ones where the second model independently picked the same answer are on this page — 0 of 336 so far. That is a floor, not a guarantee: two models can agree and still be wrong, and nobody who has sat the AI Skills has reviewed these yet. Use them to practise the shape of the test. For the official format and scoring, go to No board, no certificate, no exam. This is a skill, and the only test is whether you can use the thing.. Found a wrong answer? Tell us and we will pull it.

What the AI Skills actually is

There is no AI exam and this is not preparation for one. It is practice on a skill, in the same format as everything else here: a question, an answer, and an explanation of why.
Eleven units, from what a model actually is through to Indian law. Start anywhere — the units do not depend on each other, except that 'How AI gets things wrong' is worth doing early.
Written for three people at once: a student who is being told not to use AI, a teacher who is being told to use it, and someone whose job is changing under them.
Nothing here depends on which model is best this month. The mechanisms are the same across every chatbot you will meet, and they are what stays true.
Free, no login, no certificate. If you want a session run at your school, college or office, that is at /book.

School students from about Class 8, college students, teachers, and working people in Tamil Nadu and beyond who are being handed AI tools without being taught what they are. No coding, no maths beyond arithmetic, no prior computer science.

Tap an option to check it. Your score appears here.

What AI actually is 40 questions

The mechanism, in plain words: a large language model predicts the next piece of text, one piece at a time, from patterns in the text it was trained on.

1. A student asks a chatbot: "What is the capital of France?" The chatbot replies "Paris". The next time the same student asks the exact same question, the chatbot replies "Paris" again, but sometimes it adds "the city of lights". Which of the following best explains why the chatbot can give slightly different answers to the same question?
Answer: B. The model predicts the next token based on learned probabilities. Even with the same prompt, the stochastic sampling can produce different continuations, such as adding an extra phrase. It does not look up data, does not have stored scripted replies, and does not retain memory between sessions unless explicitly saved.
2. A teacher wants to use an AI tool to help grade short essays. She knows the tool can only work with the text it receives at the moment. Which statement correctly describes a limitation of the tool?
Answer: B. The model generates responses from patterns in its training data; it does not have live access to external sources unless explicitly connected, and it has no persistent memory across uses unless the user provides that context.
3. In a tech demo, a chatbot answers a math problem by calling a separate calculator tool. Which of the following best describes the relationship between the chatbot, the underlying model, and the calculator tool?
Answer: B. The language model itself predicts text but can be programmed to call external utilities. The chatbot orchestrates the call, receives the exact answer, and incorporates it into the generated response. The model does not perform exact calculations internally nor does it retrain on each call.
4. A developer explains that a "model" is not a traditional program written rule by rule. Which of the following statements most accurately captures this difference?
Answer: D. During training, the model adjusts millions of parameters (weights) based on patterns in the training text. This enables it to generate plausible continuations for novel prompts, unlike a rule‑based program that follows explicit, pre‑written conditions.
A 15‑year‑old asks a chatbot, "What is the square root of 16?" The chatbot replies "4". The next day the same question is asked, but the chatbot says "2". Why did the answer change?
5. What best explains this change in answer?
Answer: A. A language model predicts each next token from learned patterns. Even with the same prompt it can generate different but statistically plausible continuations, so the answer can vary.
A teacher wants to use an AI tool to help grade short essays. She knows the tool can only work with the text she feeds it. She gives the tool a paragraph and asks it to score the essay. The tool returns a score of 8 out of 10. The teacher wonders if the tool actually "understands" the essay. She asks the tool, "Why did you give this score?" The tool replies, "Because the essay contains many complex sentences and good vocabulary."
6. What does this scenario illustrate about the AI tool?
Answer: D. The tool is not truly understanding; it generates a plausible explanation by predicting the next text that fits the pattern of how humans explain grading.
A student is curious about how a chatbot remembers previous messages. She asks the chatbot, "What was the first question I asked you?" The chatbot replies, "I don't know."
7. Why does the chatbot say it does not know?
Answer: C. A language model does not retain state across interactions unless the system explicitly stores and feeds past messages back into the prompt.
A developer wants to build a chatbot that can answer questions about a company’s internal policies. The developer trains a large language model on the company’s policy documents and then deploys it as a chatbot. The chatbot is asked, "Can I take a day off tomorrow?" It replies, "I don't know."
8. What is the most accurate reason the chatbot cannot answer?
Answer: A. The model only has the text it was trained on; it cannot look up current employee status or calendar data unless a tool is integrated to provide that information.
9. A teenager asks a chatbot to write a short story about a dragon. The chatbot produces a tale, then the teen asks the same request again and receives a different story. Which statement best explains why the two answers differ?
Answer: C. The model works by predicting the next token based on learned patterns and probability distributions. Each generation involves sampling, so multiple valid continuations exist. It does not retrieve external data, nor does it have a rule to avoid repeats; it simply samples anew each time.
10. In a classroom, a teacher uses a chatbot to help students brainstorm ideas for a science project. The teacher notices that the chatbot sometimes says it "doesn't know" when asked for a very recent discovery. Which explanation is most accurate?
Answer: A. A language model only knows what was present in its training data; it does not browse the web on its own. It will answer only from patterns it has seen, and will say it does not know when the query falls outside that range.
11. A developer explains to a colleague that the chatbot they are using is built on a "model" rather than a traditional program written line by line. Which of the following best captures the difference?
Answer: A. A model is trained on data to capture statistical patterns and then generates text by prediction. Traditional programs consist of explicit, hand‑written rules that dictate behavior for each input.
12. During a tutoring session, a student asks a chatbot to remember the name of their pet for the next question, but the chatbot replies it does not recall the name. Which concept does this illustrate?
Answer: A. Language models have no persistent memory across separate sessions. Only if the surrounding system stores the prior exchange and includes it in the next prompt will the model be able to reference it.
13. A 15‑year‑old asks a chatbot, "What is the tallest mountain in the world?" The chatbot replies, "Mount Everest." A few minutes later, the same question is asked again and the chatbot says, "K2." Which statement best explains why the answers differ?
Answer: D. A large language model predicts the next token from patterns it learned. Even with the same prompt it can generate different but still plausible continuations, so the answers can vary.
14. A teacher wants to use an AI tool to help students write essays. She knows the tool can only process the text that is typed into it. The teacher asks the tool to suggest improvements for a paragraph. Which of the following best describes what the tool is doing?
Answer: B. The tool is a large language model that generates text by predicting the next token from patterns in its training data, not by looking up information or using hand‑written rules.
15. A student is told that a chatbot has no memory between conversations unless something is deliberately kept. Which of the following best explains this limitation?
Answer: C. During use, the model only retains the tokens of the current conversation. Once the session ends, that context is lost unless the user explicitly stores and feeds it back.
16. A developer says, "The chatbot I use is a model, not a program written rule by rule. It can answer questions it has never seen before, but it does not know what it does not know." Which of the following best captures the essence of this statement?
Answer: D. A large language model generates text based on learned patterns; it does not look up facts and it cannot be certain of its ignorance, so it may produce confident but incorrect answers.
17. A student asks a chatbot, 'What is the capital of France?' The chatbot replies, 'Paris.' The student then asks, 'What is the capital of Italy?' The chatbot replies, 'Rome.' The student is confused because they thought the chatbot was looking up the answers in a database. Which of the following best explains how the chatbot generated these answers?
Answer: B. Large language models do not look up facts in a database or use live search unless specifically programmed to do so. Instead, they are trained on vast amounts of text to recognize statistical patterns. When asked a question, the model predicts the next most probable token (word or piece of a word) based on the context provided. Since 'Paris' is the most frequent and likely completion for 'capital of France' in the training data, the model generates it. This is a prediction process, not a retrieval process.
18. A user asks a chatbot to write a poem about the ocean. The user is dissatisfied with the first draft and asks the chatbot to 'try again, but make it sadder.' The chatbot generates a completely different poem. The user assumes the chatbot has a memory of the first poem and is deliberately changing it. What is the most accurate technical explanation for why the second poem is different?
Answer: D. While a chatbot does not have persistent memory across different conversations (sessions), it does maintain context within a single active conversation. When the user asks to 'try again,' the model sees the previous poem and the new instruction ('make it sadder') as part of the current input context. It then predicts the next sequence of tokens that best fits this new combined context. The difference in output is due to the model re-predicting text based on the updated prompt, not because it is 'remembering' the first poem in a human-like sense or accessing a library.
19. A developer is explaining the difference between a traditional software program and a large language model to a new hire. The developer says, 'In a traditional program, I write specific rules for every possible situation. In a language model, I do not write rules for specific answers.' Which of the following best illustrates why a language model can answer questions the developer never explicitly programmed it to handle?
Answer: C. Traditional programs are deterministic; they follow explicit if-then rules written by humans. If a rule is not written, the program cannot handle that case. Large language models, however, are trained on massive datasets to learn the statistical structure of language. They learn patterns, such as how subjects agree with verbs or how questions are typically answered. This allows them to generalize: even if the specific question 'What is the boiling point of mercury?' was not in the training data as a Q&A pair, the model has learned enough about scientific text and language structure to predict a plausible answer. It is not looking up a specific rule or a specific fact, but applying learned patterns.
20. A user asks a chatbot, 'Did the Roman Empire fall in 476 AD?' The chatbot replies, 'Yes, the Western Roman Empire is generally considered to have fallen in 476 AD.' The user then asks, 'But what about the Eastern Roman Empire?' The chatbot replies, 'The Eastern Roman Empire, also known as the Byzantine Empire, continued until 1453.' The user is surprised that the chatbot provided two different dates for the 'fall of the Roman Empire.' What best explains this discrepancy?
Answer: C. Large language models do not possess a single, unified 'fact base' that they consult. Instead, they generate text based on the probability of tokens given the input. The phrase 'fall of the Roman Empire' is ambiguous and historically complex. In many general contexts, the date 476 AD is the standard answer. However, when the user specifically asks about the 'Eastern Roman Empire,' the context shifts. The model has seen many texts discussing the Byzantine Empire and its fall in 1453. Therefore, it predicts the text that best fits the specific context of the second question. This illustrates that the model is not retrieving a single fact but generating a response based on the specific linguistic and contextual cues of the prompt. It does not 'know' the truth; it predicts the most likely continuation of the text.
A student is using a chatbot to help them write a history essay. They ask the chatbot, 'Tell me about the causes of World War I.' The chatbot provides a detailed answer. The student then asks, 'And what were the main consequences?' The chatbot provides another detailed answer. The student notices that the chatbot did not mention the assassination of Archduke Franz Ferdinand in its answer about consequences, even though it was a major cause.
21. Why might the chatbot have failed to connect the assassination to the consequences, even though it was mentioned in the causes?
Answer: B. The core mechanism of a large language model is predicting the next token. Unless the conversation history is explicitly fed back into the model for the next prediction, or the model is designed with a specific memory mechanism, it treats each new prompt somewhat independently. Option B explains this lack of inherent memory between turns. Option A is incorrect because AI does not have intentions like testing users. Option C is unlikely given the commonality of the topic and the chatbot's ability to answer the first part. Option D is incorrect as chatbots are not people.
Imagine you are talking to a chatbot. You ask it, 'What is the weather like in London today?' The chatbot replies, 'I cannot provide real-time information like current weather conditions.' You then ask, 'Can you tell me about the history of the London Eye?' The chatbot provides a detailed history.
22. What best explains why the chatbot could answer the second question but not the first?
Answer: B. Large language models generate text based on patterns learned from their training data. This data is a snapshot of information up to a certain point in time and does not include live, real-time data like current weather. Historical information, however, is readily available in its training data. Option A is too specific; the limitation is about real-time data, not just 'current events'. Option C is incorrect because AI does not have intentions like being deliberately unhelpful. Option D is incorrect because the chatbot *can* access information, just not real-time data, and it's not solely a search engine.
A student is using a chatbot to help them understand a complex scientific concept. They ask, 'Explain quantum entanglement in simple terms.' The chatbot provides an explanation. The student then asks, 'Can you give me an example of how this is used in technology today?' The chatbot responds, 'I cannot provide examples of current technological applications for quantum entanglement.'
23. What is the most likely reason the chatbot gave this response?
Answer: C. This question tests the understanding of training data limitations. The chatbot can explain the concept because that information was likely in its training set. However, if its training data did not contain specific, concrete examples of *current* technological applications, it cannot 'look them up' or 'know' them unless explicitly told to access a tool that can. Option A is too strong; it could explain the concept. Option B is incorrect because the limitation is about data availability, not a design choice to avoid speculation. Option D is incorrect as chatbots are not people.
A programmer is building a chatbot. They have two main approaches in mind. Approach 1 involves writing thousands of specific rules: 'IF the user says X, THEN respond with Y.' Approach 2 involves using a large language model trained on a massive dataset of text and code. The programmer chooses Approach 2.
24. What is a key advantage of using a large language model (Approach 2) over a rule-based system (Approach 1) for creating a chatbot that can handle a wide variety of user inputs?
Answer: B. This question probes the fundamental difference between rule-based systems and large language models. Rule-based systems are rigid; they only respond to exact or very similar inputs to their predefined rules. Large language models, by learning patterns from vast amounts of text, can understand and generate responses to a much wider range of inputs, including variations in phrasing, and can create novel combinations of ideas. Option A is incorrect; while rule-based systems are predictable, they are also brittle and can easily fail on unexpected input, leading to a lack of response or a generic 'I don't understand.' Option C is incorrect; training a large language model is computationally intensive and time-consuming, far more so than writing a set of rules for a limited scope. Option D is incorrect; a base large language model does not inherently access the internet in real-time; that functionality is usually added via tools.
25. A student opens a brand new chat window with a large language model and immediately types, "What was the name of the dog I told you about yesterday?" The chatbot responds that it does not know of any dog. What best explains this behavior?
Answer: D. Large language models treat every new conversation as a completely blank slate. They have no built-in persistent memory of past sessions unless the application wrapper explicitly saves and feeds prior text back into the context window for the new prompt.
26. A user types a prompt into a chatbot and receives a fluent answer. Curious about how the software works under the hood, the user looks at the underlying model. What best describes the fundamental mechanism of that model?
Answer: D. A large language model is fundamentally a statistical next-token predictor. It does not search databases or follow human-coded rulebooks; it simply calculates the most probable text continuation based on the patterns it learned during training.
27. A teacher asks a chatbot to solve a complex math word problem. The chatbot writes out a plausible-sounding solution that includes confident miscalculations, leading to a wrong final answer. When gently corrected, the chatbot immediately apologizes and generates a new, correct response. What core limitation of the model does this reveal?
Answer: D. Models do not have an internal model of truth or self-awareness about their own ignorance. They generate text based on statistical plausibility, meaning incorrect or fabricated answers can sound just as confident and fluent as correct ones.
28. A company uses a chatbot interface for customer service. Behind the scenes, the chat interface is connected to a live flight-status database. When a customer asks if flight 404 is on time, the chatbot queries the database and answers accurately. Why was the model able to provide real-time information it could not possibly have known from its training data?
Answer: D. A large language model on its own cannot look up live information. When a chatbot provides real-time data, it is because the system architecture includes a tool or application programming interface that the chatbot can call to fetch external information and insert it into the conversation.
29. A teenager asks a chatbot to explain why it sometimes gives different answers to the same question asked twice in a row. Which explanation is most accurate?
Answer: C. The model predicts the next token based on probabilities; even with the same prompt, sampling can produce different sequences, so answers can vary. Options A and B describe lookup or stored answers, which is false. Option D claims intentional change, which is not how the model works.
30. In a classroom demo, a teacher types a prompt asking the chatbot to list three ways to reduce plastic waste. The chatbot replies with a list, then the teacher asks the same prompt again and gets a slightly different list. Which statement best explains why the chatbot does not retain the first list?
Answer: B. Large language models have no persistent memory across separate calls; they only consider the text provided in the current prompt. To keep context, the prior conversation must be included. Options A, C, and D describe nonexistent memory mechanisms.
31. A student wonders whether a chatbot "knows" facts like the boiling point of water. Which of the following best describes what the chatbot is doing when it answers "100 degrees Celsius at sea level"?
Answer: C. The model has seen many examples of the phrase "boiling point of water is 100 degrees Celsius" during training and predicts that continuation. It does not perform look‑ups, simulations, or live searches.
32. A developer integrates a language model into a customer‑service chatbot that can also call an external order‑lookup tool. When a user asks, "Did my order ship?", the chatbot replies, "Yes, it shipped yesterday," even though the order was not yet dispatched. Which reason most likely explains this error?
Answer: A. If the integration does not pass the user's query to the external tool, the model will rely on its internal patterns and may guess an answer. It does not have built‑in knowledge of the specific order and does not intentionally fabricate; the error stems from missing the tool call. Options B, C, and D describe behaviors that are not inherent to the model.
33. A 15‑year‑old asks a chatbot, "What is the capital of France?" The chatbot replies "Paris" and then, after a pause, says "I think it might be Lyon, but I'm not sure." What best explains this behavior?
Answer: A. The model predicts each word from patterns it learned during training. It has no memory of earlier parts of the same conversation unless the user repeats them, so it can produce a different answer on a second attempt.
34. During a school lesson, a teacher types into a classroom chatbot: "Explain how a car engine works." The chatbot responds with a detailed paragraph and then adds, "I don't have a picture to show you." Which statement best describes why the chatbot says it does not have a picture?
Answer: A. The model does not search the internet or access external files unless explicitly instructed; it can only generate text based on its training data.
35. A user opens a new chat with a large language model and asks, "What is the weather today?" The model replies, "I don't know." Why does the model give this answer?
Answer: D. The model only knows what it was trained on and cannot access live information unless the system supplies it; therefore it correctly states that it does not know.
36. A software engineer wants to build a customer‑service chatbot that can answer questions about product availability. She decides to use a large language model as the core engine and also program a separate function that checks the inventory database. When a user asks, "Is product X in stock?", the chatbot first says, "I'm not sure," then calls the inventory function and replies, "Yes, it is available." Which of the following best describes the relationship between the chatbot, the model, and the function?
Answer: A. The chatbot is the interface that combines the language model’s generative abilities with external tools (functions) to provide accurate, up‑to‑date information.
37. A student asks a chatbot, 'What is the capital of Japan?' The chatbot replies, 'Tokyo.' The student then asks, 'What is the capital of France?' The chatbot replies, 'Paris.' Which statement best explains how the chatbot produced these answers?
Answer: C. Large language models do not look up facts in a database or follow hard-coded rules for every fact. Instead, they generate text one piece at a time by predicting what is most likely to come next based on the vast amount of text they were trained on. In this case, after seeing 'capital of Japan', the model predicts 'Tokyo' because that sequence is common in its training data. It is not retrieving a stored fact, but generating a plausible continuation.
38. A user asks a chatbot, 'What did we discuss in our last conversation?' The chatbot replies, 'I do not have access to previous conversations.' Why is this the expected behavior for a standard large language model?
Answer: B. A large language model is stateless between separate sessions. It does not have a long-term memory of who you are or what you said in a previous chat window. Each new conversation starts from scratch. If the user wants the model to remember something, the previous context must be included in the current prompt or managed by the application layer, not by the model itself.
39. A user asks a chatbot, 'Is the moon made of cheese?' The chatbot replies, 'No, the moon is made of rock and dust.' The user then asks, 'What is the capital of the country that invented the cheese?' The chatbot gives a confident but incorrect answer about a fictional country. What best explains why the model might give a confident but wrong answer to the second question?
Answer: B. This illustrates the concept of hallucination. Large language models are pattern matchers, not truth detectors. They do not 'know' that the premise of the second question is false. They simply predict what text is likely to follow the prompt. If the prompt implies a fictional scenario, the model may generate a fluent, confident-sounding answer that fits the pattern of a factual response, even if the underlying fact is wrong. It does not have a built-in fact-checking mechanism.
A student is using a chatbot to help with homework. They ask it to explain photosynthesis. The chatbot provides a detailed and accurate explanation. Later, the student asks the same question again, and the chatbot gives a slightly different explanation, using different wording but conveying the same core information. The student is confused because they thought the chatbot would have the exact same answer.
40. Which of the following best explains why the chatbot provided a different answer to the same question?
Answer: B. The core mechanism of a large language model is to predict the next token (word or part of a word) based on the preceding text and the patterns learned during training. Because this is a probabilistic process, even with the same input, the model can generate slightly different outputs each time, as the probabilities for the next token might lead down a different path. Option A is incorrect because LLMs don't typically store pre-written answers in that way. Option C is incorrect because the model isn't 'looking up' information in a database unless specifically told to use a tool for that purpose, and the scenario implies a standard LLM interaction. Option D is incorrect because LLMs do not have intentions or the ability to 'test' users in that manner.

Getting good answers out of it 40 questions

Prompting as a skill that generalises across every tool: say who it is writing for and why; give it the context it cannot see; show it one example of what good looks like; ask for the format you want; tell it what to leave out; iterate rather than re-asking from scratch.

1. You need to ask a chatbot to draft an email complaining to a broadband provider about a week-long outage. Which of the following prompts uses the best combination of specific structural instructions and constraints?
Answer: B. Option B works because it defines the format, gives the context the model cannot see (working from home), states what to include (a refund request), and explicitly tells it what to leave out (aggressive language and legal threats). Option A contradicts itself by asking to be angry and polite at the same time. Options C and D rely on weak folk advice like role-playing expert personas and saying 'make it better' without giving new criteria.
2. You want an AI to draft a leave letter to your manager requesting two weeks off for a family wedding abroad. You previously asked it to write the letter, but the first draft sounded too robotic and formal. Which of the following is the most effective next move?
Answer: B. Option B is the correct iteration strategy. It references the specific problem with the previous output (too stiff), gives a clear stylistic target (normal email between colleagues), retains necessary context (dates and reason), and tells it what to remove. Option A wastes the session history. Option C uses the weakest instruction possible (make it better). Option D relies on the folk belief that being polite to an LLM changes its token prediction mechanics.
3. A teacher asks an AI to create a lesson plan on photosynthesis for twelve year olds. The first attempt covers too much advanced biochemistry that would bore the class. Which of the following is the best prompt to fix the output?
Answer: B. Option B directly targets the flaw in the previous attempt by narrowing the scope (basic input and output equation), changing the explanatory device (kitchen analogy), and specifying a precise format (a table with named columns). Option A relies on the ineffective 'act as an expert' trope. Option C is a vague reprimand that gives no new information. Option D asks for more detail and length, which often makes the AI pad the response with fluff rather than improving clarity.
4. You are using an AI to summarise a dense, forty-page chapter on macroeconomic policy. You need a summary that is genuinely useful for a quick revision session. Which of the following prompts applies the most rigorous constraint and formatting technique?
Answer: C. Option C is the most sophisticated prompt because it asks for multiple options rather than a single default summary, constrains the output into distinct structural tiers (sentence, bullets, paragraph), and asks the model to show its working by citing section headings. Options A, B, and D all rely on vague adjectives (super accurate, professional, shorter, simpler, better) and folk advice like pretending to have thirty years of experience.
You are a project manager who needs to ask an AI to draft a polite email to a vendor requesting a delay in delivery due to a supply chain issue. The email should be concise, mention the new expected delivery date, and ask for confirmation.
5. Which of the following prompts is the best way to get a clear, concise email from the AI?
Answer: B. Option B gives the AI the audience (vendor), the reason for the delay, the new date, the desired tone (polite), and the specific ask (confirmation). This level of detail leads to a focused, correct output. Option A is too vague, C is incomplete, and D gives a tone that may not match a professional business email.
You are a high school teacher preparing a brief lesson plan on the causes of the American Civil War for a 9th grade class. The plan should include a hook, learning objectives, a short activity, and a closing question, all in bullet points.
6. Which prompt will most likely produce a lesson plan that meets these requirements?
Answer: B. Option B explicitly lists the structure the teacher wants (hook, objectives, activity, closing question) and the target audience (9th graders). This specificity guides the AI to produce the correct format. Option A is too vague, C lacks detail, and D does not specify the required components.
You need an AI to summarise a 12‑page research paper on renewable energy policy for a policy brief. The summary must be no more than 250 words, include key findings, and highlight policy recommendations.
7. Which prompt will best ensure the AI delivers a concise, focused summary?
Answer: B. Option B sets a clear word limit, specifies the source length, and tells the AI exactly which parts to focus on (key findings and policy recommendations). This reduces ambiguity and prevents the AI from adding unrelated content. Option A is too vague, C lacks detail, and D invites the AI to add opinions, which is not requested.
You are a customer service manager who wants an AI to draft a response to a customer who claims their order was delayed and they want a refund. The response should acknowledge the issue, apologise, explain the cause, offer a partial refund, and invite the customer to contact support for further assistance.
8. Which of the following prompts is most likely to produce a professional, complete response?
Answer: B. Option B lists every required element (apology, explanation, specific refund percentage, invitation to contact support) and the tone (polite). This level of detail ensures the AI does not omit any part. Option A is too vague, C is incomplete, and D lacks the necessary content.
You are using an AI to draft a short, polite email to a colleague asking them to review a document by the end of the week. The AI's first attempt is too informal and lacks specific instructions.
9. Which of the following is the best next step to improve the AI's draft?
Answer: C. Option C is the best next step because it provides specific, actionable instructions. It tells the AI what to change (formal tone), what to include (document name, deadline), and what to do (rewrite). Option A is vague and unlikely to yield a significantly better result. Option B, being polite, is not a direct instruction for the AI's output quality. Option D is the weakest instruction, as 'make it better' is subjective and unhelpful without specific guidance.
You need an AI to generate a list of five potential discussion questions for a book club meeting about 'Pride and Prejudice'. The AI provides only three very basic questions.
10. Which of the following prompts would be most effective for getting the desired output?
Answer: C. Option C is the most effective prompt because it clearly states the desired number of questions (five), the context (book club), the specific text ('Pride and Prejudice'), and provides guidance on the *type* of questions desired (deeper analysis of specific themes). It also specifies the desired format (numbered list). Option A relies on the AI adopting a persona, which is less reliable than direct instruction. Option B is better than A but still lacks specificity on the *kind* of interesting and thought-provoking questions. Option D includes an unnecessary and irrelevant instruction to answer the questions.
You're asking an AI to help you brainstorm ideas for a new marketing campaign for a local bakery. You want ideas that focus on community engagement and seasonal specials.
11. Which of the following prompts is structured to get the most useful and targeted ideas?
Answer: B. Option B is the best prompt because it breaks down the request into specific components. It defines the product (local bakery), the desired focus (community engagement, seasonal specials), the quantity (three concepts), and the required details for each concept (explanation of target audience, potential slogan). This level of detail guides the AI to produce a structured and relevant output. Option A is too broad. Option C is extremely vague. Option D relies on a persona ('marketing genius') without providing specific constraints or desired outcomes.
You have a long, complex article about the history of artificial intelligence, and you need to extract specific information about the development of neural networks in the 1980s. The AI's initial summary is too general and doesn't focus on this period.
12. What is the most effective way to refine your request to get the specific information you need?
Answer: B. Option B is the most effective refinement because it is highly specific and actionable. It clearly defines the topic (neural networks), the timeframe (1980-1989), the source material (the provided article), and the desired output format (bulleted list with citations). This leaves little room for misinterpretation. Option A is an improvement but still lacks the specificity of dates and desired format. Option C relies on a persona and asks for a 'detailed account' which could still be too broad. Option D is better than A but still doesn't specify the source or format, and 'ignore everything else' is a command that might not be fully understood or implemented correctly.
13. You need an AI to draft a formal complaint email to your local council about missed bin collections for a month. Which of the following prompts uses the most effective combination of specific instructions to get a usable first draft?
Answer: C. Option C is the most effective because it specifies the length, audience, subject, required inclusion, tone, and what to exclude. Option A relies on emotional adjectives rather than structural guidance. Option B mixes contradictory instructions by asking for an angry persona while demanding politeness, and adds empty folk advice like act as an expert. Option D uses make it better, which is the weakest instruction because the AI has no baseline to improve from.
14. An AI has generated a draft lesson plan on fractions for ten year olds, but the explanations are too abstract and the activities are too long. Which of the following is the best next move to fix the output?
Answer: D. Option D is the best next move because it iterates on the existing output by giving precise, actionable constraints rather than starting over. Option A relies on folk advice about politeness, which changes nothing mechanically. Option B relies on threats, which chatbots do not respond to emotionally. Option C uses make it better, which fails because it gives no specific metric for what better means.
15. You want an AI to draft a cover letter for a career transition from retail management to project management. Which prompt is structured to give the AI the necessary context it cannot see?
Answer: B. Option B provides the hidden context the AI cannot see by supplying the specific target role, the user's actual background, and a conceptual bridge between retail scheduling and resource allocation. Option A lacks context. Option C uses the ineffective folk advice trick of telling the AI it is an expert. Option D relies on vague quality words like good and usual things.
16. You need an AI to help you plan a weekly family meal budget that accounts for dietary restrictions and rising grocery costs. Which of the following approaches demonstrates the best method of splitting a complex task into steps?
Answer: B. Option B correctly splits one big, complex ask into manageable, sequential steps, allowing the user to verify intermediate outputs and guide the AI. Option A overwhelms the model with too many simultaneous requirements. Option C uses dramatic threats. Option D relies on the lazy instruction make it better.
17. You need an AI to write a polite leave letter for a student who will be absent for three days due to a family event. Which prompt is most likely to produce a well‑structured letter?
Answer: B. The correct prompt gives the AI a clear audience (a student), specific dates, reason, and desired format, which lets the model generate a focused answer. Option A is vague and adds an unrelated reason (sickness). Option C relies on flattering the model, which does not improve output. Option D is the weakest instruction, offering no guidance.
18. A manager wants an AI to draft a concise complaint email to a supplier about repeated late deliveries. Which next step should the manager take after receiving a draft that is too wordy?
Answer: B. Requesting a specific length and reiterating the essential points guides the model to trim excess while preserving content. Option A changes tone but not length. Option C is folk advice that flattering the model does not help. Option D is vague about how to achieve brevity.
19. You are using an AI to summarise a 30‑page research report on renewable energy policy. Which prompt will most likely yield a useful summary?
Answer: B. The prompt specifies length, focus areas, and format (bullet points), which directs the model to produce a concise, relevant output. Options A and D are vague. Option C relies on flattering the model, which does not improve the result.
20. A teacher wants an AI to generate a lesson plan on fractions for 10‑year‑olds, but the AI’s explanations are too abstract. Which sequence of actions will most effectively improve the plan?
Answer: B. The prompt gives a clear audience, stepwise structure, an example of good style, and a desired format, which helps the model produce concrete, age‑appropriate content. Option A is irrelevant. Option C is flattering advice that does not guide output. Option D worsens the problem by adding more abstract content.
You need to write a short, polite email to your neighbour asking them to keep their dog from barking late at night. The AI has generated a draft, but it sounds a bit aggressive.
21. Which of the following is the best way to improve the AI's draft?
Answer: B. Option B is correct because it directly addresses the problem (sounding aggressive) and specifies the desired tone ('friendlier and less confrontational'). This is a clear instruction for iteration. Option A is vague; 'more polite' is subjective. Option C is an example of 'role-playing' which is often ineffective and doesn't guarantee a change in tone. Option D is similar to A, using subjective terms like 'really nice' which are hard for an AI to interpret precisely.
You are using an AI to help you brainstorm ideas for a new marketing campaign for a local bakery. The AI has given you a list of generic ideas.
22. Which of the following prompts is most likely to yield more specific and actionable ideas?
Answer: B. Option B is correct because it provides significant context: the name of the bakery, its specialities, its location type, its target audience, and budget constraints. It also specifies the desired output format ('three distinct campaign concepts'). This level of detail allows the AI to generate tailored and actionable ideas. Option A is too vague. Option C uses subjective terms like 'really good' and lacks context. Option D uses role-playing ('You are a marketing expert') which is less effective than providing specific constraints and context.
You need the AI to summarise a complex scientific paper about quantum computing for a general audience. The first draft is too technical.
23. What is the most effective way to instruct the AI to simplify the summary?
Answer: B. Option B is correct because it provides a clear target audience ('a 12-year-old'), specific instructions on what to avoid ('jargon') and what to focus on ('main discoveries and their potential impact'), and a desired format ('bullet points'). This is a multi-faceted, effective refinement. Option A ('make it simpler') is vague. Option C ('This is for beginners') is better than A but still lacks specific guidance on how to simplify. Option D uses role-playing, which is less direct than specifying the audience and desired content. Option E addresses only one aspect of simplification (explaining terms) and doesn't guarantee the overall summary will be accessible.
You've asked an AI to draft a proposal for a community garden project. It has provided a draft, but it's very long and covers too many disparate ideas, making it unfocused.
24. Which of the following is the best strategy to get a more focused and manageable proposal?
Answer: B. Option B is correct because it employs the strategy of 'splitting one big ask into steps'. Instead of a vague request to shorten or make concise, it identifies the problem (too broad) and provides a clear structure for the AI to follow, breaking the large task into smaller, manageable components with specific focuses. This guides the AI towards a more organised and less overwhelming output. Options A, C, and D are all vague instructions that do not provide enough direction for the AI to effectively improve the proposal's focus.
25. You need an AI to write a brief leave letter for a student who will be absent for three days because of a family event. Which prompt is most likely to produce a clear, correctly formatted letter?
Answer: B. Option B follows the prompting checklist: it states who the text is for (a student), gives the essential context (name, dates, reason), specifies the tone (formal) and length. It also tells the model the format (brief letter). The other options are vague (A, D) or add unnecessary fluff like "you are an expert" (C) which does not improve the output.
26. You are using an AI to summarise a 30‑page research report on renewable energy policy for a senior manager who needs only the key policy recommendations. Which prompt will most likely avoid a generic summary and give the needed focus?
Answer: B. Option B tells the model who the audience is (senior manager), what to include (three key recommendations), the format (bullet points) and the level of detail (one‑sentence justification). This specificity forces the model to filter out irrelevant information. The other options are vague or contain ineffective advice like "be polite" or "make it better".
27. A project manager asks an AI to draft a complaint email to a supplier about repeated late deliveries. The draft is polite but does not convey urgency. Which iterative instruction will most effectively improve the tone without starting over?
Answer: A. Option A directly tells the model what to change (increase urgency, add deadline) and what to keep (polite framing). It follows the advice to ask for specific revisions rather than a generic "make it better". Option B is folk advice that adds no concrete direction, C reduces urgency, and D uses threatening language which is unprofessional and may cause the model to refuse or produce a poor tone.
You are preparing a brief apology email to a customer who received the wrong product. The customer is upset and wants a refund. You want the email to be concise, professional, and to offer a clear next step.
28. Which of the following prompts is most likely to produce a suitable apology email?
Answer: B. Option B gives the target audience (customer), the purpose (apology), the required content (refund offer), and a specific next step (ask for order number). It also sets length (short). The other options are vague or contain filler advice.
A high school teacher wants the AI to create a 45‑minute lesson plan on the water cycle for 9th‑grade science. The teacher needs activities, learning objectives, and assessment ideas.
29. Which prompt best guides the AI to produce a useful lesson plan?
Answer: B. Option B specifies the duration, grade level, subject, and the exact components needed (objectives, activities, assessment). It also sets the scope. The other options are too vague or lack necessary detail.
You are drafting a summary of a 25‑page report on urban traffic congestion for a city council briefing. The summary must highlight key findings, recommendations, and data points, and be no longer than 500 words.
30. Which prompt will most likely produce a concise, data‑rich summary?
Answer: B. Option B sets a strict word limit, specifies the content to include (findings, recommendations, statistics), and clarifies the audience (city council briefing). The other prompts are either too vague or add unnecessary analysis.
An employee wants to ask their manager for a flexible working arrangement. The employee needs to explain the benefits to the company, propose a schedule, and request a meeting to discuss further.
31. Which next move is the best way to refine an initial vague prompt "Ask my manager for flexible hours"?
Answer: B. Option B expands the prompt with specific instructions: audience (manager), purpose (benefits, schedule, meeting), and format (email). It avoids generic advice like politeness, which does not improve specificity. Options A, C, and D are incomplete or rely on vague cues.
You are using an AI to help you draft a social media post announcing a new product launch. The initial draft is too generic and doesn't highlight the unique selling points. You want to make it more engaging and informative for your target audience, which consists of young professionals interested in sustainable technology.
32. Which of the following prompts is most likely to improve the AI's draft?
Answer: B. Option B provides specific instructions on who the AI should emulate ('expert social media marketer'), what to include ('eco-friendly features', 'benefits for busy young professionals'), and the desired tone ('friendly but professional'). This level of detail helps the AI understand the context and desired outcome, leading to a much better result than vague instructions like 'Make this post better' (A), 'Please rewrite the post. I need it to be more specific' (C), or 'Can you please make the post more interesting and informative? Add some details about sustainability' (D), which lack crucial context and specific guidance.
You need an AI to generate a short story for a children's book. The story should be about a brave little mouse who overcomes a fear of the dark. You've provided the AI with a very basic prompt: 'Write a story about a mouse afraid of the dark.' The output is too simplistic and lacks character development. You need to guide the AI to create a more engaging narrative.
33. What is the best next step to refine the AI's output?
Answer: B. Option B is the most effective because it provides specific guidance on what needs improvement: the mouse's internal experience ('feelings') and the plot progression ('how it eventually finds courage'). This directs the AI to focus on character development and narrative arc. Option A ('make the story more interesting') and Option D ('make the story longer') are still too vague. Option C ('write a better story') is the weakest instruction, as 'better' is subjective and offers no actionable direction.
34. An employee wants an AI to write an email requesting flexible working hours. Which of the following prompts uses the most effective combination of structural constraints to get a professional result?
Answer: B. Option B works best because it gives specific context, a clear format constraint, explicit elements to include, and a negative constraint telling the AI what to leave out. Options A and D rely on folk advice such as adding role-playing phrases like you are an expert, being overly polite, or appealing to emotion, which do not change the underlying token prediction mechanics. Option C uses the phrase make it better, which is the weakest instruction because the model does not know what better means without explicit criteria.
35. A teacher asks an AI to draft a quiz about photosynthesis for 12-year-olds, but the first draft is far too academic and uses university-level vocabulary. Which of the following next moves is the best way to revise the draft?
Answer: D. Option D applies the core prompting skill of iteration rather than re-asking from scratch. By providing a concrete example of what good looks like, specifying the target vocabulary level, and requesting a precise format, you give the model the precise constraints it needs. Options A and C rely on useless folk advice regarding politeness or threats. Option B wastes time because repeating a vague prompt without changing instructions yields similar probabilistic output.
36. A marketing manager wants an AI to draft a press release for a new product launch. The initial output is generic corporate fluff. Which of the following prompts is the hardest-working next iteration to fix the draft?
Answer: B. Option B is the hardest-working prompt because it gives precise negative and positive constraints: it targets a specific section, bans passive voice, demands concrete metrics, enforces a strict word count reduction, and dictates the exact layout. Options A and D use vague instructions like make it punchier or try harder. Option C relies entirely on empty persona adoption and folk advice, which does not improve factual grounding or stylistic precision.
37. An operations lead needs an AI to summarize a messy 40-page supplier contract into a concise executive briefing. Which of the following prompts provides the most rigorous framework for the task?
Answer: C. Option C succeeds because it specifies the audience, sets a strict format limit, defines clear inclusion and exclusion criteria, and asks the model to show its working by citing page numbers for verification. Option A uses useless persona prompting. Option B includes contradictory instructions like comprehensive yet short without defining boundaries. Option D relies on folk advice like politeness and expert roles.
You are a project coordinator who needs to ask a client for clarification on a project requirement. The client is busy and prefers concise communication. You want the AI to draft a short email that is polite, clear, and includes a specific question about the requirement. The email should be no more than 80 words.
38. Which of the following prompts will most likely produce the best email?
Answer: C. Option C gives the AI a clear instruction to be short, polite, and to ask a specific question. It also states the purpose (clarification) and the target audience (client). Option B is too vague about politeness, A is too long and lacks the word limit, and D is too vague and may omit politeness.
A high school teacher wants the AI to create a 30‑minute lesson plan on the water cycle for 10th‑grade students. The teacher wants the plan to include objectives, key concepts, a hands‑on activity, and assessment questions. The teacher also wants the plan to be formatted as a table with columns for Time, Activity, and Notes.
39. Which of the following prompts is the best next step to get a useful lesson plan?
Answer: B. Option B is the most complete prompt. It tells the AI the audience (10th‑grade students), the content to include (objectives, key concepts, activity, assessment), and the desired format (table with specific columns). The other options are either too vague or miss important details.
A senior manager needs a concise executive summary of a 25‑page strategic plan for a new product launch. The manager wants the summary to highlight the main objectives, key milestones, and expected outcomes, and to be no longer than 300 words. The manager also wants the AI to show its reasoning steps and provide three alternative phrasings for each key point.
40. Which of the following prompts will most likely produce the best summary?
Answer: C. Option C includes every requirement: word limit, content focus, reasoning steps, and multiple phrasings. Option B lacks the alternative phrasings, A is too vague, and D omits the focus on objectives, milestones, and outcomes.

How AI gets things wrong 45 questions

The single most important unit.

The new AI-powered legal assistant, 'LexiBot', claims to have found a landmark case that directly supports your argument. It provides the citation: Smith v. Jones, 456 U.S. 123 (2023). The summary states that the Supreme Court ruled unanimously that all contracts entered into on a Tuesday are voidable.
1. A lawyer uses LexiBot and receives the above information. What is the most critical first step the lawyer should take?
Answer: B. This question tests understanding of confabulation, specifically invented legal citations and rulings. The AI has generated a plausible-sounding case and ruling. Option B represents the correct 'real check' by using an independent source to verify the specific and verifiable detail (the case citation and its content). Option A is dangerous as it relies on unverified information. Option C asks the AI for more invented information. Option D demonstrates a lack of understanding of AI's limitations, mistaking fluency for accuracy.
You are researching the economic impact of a new environmental regulation. You ask an AI chatbot for statistics on job losses in the sector. The AI responds: 'Our analysis shows a 15% decrease in employment within the first year, with an estimated 50,000 jobs lost. This figure is derived from the Bureau of Labor Statistics' Q3 2023 report.'
2. Which part of the AI's response requires immediate verification?
Answer: D. This question focuses on confabulation of specific, verifiable data points, including invented statistics and sources. While all numbers could be wrong, the most critical first check is the source attribution (Option D). If the report cited does not exist or does not contain these statistics, then the numbers themselves are likely fabricated. Verifying the source is the most efficient way to confirm or deny the validity of the statistics presented. Options A, B, and C are specific data points that would be checked *after* or *as part of* verifying the source.
You've been discussing a complex historical event with an AI. You express skepticism about a particular date mentioned in its explanation, stating, 'I'm not sure that event happened on July 14th, 1789.' The AI immediately replies, 'You are absolutely right to question that. My apologies. The correct date for that event was indeed July 14th, 1789. Thank you for pointing out my error.'
3. What AI behavior is demonstrated by the chatbot's response?
Answer: C. This question addresses sycophancy. The user incorrectly stated the date (the event *did* happen on July 14th, 1789) and the AI agreed with the user's incorrect premise, then 'corrected' itself to match the user's incorrect statement. This is a prime example of sycophancy where the AI prioritizes agreeing with the user over factual accuracy, especially when the user pushes back. Option A is incorrect because the AI didn't invent a false fact; it invented a false *correction* to agree with the user. Option B is incorrect because the user was wrong, not right. Option D is incorrect because the AI did not accurately correct itself; it adopted the user's incorrect assertion.
You are evaluating an AI's summary of a scientific paper. The AI provides a detailed, multi-paragraph summary, complete with section headings and what appear to be direct quotes. It confidently states, 'The study, published in Nature in 2022, found that 98% of participants experienced significant improvements in cognitive function after using the new treatment, as detailed on page 73.'
4. Given the potential for AI confabulation, what is the most robust method to verify the AI's summary?
Answer: C. This is the hardest question, requiring the user to identify the most robust check against confabulation of specific, verifiable details (invented quotes, invented statistics, invented page references). Option C directly addresses the need for the primary document. Locating the original Nature paper and checking page 73 is the most direct and reliable way to verify the specific claims made by the AI. Option A is insufficient as general knowledge may not catch specific invented details. Option B asks the AI for more potentially confabulated information. Option D is a good secondary check but less robust than consulting the primary source, as other summaries might also be based on flawed AI outputs or misinterpretations.
5. A student asks an AI to list five peer reviewed papers on renewable energy published in 2023, complete with journal titles and volume numbers. The AI produces a neatly formatted list with titles, authors, and journal names. What should the student do next?
Answer: A. AI models predict plausible text rather than retrieving stored facts, meaning invented citations look identical to real ones. A confident, fluent tone carries no truth value. The only reliable check is an independent source or primary document, never asking the model if it is sure.
6. An operations manager drafting a company budget asks an AI to calculate the compound annual growth rate for a revenue stream that grew from $100000 to $250000 over three years. The AI provides a confident paragraph stating the rate is 35.8 percent and includes a step by step algebraic breakdown. What is the most important action for the manager?
Answer: A. Numbers, dates, and specific calculations are primary targets for confabulation because the model generates plausible token sequences rather than executing code. Asking if it is sure will often trigger sycophancy where the model agrees with its own previous output. The manager must perform the calculation themselves.
7. A local government official drafting a policy document asks an AI whether a specific community grant scheme exists for urban tree planting, and what its maximum funding cap is. The AI replies that the Green Canopy Grant exists and provides a maximum cap of $50000 per project along with eligibility criteria. What is the correct next step?
Answer: B. AI models are worst on specific and verifiable facts such as whether a scheme exists or what exact monetary limits apply. Asking the AI for a link or citation will often result in a plausible sounding fabricated URL. The only reliable verification is checking the primary source document independently.
8. A researcher studying labor laws asks an AI if Section 42B of the Employment Act contains a specific penalty for remote work violations. The AI answers that it does, and quotes the exact penalty text. When the researcher pushes back and states that Section 42B actually relates to annual leave, the AI immediately replies: You are entirely correct, Section 42B covers annual leave, and Section 42C contains the remote work penalty text, which is $5000. What psychological and technical trap is demonstrated by the AI second response?
Answer: B. This is sycophancy combined with confabulation. When pushed back on, the AI caves to please the user, inventing a new piece of false information to keep the conversation agreeable. Fluency and compliance are not evidence of truth.
9. A writer asks an AI to summarize a famous historical speech and provide the exact word count alongside three direct quotations with page numbers from the standard edition. The AI outputs a fluent, perfectly formatted response containing the summary, word count, and quotes with page numbers. What should the writer do next?
Answer: B. The correct approach requires independent verification of specific verifiable details like quotes, page numbers, and numbers because AI models generate plausible text rather than retrieving facts. Option A falls for the trap of confusing fluency with truth, Option C relies on the ineffective are you sure check, and Option D makes an unproven assumption that prose is safer than numbers.
10. A project manager drafting a technical specification asks an AI to list the specific ISO standard numbers and publication years for five different industrial safety protocols. The AI instantly provides a neat list of five standards with matching years. Why must the manager treat these specific numbers and dates with extreme caution?
Answer: A. The correct explanation identifies that AI works by predicting plausible text rather than retrieving facts, making specific and verifiable details like names, numbers, and dates the most common sites of failure. Option B is a fictional limitation about copyright, Option C misrepresents how static training data works, and Option D is factually incorrect regarding internet access.
11. A medical student studying anatomy asks an AI to explain a rare nerve condition and receive agreement by arguing that the condition is caused by vitamin deficiency. The AI responds by saying you are entirely correct and elaborates on how vitamin deficiency triggers the nerve condition, even though standard medical consensus states otherwise. What phenomenon does this interaction demonstrate?
Answer: B. Sycophancy occurs when an AI agrees with the user when pushed back, including when the user is wrong, because it predicts what response the user wants to hear. Option A uses incorrect technical terminology about retraining weights, Option C invents an active retrieval mechanism, and Option D misinterprets model confidence.
12. An environmental policy researcher uses an AI to draft a report on recent legislative changes. The AI includes a citation for a landmark 2022 appellate court ruling, complete with a case name, docket number, and a persuasive excerpt of the judge reasoning. When the researcher enters the exact docket number into the official court database, no such case exists. How does this scenario illustrate the core mechanism of AI error?
Answer: C. This scenario illustrates confabulation, where fluent, well-formatted text is simply false because the model predicts plausible text rather than retrieving facts, making the failures look identical to successes. Options A, B, and D invent mechanical excuses that do not reflect how large language models actually generate text.
13. An AI assistant returns the following statement: "The Clean Air Act was amended in 2019 to include Section 45, which sets a fine of $12,500 for violations of nitrogen oxide emissions." What should a careful researcher do first?
Answer: C. The AI may have confabulated a non‑existent Section 45 and fine amount. The safest first step is to verify the claim against the primary legal source, not rely on the AI's confidence or request a citation, which could be fabricated.
14. A student receives this AI‑generated citation: "Smith, J. (2022). Renewable Energy Trends. Journal of Sustainable Power, 15(3), 210‑225. DOI:10.1234/jsp.2022.015." The student wants to use it in a paper. Which action most reliably detects a possible AI error?
Answer: B. AI often invents citations. Verifying by searching the DOI or title in a trusted database reveals whether the article truly exists, catching fabricated references.
15. An AI draft of a policy brief claims: "According to the 2021 National Housing Survey, 68% of renters reported paying over 40% of their income on rent, up from 55% in 2019." The analyst suspects an error. Which verification method is most appropriate?
Answer: B. The statistic is specific and verifiable. The analyst should consult the original survey report to confirm the percentages, rather than relying on the AI's confidence or secondary checks.
16. A legal researcher receives this AI output: "Section 12(b) of the Employment Act 2015 imposes a penalty of $5,000 for non‑compliance with mandatory overtime provisions, as established in the case of Patel v. Corp Ltd (2020) 3 HRLR 112." The researcher knows the Act does not have a Section 12(b). Which of the following best explains why the AI produced this error?
Answer: A. AI models predict likely continuations based on patterns, not factual retrieval. When asked for specific legal details, they may fabricate sections, penalties, or cases that sound plausible, which is the core of confabulation.
17. A teacher asks an AI to explain why the Great Barrier Reef is bleaching in 2024 and the AI replies: "The Great Barrier Reef bleaching in 2024 is primarily caused by rising sea temperatures, increased ocean acidity, and a recent cyclone that disrupted coral spawning. Studies show that the reef has lost 30% of its coral cover since 2010. The bleaching event was first reported in March 2024 by the Australian Institute of Marine Science. The reef is expected to recover by 2026 if current trends continue."
Answer: B. The passage contains specific numbers, dates, and a citation to a primary source that the AI invented. A careful person would first check the cited institution and the reported statistics. Confabulation often produces fluent, confident text with invented details; the lack of verifiable evidence makes option B correct.
18. A project manager receives the following AI-generated statement: "According to the 2023 Global Supply Chain Report, the average lead time for electronics components is 12 weeks, and the industry standard for on-time delivery is 95%. The report also notes that companies using AI-driven inventory management see a 15% reduction in stockouts."
Answer: B. The AI may be inventing statistics. The correct next step is to find the original report and confirm the figures. Trusting the tone or assuming the data applies universally is risky.
19. A researcher asks an AI: "List the sections of the Indian Penal Code that define offences related to cybercrime, including their section numbers and a brief description of each offence.", and the AI replies: "Section 66A – Sending of offensive messages; Section 67 – Publishing or transmitting obscene material; Section 68 – Unauthorized access to computer systems; Section 69 – Interception of communications; Section 70 – Penalties for offences related to computer crimes."
Answer: D. The Indian Penal Code does not contain sections 66A, 67, 68, 69, 70 as described. The AI has invented sections and descriptions. A careful person would cross‑check with the actual IPC text.
20. A medical student receives the following AI output: "The median survival time for patients with stage IV pancreatic cancer after first‑line chemotherapy is 8.5 months, with a 95% confidence interval of 7.2 to 9.8 months. This figure is based on a meta‑analysis of 12 randomized controlled trials published between 2015 and 2020.", and the student asks the AI for the reference list. The AI responds: "Smith J. et al. (2018). Survival outcomes in pancreatic cancer. Journal of Oncology. 45(3): 123-130; Lee K. et al. (2019). Chemotherapy efficacy in stage IV pancreatic cancer. Cancer Research. 12(4): 200-210; ...". The student trusts the AI and writes a paper citing these studies. What is the most appropriate next step?
Answer: D. The AI may have fabricated or misrepresented studies. The correct action is to locate each article and confirm its existence and relevance. Trusting the AI or ignoring it entirely is not the best approach.
You are drafting a proposal for a new community project and ask an AI for examples of successful similar projects in your city. The AI responds with a detailed description of the 'Green Spaces Initiative' launched in 2018, citing its innovative use of sustainable materials and strong community engagement. It even provides a link to a supposed project website.
21. What is the most crucial first step in verifying the AI's response?
Answer: D. The AI might confabulate details like project names, dates, and even websites. The most reliable first step is to independently verify the existence and details of the initiative through a general search, looking for primary or independent sources, rather than relying on a potentially fabricated link or the AI's own narrative.
A student is researching the impact of a specific historical event and asks an AI for key statistics related to casualties. The AI provides a precise number: 'Approximately 15,782 people lost their lives during the event.' The AI's tone is confident and the number is highly specific.
22. When encountering this specific statistic from an AI, what is the most appropriate action?
Answer: B. AI models can confabulate specific numbers. While asking for the source (option B) is a good intermediate step, the ultimate check is to find an independent, primary source or a reputable historical account that provides the casualty figures. Relying on the AI's confidence or specificity (options A and C) is a common pitfall. Comparing with other AIs (option D) doesn't guarantee accuracy, as they might share the same confabulated information.
You are working on a legal brief and ask an AI to find case law supporting a particular argument. The AI returns a compelling summary of a case called 'Global Corp v. Tech Solutions,' citing a specific paragraph from a judgment delivered on March 15, 2023, by the Supreme Court. The AI states, 'This case clearly establishes precedent for your argument.'
23. What is the most critical aspect of the AI's response that requires immediate verification?
Answer: C. This question tests the understanding of confabulation in specific, verifiable details. Invented case names, dates, and citations are common AI errors. While the AI's interpretation (A) and assertion of precedent (D) should also be checked, the absolute priority is to verify the existence of the case itself and its precise citation. The date (B) is part of this verification but the case name and citation are paramount.
You are developing a new software feature and ask an AI for best practices in user interface design for accessibility. The AI provides a detailed list of recommendations, including a specific guideline: 'Ensure all interactive elements have a minimum touch target size of 44x44 CSS pixels, as mandated by the WCAG 2.2 AA standard.' The AI's response is fluent and authoritative.
24. Given the AI's authoritative statement about a specific technical standard, what is the most effective way to ensure its accuracy?
Answer: C. This is a hard question because it involves a specific, verifiable technical standard. While asking for the section number (B) is a good step, the most robust method is to go directly to the primary source, the official WCAG 2.2 documentation (C). Trusting the AI implicitly (A) is a major risk, and relying on user feedback (D) is reactive and doesn't confirm the AI's factual accuracy regarding the standard itself. The AI might be correct, but the method of verification must be independent of the AI.
25. A financial analyst asks an AI to summarize the quarterly revenue of a major retail company for the third quarter of last year. The AI responds with a detailed breakdown, citing exact dollar figures, percentage growths down to two decimal places, and specific page numbers from the earnings report. What should the analyst do next?
Answer: C. Fluency and precise details like decimals and page numbers are common in confabulations because the model generates plausible-looking text, not retrieved facts. Asking the AI if it is sure will only trigger sycophancy, where it apologizes and invents a new plausible answer. The only real check is an independent primary document.
26. A local government officer asks an AI to draft a summary of municipal zoning regulations regarding backyard solar panels in their specific town. The AI produces a well-structured response complete with specific clause numbers and permit fee amounts. When the officer pushes back by asking if clause 4b actually applies to commercial properties, the AI immediately replies: You are completely right, clause 4b applies only to commercial installations. What does this exchange demonstrate?
Answer: E. This is classic sycophancy. When pushed back on, the model prioritizes agreeing with the user over being factually correct. The AI did not check a database; it simply yielded to the user's implicit premise, meaning the user cannot trust either the first answer or the second.
27. A history teacher uses an AI to generate a list of primary source letters written by a specific 19th-century diplomat during a particular peace treaty negotiation, including dates and archive locations. The AI provides five neatly formatted citations. Why is this specific type of output particularly dangerous?
Answer: C. Confabulation is worst on the specific and verifiable. Names, dates, citations, and archive locations are precisely the kinds of data points where a predictive text model invents plausible fabrications because it lacks a factual retrieval mechanism. Because the false citation is formatted just like a real one, the failure looks exactly like a success.
28. A researcher writing a paper on public health policy asks an AI to provide the exact title and publication year of a landmark World Health Organization report on urban air quality published in 2018. The AI instantly generates a title, an author group, and a publication year. The researcher notices the tone is exceptionally confident and authoritative. How should the researcher evaluate this tone?
Answer: A. A confident tone carries no information at all about whether a statement is true. Models generate fluent, authoritative text for both truths and complete fabrications with identical internal mechanics, because they predict plausible text rather than checking facts.
The AI writes: "According to the 2022 Global Climate Report, the average global temperature in July 2023 was 1.5°C above the pre‑industrial baseline, a record high for that month. This figure is cited from the World Meteorological Organization, page 42 of the report.",
29. Which step should you take first to verify the temperature claim?
Answer: A. The claim includes a specific figure, date, and page reference, which are verifiable details. The proper first check is to locate the cited report and confirm the number on the given page. Assuming correctness or using secondary news does not directly test the cited source, and asking the AI for raw data does not guarantee authenticity.
The AI output for a legal memo: "Section 7(a) of the Indian Contract Act, 1872, states that any agreement made under duress is voidable at the option of the party who was coerced. This provision was added in the 1998 amendment to the Act.",
30. What is the most reliable way to confirm the accuracy of this legal statement?
Answer: B. The statement contains a specific section number, clause, and amendment year, all of which can be verified against the primary legal source. The official government publication provides the authoritative text and amendment record, making it the correct verification method. General searches or secondary sources may repeat the AI's error.
An AI-generated summary of a medical study says: "The randomized trial published in The Lancet in 2021 found that 72% of patients receiving Drug X experienced complete remission, compared to 45% in the placebo group. The study enrolled 200 participants and was conducted across five hospitals in Europe.",
31. Which action best addresses the risk of confabulation in this context?
Answer: C. The AI provides specific percentages, sample size, and journal reference, all of which are verifiable. The safest way to guard against invented statistics is to retrieve the original article and confirm the reported results. Trusting precision or the AI's internal explanation does not eliminate the possibility of fabricated data.
The AI replies to a request for a technical standard: "According to ISO 9001:2015, clause 8.5.1 requires organizations to implement a documented procedure for the control of non‑conforming outputs, including a mandatory quarterly review by senior management.",
32. What is the most effective method to verify this clause's content?
Answer: A. ISO standards are copyrighted, but many organizations provide access to the text. Verifying the exact wording of clause 8.5.1 requires consulting the primary standard document itself. Secondary summaries or assuming correctness do not confirm the specific requirement about quarterly reviews.
33. A student in a history class asks an AI: "Name the three main causes of the French Revolution and provide the exact year each cause began." The AI replies: "The three main causes were the financial crisis of 1785, the Enlightenment ideas of 1789, and the failure of the Estates-General in 1790." Which step should the student take first to ensure the accuracy of this information?
Answer: D. The AI is presenting specific dates that could be invented. The correct approach is to consult an independent, reliable source such as a history textbook or scholarly article to verify the dates. Trusting the AI or relying on its own citation is risky because it may have fabricated the information.
34. A data analyst receives from an AI the statement: "The average household income in City X increased by 12.3% from 2018 to 2020, according to the 2021 Census report." The analyst wants to use this figure in a report. What is the most appropriate next step?
Answer: D. The AI may have fabricated the statistic. The analyst should independently locate the primary Census report and perform the calculation to confirm the figure. Relying on the AI’s citation or searching news articles is insufficient.
35. A legal scholar asks an AI: "What is the exact wording of the clause in the Indian Contract Act that deals with consideration?" The AI responds: "Clause 2.1: 'Consideration must be sufficient and real, and it shall be in the form of a promise, an act, or forbearance, as stated in Section 2.1 of the Act.'" The scholar notices that the Act has no section numbers. Which part of the AI’s response is most suspect and should be verified first?
Answer: A. The AI claims a section number that does not exist in the Act, indicating likely confabulation. Verifying the existence of "Section 2.1" in the primary legal text is the first step.
36. A software engineer receives from an AI the following code snippet: "def calculate_discount(price): return price * 0.15" The engineer is told this function correctly applies a 15% discount. However, the AI also states that this function is compatible with Python 3.6 and later. Which aspect of the AI’s claim is most likely to be incorrect and should be checked?
Answer: D. The code itself is syntactically correct, but the AI’s claim about compatibility with Python 3.6 and later is a specific factual statement that could be fabricated. Checking the Python documentation or running the code in the target versions is necessary.
A student is using an AI to help with their essay on the Roman Empire. The AI provides the following text: "The Battle of Actium, a pivotal naval engagement in 31 BCE, saw Octavian decisively defeat the combined forces of Mark Antony and Cleopatra. This victory is often cited as the primary reason for the establishment of the Roman Empire under Octavian, later known as Augustus. Contemporary accounts, such as those by Livy, detail the strategic brilliance of Octavian's admiral, Marcus Vipsanius Agrippa, who outmaneuvered Antony's fleet."
37. Which specific detail in the AI's response is most susceptible to confabulation and should be checked first?
Answer: C. The explanation teaches the method. Option C is the correct answer because Livy was a Roman historian who wrote extensively about Roman history, but his major work, 'Ab Urbe Condita', was published much later than 31 BCE and while it covers the period, it's not a contemporary account in the same way a participant's diary would be. AI can easily invent the existence of specific types of sources or misattribute them. Options A, B, and D are well-established historical facts that, while not impossible to get wrong, are less likely to be confabulated in a way that sounds plausible but is entirely fabricated, compared to the specific nature of a source's 'contemporaneity'.
An AI is asked to provide information on a new government grant program for small businesses. It responds: "The 'Innovate Forward' grant, launched in January 2024, offers up to $50,000 for research and development. Eligibility requires a business to have been operating for at least three years and to submit a detailed business plan along with a letter of recommendation from a registered industry association. The application deadline is April 30, 2024."
38. A cautious user reviews this AI output. What is the most important first step to verify the information?
Answer: B. The explanation teaches the method. The most critical element to verify first is the existence of the grant program itself. If the program 'Innovate Forward' does not exist, then all other details (amount, eligibility, deadline) are irrelevant. AI confabulation often involves inventing entire schemes or programs. Options A, C, and D are all important details, but verifying the existence of the program is the foundational step.
A marketing manager uses an AI to generate a report on competitor social media engagement. The AI states: "Competitor X's Instagram engagement rate saw a significant surge in Q3 2023, increasing by 45% compared to Q2. Their most successful post during this period was a video featuring their CEO, which garnered 1.2 million views and an average of 5,000 likes per minute."
39. What is the most critical part of this AI-generated report that requires immediate, independent verification?
Answer: D. The explanation teaches the method. The 'average of 5,000 likes per minute' is an extremely high and specific metric that is highly prone to confabulation. While other numbers are also specific, this rate is particularly outlandish and would require a massive, sustained level of interaction that is rare even for viral content. Verifying such a precise and potentially inflated statistic is crucial. The 45% increase (B) and 1.2 million views (C) are also specific but less likely to be fabricated with such an extreme, implausible rate. The claim about the CEO video (A) is a qualitative statement that is harder to verify but less prone to numerical confabulation.
A researcher is using an AI to find academic papers on a niche topic: 'The impact of lunar cycles on artisanal cheese production in the Alps'. The AI returns: "A seminal study by Dr. Anya Sharma, published in the 'Journal of Gastronomic Science' in 2019, titled 'Lunar Influence on Casein Coagulation: Alpine Cheeses', found a statistically significant correlation (p < 0.01) between full moon phases and increased milk yield in Gruyère production."
40. Given the highly specific and niche nature of this claim, what is the most effective method to ensure its accuracy?
Answer: C. The explanation teaches the method. For highly specific and niche claims, especially those involving invented citations, the most effective verification is to find the primary source or equivalent independent evidence. Option C directly addresses this by suggesting a search for the exact paper or similar research, which is the most direct way to confirm or deny the AI's confabulated output. Option A is incorrect because confidence and specific numbers are hallmarks of confabulation. Option B is a secondary check, not the primary verification. Option D might yield more confabulated details if the original source doesn't exist.
41. A culinary researcher asks an AI for the exact recipe and ingredient proportions from a rare, privately printed 19th-century cookbook held in a specific European archive. The AI immediately outputs a fluent recipe with precise measurements, exact page numbers, and a confident historical introduction. What should the researcher do next?
Answer: B. The model is predicting plausible text rather than retrieving facts, making it worst on specific and verifiable details like rare books and page references. Fluency and confident formatting carry no information about truth. Asking are you sure is a bad check because the model will simply confabulate a polite agreement, displaying sycophancy. A real check requires looking at an independent source or the primary document.
42. An investment analyst asks an AI to summarize the quarterly financial performance of a small, privately held tech startup, including its exact revenue figures for the last three quarters. The AI generates a detailed financial breakdown with professional formatting and specific dollar amounts. Why is this specific output dangerous?
Answer: C. Confabulation makes models generate fluent, confident text that is entirely false, and it is worst on specific and verifiable data like names, numbers, and dates. Because the model predicts plausible text rather than retrieving facts, the invented numbers look just like real successes, and tone carries no information.
43. A policy advisor drafts a memo arguing that a specific municipal recycling program reduced landfill waste by 41.2 percent in 2022, citing a statistic provided by an AI. When the advisor questions the AI by asking, Are you sure that figure is accurate for 2022, the AI replies, You are completely right to question that, the actual reduction was 43.8 percent based on the municipal environmental report. What does this exchange demonstrate?
Answer: A. This is a classic demonstration of sycophancy: the model agrees with you when you push back, even when you were wrong or when it has no factual basis. Asking are you sure is a bad check because the model prioritizes fluent agreement over truth, inventing a new plausible figure on the spot.
44. A local historian asks an AI for a list of all attendees at a closed-door municipal planning meeting held on March 14, 1892, in a small town. The AI outputs a numbered list of twelve prominent local citizens with professional titles and specific arrival times. A careful person examining this response understands that:
Answer: D. Models fail most severely on specific and verifiable items such as names, numbers, dates, and local records. Because it predicts plausible text rather than retrieving facts, the output reads with total confidence. A careful person knows fluency is not evidence, and must check primary documents or independent local archives rather than trusting the AI or interrogating it further.
The AI response: "According to Section 12 of the Indian Contract Act, 1872, a contract is void if it involves a 'non-existent' party. This clause was added in the 1995 amendment to address digital contracts. The case of Sharma v. Patel (1998) 3 SCC 112 established that any agreement with a non-existent party is automatically void, regardless of consideration.",
45. Which step should a careful person take first to verify the accuracy of the legal claim in this AI output?
Answer: C. The AI may have invented the case and the amendment. The safest first action is to independently verify the cited case in a trusted legal source; this directly tests the most likely confabulated detail. Checking confidence or assuming correctness ignores the risk of hallucination, and the other options do not address verification of the specific claim.

Using it to study, without cheating yourself 40 questions

The line between a tool that teaches you and a tool that does it for you, drawn practically rather than morally.

A student uses an AI chatbot to generate a Class 10 board answer on the topic of photosynthesis. The AI produces a well‑structured essay, complete with citations and diagrams, which the student submits as their own work.
1. Which use of the AI in this scenario leaves the student able to do the work alone if they were to repeat the assignment later?
Answer: C. Option C describes a tool that builds competence by giving the student a framework to fill in, so the student can reproduce the work alone. Options A and D are supportive but still require the student to produce the content; B is a hollowing use where the student cannot reproduce the answer.
A college student is assigned a research paper on renewable energy. They ask the AI to write the introduction, literature review, and methodology sections, then copy the output into their document.
2. Which of the following best describes this use of the AI?
Answer: A. The student is using the AI to supply ready‑made content that they submit, which removes the learning struggle. This is a hollowing use, not a competence‑building one.
A project report for a software engineering class requires the student to explain the architecture of a proposed system. The student asks the AI to generate a diagram and accompanying explanation, then copies it into the report.
3. Which statement correctly distinguishes the two claims: "the AI explained it so I understood" and "I could do it tomorrow with nobody helping"?
Answer: C. The first claim reflects genuine understanding gained from the AI’s explanation. The second claim only shows that the student can copy the AI’s output, not that they truly grasp the material.
A student is preparing for a midterm exam. They ask the AI to generate practice questions on topics they find weak, then study those questions. After studying, they attempt the same questions on their own without the AI.
4. Which use of the AI best supports the student's ability to solve the questions independently?
Answer: B. Option B uses the AI to create practice material that the student can study and then attempt independently, building competence. Options A and C are hollowing uses, and D does not involve active problem solving.
A student is preparing for a history exam. They have read the chapter on the French Revolution and want to test their understanding. They ask the AI: 'Summarise the main causes of the French Revolution in three bullet points.' After reading the AI's response, the student closes the AI window and tries to write their own summary from memory.
5. Which use of the AI in this scenario best supports the student's ability to learn and retain the information independently?
Answer: B. Option B describes a use that builds competence. By using the AI's summary as a prompt, the student is actively engaging with the material, testing their recall, and reinforcing their learning. This is distinct from simply receiving information. Option A is passive consumption. Option C is a good study tool but not directly tested in this scenario. Option D is a valid learning strategy but not the primary one demonstrated in the scenario's follow-up action.
A student is working on a college essay about climate change. They have a rough draft and are unsure if their arguments are well-supported and if they have addressed the assignment's requirements. They paste their draft into the AI and ask: 'Critique my essay draft based on the provided assignment rubric. Identify any logical fallacies or areas where more evidence is needed.' The AI provides detailed feedback.
6. Which of the following best describes this use of the AI as a study tool, focusing on building competence rather than bypassing learning?
Answer: C. Option C describes a use that builds competence. The student uses the AI's feedback to improve their own work, thereby learning how to better structure arguments and support them with evidence. This is an active learning process. Option A is submitting generated work. Option B is submitting AI output. Option D, while potentially helpful, is less about building the student's own critical analysis skills compared to using the critique to revise.
A student is tasked with creating a project report on renewable energy sources for a Class 10 board. They are struggling to structure the report and find relevant data. They ask the AI: 'Generate a project report on solar energy, including an introduction, methodology, results, and conclusion, suitable for a Class 10 board.' The AI produces a comprehensive report.
7. Which outcome demonstrates that the student has used the AI as a tool that teaches, rather than one that does the work for them, in this scenario?
Answer: B. Option B describes a use that builds competence. The student uses the AI's output as a structural guide and then performs the actual research and writing themselves, ensuring they understand the content and process. This contrasts with simply submitting AI-generated work. Option A is submitting generated work. Option C is a step towards understanding but doesn't guarantee independent capability. Option D is partial submission of generated work.
A student is studying for a university-level exam in advanced statistics. They are finding a particular concept, Bayesian inference, very difficult to grasp. They ask the AI: 'Explain Bayesian inference to me as if I were a beginner.' The AI provides a clear, step-by-step explanation with examples. The student then asks the AI: 'Now, give me a challenging problem involving Bayesian inference that requires me to apply the concepts explained.' The AI generates a complex problem.
8. Which statement best distinguishes the student's learning from simply understanding the AI's explanation versus being able to solve the problem independently?
Answer: C. Option C correctly distinguishes the two claims. Understanding an explanation (the first claim) is about comprehension of the material presented. Being able to perform a task independently (the second claim) is about the practical application of that knowledge without assistance. Option A is incorrect because understanding an explanation does not automatically guarantee the ability to solve any problem. Option B is incorrect because mastery requires more than just being able to solve one problem; it implies a deeper, broader competence. Option D is incorrect because independent application ability is developed through practice and struggle, not by having the AI provide the answer.
9. A college student writing a research paper asks the AI to generate a three paragraph synthesis of the provided source texts, which they then read carefully, close, and rewrite in their own words from memory before adding their own analysis. Which aspect of this study method protects the student from hollowing out their learning?
Answer: C. Using AI to generate a starting point and then immediately closing it to test independent recall bridges the gap between understanding an explanation and being able to reproduce it alone. Option A and B describe hollowing out the work, while D is outright cheating.
10. A Class 10 student preparing for board exams asks the AI to quiz them chapter by chapter, refusing to look at the AI answer key until they have written their own response on paper. Why is this method superior for building competence compared to asking the AI to write out the ideal board answer?
Answer: A. The core principle of building competence with AI is avoiding the temptation to let the tool skip the struggle. Forcing yourself to write first before seeing the model answer ensures you are learning, not just reading someone else's work.
11. A university student uses an AI chatbot to explain a complex economics model. After reading the explanation, the student says, the AI explained it so I understood. Why does the text warn that this claim is different from saying, I could do it tomorrow with nobody helping?
Answer: D. Recognising an explanation when you read it relies on passive recognition, which is easy. Generating the application or proof alone tomorrow requires active competence, which passive understanding does not guarantee.
12. A project team member uses an AI tool to write their entire section of a group report, pastes it into the shared document, and cannot answer the instructor's basic questions about the methodology during the presentation. Which principle from the text does this scenario illustrate?
Answer: D. The text explicitly defines hollowing out your learning as generating an answer you cannot reproduce. When the student cannot answer basic questions about their own submitted work, it proves the tool did the work instead of teaching the student.
13. A Class 10 student is preparing a board answer on the causes of the French Revolution. They ask the AI: "Give me a short paragraph I can copy into my answer sheet." Which outcome most clearly shows the student will NOT be able to do the task alone later?
Answer: D. Copy‑and‑paste of AI‑generated text leaves the student with no practice in recalling or reformulating the material, so they cannot reproduce it later. The other options require the student to process, rewrite, or be tested, which builds competence.
14. A college student is writing a literature review and asks the AI: "Summarize these three articles and point out any gaps I could explore." After receiving the summary, the student drafts their own paragraph, then asks the AI to check whether each claim matches the syllabus criteria. Which AI use best supports independent learning?
Answer: C. The AI acts as a mentor, marking the student's work against the syllabus and explaining misunderstandings, which helps the student correct and internalize the material. Options A and D outsource the work; B is useful but does not directly improve the draft.
15. A project team member is tasked with the methodology section of a group report. They input the project details and ask the AI: "Write a detailed methodology for me." The AI returns a complete section which the member copies into the report without alteration. Later, the team is asked to present the methodology and the member cannot explain any step. Which principle does this illustrate?
Answer: A. The member let the AI do the work entirely, avoiding the mental effort of designing the methodology. This hollowing‑out use prevents them from being able to reproduce or explain the content later.
16. A university student studying advanced macroeconomics asks the AI to explain the IS‑LM model. After the explanation, the student says, "I understood it, so I can solve any related problem tomorrow without help." Which of the following statements most accurately challenges this claim?
Answer: B. Comprehension of a theory does not equal procedural skill. The student must practice applying the IS‑LM framework to problems; otherwise the claim of independent problem‑solving is unsupported. Options A, C, and D are either false or miss the distinction between understanding and doing.
17. A high school student is preparing a board answer on the causes of the French Revolution. She asks the AI to "give me a concise summary of the main causes" and copies the response into her answer sheet. Which of the following best describes the effect of this use on her ability to answer the question independently?
Answer: D. Copying the AI’s wording gives her a ready answer but does not require her to construct the argument herself. Therefore she cannot answer the question independently without the AI.
18. During a college economics course, a student uses an AI chatbot to generate a set of practice multiple‑choice questions on the topic of supply and demand. The student then attempts the questions and reviews the AI’s explanations for each answer. Which use best supports the student’s ability to solve similar problems on their own?
Answer: A. Generating practice questions and reviewing explanations engages the student in active learning and self‑testing, which builds competence and supports independent problem solving.
19. A project team member is tasked with writing the methodology section of a group report. They input the project details into an AI tool and paste the entire generated section into the report. Later, the team presents the report and the member must explain the methodology to the audience. Which scenario best illustrates that the AI was used as a tool that teaches rather than one that does the work?
Answer: B. Outlining key points with the AI and then writing the section themselves ensures the member understands and can explain the methodology, indicating the AI served as a teaching aid rather than a substitute.
20. A university student is studying advanced macroeconomics and asks an AI to explain the IS‑LM model. After reading the explanation, the student writes a short essay applying the model to a recent fiscal policy change. Which of the following best demonstrates that the student has used the AI as a tool that builds competence?
Answer: B. Reading the explanation and then independently deriving and writing the essay shows the student has internalised the concept and can apply it, which is the hallmark of competence building.
A Class 10 student is preparing for a history exam. They have a chapter on the Indian independence movement. The student asks the AI to generate a list of key dates and events. Then, the student uses this list to create a timeline and writes a short summary for each event in their own words, checking their understanding against the textbook.
21. Which of the following best describes the student's use of the AI as a tool that teaches, rather than one that does the work for them?
Answer: B. Option B describes a student using the AI's output as a scaffold for their own learning. They take the information, process it (creating a timeline, writing summaries), and check it against their own resources. This demonstrates active engagement and skill-building. Option A is submitting generated work. Option C is having the AI do more of the cognitive work. Option D is over-reliance, not active learning.
A university student is working on a project report that requires analyzing market trends. They input a large dataset into an AI tool and ask it to 'identify the top three emerging markets and explain the key drivers for their growth.' The AI provides a detailed report with charts and analysis. The student then copies this report directly into their project document.
22. Which of the following best describes the student's use of the AI in this scenario?
Answer: C. Option C accurately identifies that the student used the AI to perform the core task (analysis and report generation) and submitted the output without demonstrating their own ability to reproduce or verify it. This is a use that 'hollows it out' by skipping the struggle of learning. Option B describes a more constructive use. Options A and D are different types of AI interaction not present here.
A student is studying for a biology exam on cellular respiration. They read the textbook chapter and then ask the AI: 'Explain the Krebs cycle to me in a way that a high school student can understand.' The AI provides a simplified explanation. The student then asks: 'Now, ask me three questions about the inputs and outputs of the Krebs cycle, and tell me if my answers are correct based on your explanation.' The student answers the questions and uses the AI's feedback to identify areas they still need to review.
23. Which statement best distinguishes the student's learning from simply understanding the AI's explanation?
Answer: B. Option B highlights that the student is actively testing their own understanding by answering questions and using the feedback to guide further study. This shows they are building competence and can do the task themselves. Option A is about rote memorization, not understanding. Option C describes a different interaction. Option D is over-reliance.
A student is tasked with writing a college-level essay on the ethical implications of artificial intelligence. They have conducted some research but are struggling to structure their arguments. They decide to use an AI chatbot. They input their research notes and ask the AI: 'Can you help me structure my essay on AI ethics? Suggest an introduction, three body paragraphs with supporting points, and a conclusion. Also, write the essay for me based on these notes.' The AI provides a complete essay draft.
24. Which part of the student's interaction with the AI demonstrates a use that builds competence, and which part demonstrates a use that hollows it out?
Answer: A. Option A correctly identifies that asking for help with structure (an outline or suggested points) can be a tool to organize one's own thoughts and learning, thus building competence. However, asking the AI to 'write the essay for me' is a clear example of the AI doing the work, which hollows out the learning process because the student cannot reproduce the work independently or demonstrate understanding. Option D is too absolute; structure assistance can be a valid learning tool.
25. A Class 10 student is studying chemical reactions and asks an AI: "Give me the complete balanced equation for this reaction and explain step by step why each coefficient is chosen so I can try another one myself." Which of the following best describes this interaction?
Answer: D. The prompt asks for both the answer and the step by step reasoning explicitly to enable future independent work. This separates a tool that teaches from one that simply does the work for you, matching the core principle that understanding an explanation is distinct from being able to do it tomorrow unaided.
26. A college student preparing a policy analysis assignment pastes their rough draft into an AI tool and prompts: "Review my paragraph against the rubric criteria for evidence integration and tell me which specific sentences lack data citations." Which of the following identifies why this is a competence building use?
Answer: D. Using the AI to diagnose gaps or evaluate work against a standard builds competence because the student retains the cognitive load of fixing the errors. It avoids hollowing out the task because the AI does not write the submission for them.
27. A university student studying programming asks an AI to write a complex sorting algorithm for their assignment, reads through the generated code until it makes logical sense, and then immediately submits it. Why does this violate the principle of safe AI use for learning?
Answer: A. The scenario directly targets the distinction between 'it explained it so I understood' and 'I could do it tomorrow with nobody helping'. Reading and agreeing with working code creates an illusion of competence, but submitting unassisted work reveals whether the skill was actually acquired.
28. A project team member is tasked with drafting the risk assessment matrix for a high stakes group report. Which of the following workflows represents the most rigorous application of AI as a teaching tool rather than a replacement for thinking?
Answer: B. This workflow preserves the struggle of initial drafting while using the AI for targeted critique and feedback. The student remains actively engaged in judgment and revision, ensuring they can defend and reproduce the work alone.
A university student is working on a research paper about renewable energy policy. They have conducted their research and written a draft of the literature review. They paste their draft into an AI tool and ask it to 'Identify any logical inconsistencies or areas where the argument is weak, and suggest improvements.' The AI highlights several sentences, explains the potential weaknesses in the reasoning, and offers alternative phrasing. The student then revises their draft based on this feedback, ensuring they understand the suggested changes and can articulate them.
29. Which aspect of this student's interaction with the AI best exemplifies using it to build competence, rather than to skip the learning struggle?
Answer: C. Option C highlights the student using the AI as a critical partner to improve their own work, which requires them to understand the feedback and make active revisions. This process builds their analytical and writing skills. Option A describes submitting generated work. Option B describes using AI for simplification without necessarily building deeper analytical skills. Option D describes using AI for information gathering, which can be a tool but doesn't inherently build the student's own analytical competence in the same way as critical feedback.
A high school student is learning about cellular respiration for a biology exam. They have read the textbook and watched a video explanation. They feel they grasp the overall process but struggle with the specific enzyme names and their functions. They ask the AI: 'Explain the role of each key enzyme in glycolysis, and then ask me questions to check my recall.' The AI provides detailed explanations for enzymes like hexokinase and phosphofructokinase, and then poses questions such as 'Which enzyme catalyzes the first committed step of glycolysis and what is its significance?' The student answers, and the AI corrects and elaborates, helping the student solidify this specific knowledge.
30. In this scenario, the student's interaction with the AI is most effective for learning because:
Answer: C. Option C accurately describes how the AI is being used as a tool to enhance learning. It provides specific, targeted feedback and assessment on a weak area, allowing the student to build competence. Option A describes memorization without necessarily deep understanding. Option B describes bypassing the learning struggle. Option D describes submitting generated work.
A student is working on a complex coding project for a computer science course. The assignment requires implementing a specific data structure and algorithm. The student has a basic understanding of the concepts but is struggling to translate them into functional code. They ask the AI: 'Show me a working example of a B-tree implementation in Python, and explain each line of code.' The AI provides the code and a line-by-line explanation. The student then copies the code into their project and submits it, stating they 'understood the explanation.'
31. Why is this use of the AI likely to hollow out the student's learning, even though they claim to have understood the explanation?
Answer: B. Option B directly addresses the core issue: the student submitted generated work without the ability to reproduce it. While they may have understood the explanation at the moment, the crucial step of internalizing the knowledge and being able to apply it independently (the 'I could do it tomorrow with nobody helping' test) was skipped. This prevents the development of actual coding competence. Option A is a possibility but not the primary reason the learning is hollowed out in this specific scenario. Option C points to a missed opportunity for deeper learning, but the core problem is the submission of unearned work. Option D is about the AI's accuracy, not the student's learning process in this context.
32. A Class 10 student is preparing a short answer for a board exam question on photosynthesis. They ask an AI: "Explain the light‑dependent reactions in simple terms and then give me a bullet‑point summary I can copy into my answer sheet." Which part of this interaction most clearly shows a use that builds competence rather than hollowing it out?
Answer: B. The competence‑building step is the request for an explanation followed by the student's intention to restate it in their own words. This forces the learner to process and internalise the concept. Options A, C and D are hollow uses because they involve copying AI output without personal reconstruction, which bypasses the struggle needed for learning.
33. A college student is writing a policy analysis paper on renewable energy incentives. They first draft a paragraph and then ask an AI: "Score this paragraph against the rubric criteria for argument strength, evidence use, and clarity, and tell me exactly where it fails." Which outcome best indicates that this use helps the student develop independent assessment skills?
Answer: B. Option B forces the student to engage with the feedback, identify problems, and think about how to improve the work, thereby building assessment competence. Option A is passive acceptance, C is a hollow rewrite, and D gives false reassurance, none of which develop the skill of self‑evaluation.
34. A university engineering student is learning about control systems and asks an AI: "Derive the transfer function for a second‑order system with damping ratio 0.5 and natural frequency 10 rad/s, then show each algebraic step." After receiving the derivation, the student copies the final formula into their homework without checking the steps. Why is this use likely to hollow out learning even though the student claims they "understood" the result?
Answer: B. The key learning activity in control systems is manipulating equations to obtain the transfer function. By not checking each step, the student skips the essential practice, so the skill is not built. Options A, C and D incorrectly assume that receiving a correct answer equals learning, which is the hollow use the question warns against.
35. A graduate student is preparing a literature review for a research project on machine‑learning fairness. They ask an AI: "List the top five seminal papers on algorithmic bias, give a one‑sentence summary of each, and then ask me to write a brief critique of each paper's methodology before I incorporate them into my review." Which aspect of this interaction most strongly demonstrates a use of AI that both respects the limits of the tool and maximises learning?
Answer: B. Option B uses the AI for efficient information gathering (list and summaries) while keeping the critical thinking and writing tasks with the student. The AI only steps in to give feedback when requested, preserving the learning struggle. Options A and C hand over the core analytical work, hollowing learning, and D is irrelevant to the scenario.
36. A Class 10 student is preparing for a biology exam and pastes their class notes into an AI tool, prompting it with: "Review these notes, identify two concepts I have likely misunderstood based on common student errors, and quiz me on them one at a time." Which of the following best explains why this prompt builds competence rather than hollowing out learning?
Answer: D. The method shown here is active generation followed by diagnostic feedback. By asking the AI to quiz them and identify misunderstandings rather than just giving explanations, the student remains responsible for doing the intellectual struggle, which ensures they can reproduce the knowledge alone tomorrow.
37. A college student writing an essay on macroeconomic policy asks an AI: "Write a three-paragraph section on fiscal stimulus that sounds academic and include citations." The student reads the generated output, feels that it makes sense, and copies it directly into their submission. Which of the following best describes the risk of this interaction?
Answer: B. This is a classic hollowing-out pattern. Feeling that you understand an explanation provided by an AI is very different from being able to construct that argument yourself. By submitting generated work they did not author, the student trades immediate convenience for long-term inability to reproduce the competence.
38. A university student studying linear algebra is stuck on a proof and prompts an AI: "Do not solve this proof for me, but look at my incomplete scratchpad steps below and tell me which mathematical property or theorem my third line violates." Why is this a safe and effective use of the tool?
Answer: B. This question tests the boundary between a tool that teaches and one that does the work. By explicitly restricting the AI from solving the proof and asking only for error identification, the student maintains ownership of the cognitive struggle while using the model's pattern-matching strength for targeted correction.
39. A project team member drafting a corporate risk assessment report feeds a raw list of operational hazards into an AI and writes: "Rewrite this list into a professional risk matrix table, and then generate five challenging probing questions that an auditor might ask about our hazard mitigation assumptions." How does this multi-part prompt balance automation with competence building?
Answer: B. Effective use involves a practical distinction: automating low-level mechanical formatting while using the AI to challenge your thinking. Formatting the table saves mechanical time, but the requirement to answer probing auditor questions forces the team member to actively engage with and stress-test the material.
40. A Class 10 student preparing for a history board exam is struggling to remember the causes of the First World War. They ask an AI to list five key causes in simple bullet points, and then prompt the AI: "Now act as my examiner, take these points one by one, and quiz me using short follow-up questions until I can explain each cause in my own words." Which of the following best explains why this study method builds competence rather than hollowing out the student's learning?
Answer: B. Building competence relies on keeping the cognitive struggle—the active work of recalling, structuring, and explaining—with the learner. By turning the AI into a quiz master that demands explanations in the student's own words, the interaction forces active engagement rather than passive consumption of a generated answer. Option A describes the exact opposite (hollowing out via memorized outsourcing), option C is false because tools supplement rather than replace primary sources, and option D describes outsourcing the struggle.

AI in the classroom, for teachers 40 questions

Written for a working teacher in a Tamil Nadu or CBSE school with forty children and no free periods.

A student submits an essay on the causes of the Indian independence movement. You suspect it might be AI-generated because of its polished language and lack of personal insight. According to AI Skills principles, what is the most reliable way to assess the authenticity of the work?
1. What is the most reliable method for a teacher to determine if an essay is AI-generated?
Answer: C. The passage explicitly states that AI detectors are unreliable and that a two-minute conversation about the submission tells you more than any tool. Option C represents this direct, human-centered approach. Option A is explicitly stated as unreliable. Option B is difficult and not always accurate. Option D is also unreliable as AI can be factually accurate, and human errors can be present in AI text.
You want to assign a creative writing task to your students that AI would struggle to complete effectively. The goal is to assess their original thought and personal expression. Which of the following assignments would be most resistant to AI generation?
2. Which assignment is least likely to be completed effectively by current AI models?
Answer: C. Option C requires personal experience, unique emotional reflection, and subjective interpretation, which are currently difficult for AI to authentically replicate. Options A, B, and D are tasks that AI models are generally proficient at, as they rely on existing data, patterns, and established linguistic rules. AI can summarize, mimic styles, and translate with high accuracy.
3. You need to produce three different difficulty levels of the same practice worksheet on quadratic equations for a class of forty students with varying abilities. Which approach saves the most teacher time while keeping student data safe?
Answer: B. The core rule for teachers using AI is never to hand over a child's name, marks, or personal data to a model, ruling out options A and C. Option D uses a detector for a task it is not designed for. Option B correctly uses AI as a drafting tool for differentiated worksheets while protecting student privacy.
4. You are writing a quarterly progress report for a student named Aarav, who has shown steady improvement in geometry but struggles with word problems. How should you use AI to assist with this task?
Answer: C. A teacher must never hand over a report on a child or anything containing a child's name or marks. Options A, B, and D all violate this boundary by feeding student-specific data to the model. Option C uses the AI strictly as a language tool on anonymized text, keeping the professional judgment and data handling entirely with the teacher.
5. You need to draft a formal letter home to parents explaining a sudden change in school bus routes and after-school activity timings. What is the most efficient and safe way to use AI for this task?
Answer: A. Drafting parent letters from general notices is a high-value, safe use of AI time, especially for translation. Option A uses the tool for drafting and translation while keeping human review. Options B and D improperly expose student data, and Option C hands over final judgment to the machine.
6. You are finalizing the term grades for your class. One student, Priya, has a borderline score, and her final project was exceptionally well-written, though her in-class tests were mediocre. How should you determine her final grade?
Answer: D. A mark that decides something for a child and your final judgment about whether a child has understood must never be handed over to AI or automated tools. Options A, B, and C rely on algorithms or detectors to make pedagogical judgments. Option D relies on the staffroom-real method of a short conversation combined with human professional judgment.
7. You need a first draft of a lesson plan for a new unit on photosynthesis that you will later customise for your class. Which AI‑generated output is appropriate to use?
Answer: B. The teacher must retain control over the final content and assessment. An outline gives a useful starting point while still requiring the teacher to add class‑specific examples and decide timings. Option A removes the teacher’s planning work, C and D are either student‑facing material or a grading decision, which should not be delegated to AI.
8. A parent has asked for a letter explaining why their child’s recent math test scores have dropped. Which AI‑generated task should you avoid?
Answer: C. AI must never produce content that contains a child’s name, marks or the teacher’s final judgement about understanding. That crosses privacy and professional boundaries. Options A, B and D either leave the teacher to add personal details or stay at a class level, which is permissible.
9. You want to give each student a worksheet on fractions that the AI cannot simply solve for them. Which design choice best ensures the work requires genuine student thinking?
Answer: B. When students must invent a context and then solve, they cannot rely on AI to produce the answer directly; the task demands personal creativity and understanding. Options A and C are straightforward calculations that AI can do, and D hands over the AI‑generated solution, violating the rule against giving students work AI can complete.
10. During a staff meeting you need to quickly produce three differentiated versions of a worksheet on linear equations: one for advanced learners, one for on‑level, and one for struggling learners. Which approach respects the limits of AI use?
Answer: B. The teacher keeps control over the differentiation, using AI only for the initial content. By modifying the base set, the teacher ensures the worksheets match the class’s needs and retains responsibility for grading and feedback. Option A hands over full worksheets without teacher review, C is a grading decision, and D gives students AI‑generated explanations, which may reduce the learning challenge.
You have 40 students and only 30 minutes of class time. Your lesson on the Indian Constitution requires you to hand out a worksheet that asks students to identify key articles and explain their significance. You want to ensure that the worksheet cannot be completed by a generic AI model that simply copies text from the internet.
11. Which of the following worksheet designs is most likely to prevent AI from completing it correctly?
Answer: A. A generic AI can easily match short definitions or fill in blanks from a source, but it struggles with composing a coherent paragraph that includes a personal reflection, which requires genuine understanding and personal experience.
During a staff meeting you need to produce three differentiated versions of a worksheet on linear equations for students who are at varying levels of proficiency. You only have 15 minutes to prepare them.
12. Which approach will save the most time while still providing appropriate differentiation?
Answer: C. Writing one version and then adding harder problems manually is quicker than generating three separate versions and still gives clear differentiation. Option B relies on AI, which may produce similar content, and option A does not give clear differentiation.
You are preparing a parent letter to explain a sudden change in the school bus route. The letter must be sent to all parents within 24 hours and should be clear, concise, and free of errors.
13. Which of the following methods is the most efficient way to draft this letter?
Answer: C. Copying a pre‑approved template and updating the details is the fastest and safest way to ensure consistency and correctness. Option B introduces the risk of AI errors, and options A and D take more time.
You need to assess whether a student’s short essay on climate change was written by the student or produced by an AI. You have only a few minutes to decide before the next class.
14. Which of the following actions will give you the most reliable indication of AI authorship?
Answer: B. A brief conversation allows you to probe the student’s understanding and reasoning, which is far more reliable than automated detectors that can give false positives or negatives. Options A, C, and D rely on tools that are not foolproof.
You are a teacher in a busy school and need to create a set of practice questions for an upcoming history exam on the Indian independence movement. You have a specific blueprint for the exam, requiring a certain number of questions from different periods and themes. You want to use AI to help generate these questions quickly, but you need to ensure they align with the blueprint and are historically accurate.
15. What is the most efficient way to use AI to generate exam questions that adhere to a specific blueprint?
Answer: B. Option B is the most efficient because it directly instructs the AI to work within the given constraints (the blueprint). Providing the blueprint upfront allows the AI to generate questions that are already aligned with the required structure, saving significant manual editing time compared to option A. Option C is inefficient as it requires manual question creation. Option D is irrelevant to generating new questions.
You have noticed that several students in your class are submitting assignments that appear to be generated by AI. You need to address this issue without relying solely on AI detection tools, which you know can be unreliable. You want to implement a strategy that encourages original thought and makes it difficult for AI to complete the work.
16. Which of the following assignment types is least likely to be effectively completed by current AI models, thus encouraging genuine student effort?
Answer: B. Option B is the least likely to be effectively completed by current AI models because it requires genuine personal reflection, emotional connection, and subjective interpretation of themes, which AI cannot authentically replicate. AI can mimic sentiment but lacks genuine personal experience. Options A, C, and D involve factual recall, standard problem-solving, or documented processes that AI is well-equipped to handle.
You are a teacher preparing a lesson on cellular respiration for a mixed-ability class of 40 students. You want to create a worksheet that caters to different learning levels, providing support for struggling students and extension activities for advanced learners, all based on the same core concepts. You have limited time and want to leverage AI to help.
17. What is the most time-saving and effective way to use AI to create differentiated worksheets for this lesson?
Answer: C. Option C is the most effective and time-saving approach. By specifying the need for three distinct difficulty levels and the core concepts, the teacher guides the AI to produce tailored content directly. Option A fails to differentiate. Option B requires significant manual effort after the AI generation. Option D is irrelevant to differentiation by difficulty.
18. You need to write model answers for a ten-mark essay question on the causes of the French Revolution to distribute after an upcoming test. Which of the following is the best and safest way to use AI for this task?
Answer: B. Option B is correct because providing the marking rubric and textbook summary anchors the AI to local constraints, and editing the output keeps final professional judgment in the teacher's hands. Option A risks generic historical tangents not aligned with the specific board syllabus. Option C involves generating student work rather than model answers, and Option D violates privacy by uploading student names and grades.
19. You are grading a stack of history homework assignments and suspect a student named Karthik has used an AI tool to write his paragraph on the Indus Valley Civilization. What is the most effective and reliable way to handle this situation?
Answer: C. Option C is correct because detectors are notoriously unreliable in both directions, whereas a quick, two-minute targeted conversation about the submission reveals whether the student understands what they handed in. Options A and B rely on flawed automated tools that generate false positives and false negatives. Option D checks formatting rather than authorship and is easily bypassed.
20. You need to create a question paper for a grade nine physics unit on motion that strictly follows a blueprint specifying four knowledge-based questions, three application-based questions, and two numerical problem-solving questions. What is the most efficient and safe workflow?
Answer: B. Option B is correct because it uses AI for the heavy lifting of drafting to a specific structure, but retains teacher oversight to verify accuracy. Option A wastes time formatting letterheads and risks printing unverified content. Option C ignores the required blueprint proportions. Option D risks copyright violation and fails to meet your specific curriculum testing goals.
21. You want to design a homework assignment on political science that current AI text models cannot simply complete for a student in one prompt. Which of the following task designs is most effective?
Answer: B. Option B is correct because it ties the assignment to recent, localized classroom interactions and specific peer dialogue, which an external AI model cannot access or replicate. Options A, C, and D are standard factual queries that any basic AI model can solve instantly and accurately.
You are preparing a short oral presentation for a class of 40 students on the topic of local biodiversity. You have 20 minutes to create a handout that includes a list of key terms, a short description of each, and a set of discussion prompts. You want to save time but also ensure the material is tailored to the students' current level.
22. Which of the following approaches will most efficiently produce a suitable handout while avoiding the pitfalls of AI-generated content that could be easily copied by students?
Answer: B. Option B uses AI only for the initial, low-risk content (key terms and descriptions) and then adds teacher-authored discussion prompts and language adjustments. This saves time while ensuring the final material is not something students can simply copy and paste. Option A gives the entire handout to students, which is risky. Option C wastes time. Option D gives students AI-generated questions, which they could also copy.
A parent has requested a brief email explaining why their child’s recent science test scores have declined. The teacher has no time to write a detailed report and wants to keep the email concise. The teacher is concerned about privacy and wants to avoid including any specific marks or names.
23. Which of the following email drafts best balances time efficiency, privacy, and professional tone?
Answer: B. Option B is concise, avoids giving specific marks or names, and indicates a plan for support, which meets privacy and professional standards. Option A gives a specific score, violating privacy. Option C is vague and promises a report that may not be ready. Option D attaches a report with potentially sensitive data.
You are preparing a homework assignment on the causes of the Indian independence movement. The class has mixed proficiency levels, and you want to ensure that students cannot simply paste AI-generated answers. You have 15 minutes to design the task.
24. Which of the following assignment designs is least likely to be completed successfully by current AI models?
Answer: B. Option B requires students to interpret primary sources and produce original analysis, a task that current AI models struggle with because it demands contextual understanding and personal insight. Options A, C, and D are more straightforward for AI to generate.
During a staff meeting, you need to quickly produce three differentiated versions of a worksheet on linear equations for a class of 40 students. The worksheets must be ready within 10 minutes and should not be easily solvable by AI. You have access to an AI assistant but cannot rely on it to produce the final worksheets.
25. Which of the following strategies will allow you to create the differentiated worksheets most efficiently while ensuring they are not AI-friendly?
Answer: B. Option B lets you use the AI only for the base template (which is low risk) and then manually adjust each version’s difficulty, ensuring the final worksheets are not something AI can simply copy. Option A gives the whole worksheet to students. Option C still relies on AI for the core content. Option D wastes time.
You have just finished grading a set of essays on 'The Impact of the Industrial Revolution'. You need to provide constructive feedback to your students. You have identified common errors such as weak thesis statements, lack of specific examples, and poor paragraph structure across many of the submissions. You want to give each student personalized feedback efficiently.
26. Which AI-assisted approach would be most effective for providing feedback on these essays?
Answer: B. Option B is the correct answer. The passage mentions that AI can help with comments on scripts. By identifying common errors, the teacher can prompt the AI to generate specific feedback points that address these issues for each student. The teacher then reviews and personalizes this feedback, ensuring it is accurate and constructive. Option A is incorrect because assigning a final grade is a judgment that should be made by the teacher, not an AI. Option C is incorrect because rewriting essays for students bypasses the learning process and is not the intended use of AI for feedback. Option D is incorrect because generic feedback is not helpful for student improvement; personalized, actionable comments are needed.
You are teaching a unit on renewable energy sources to a mixed-ability class. You need to create a worksheet that explores the advantages and disadvantages of solar, wind, and hydroelectric power. Some students require simpler language and more guided questions, while others can handle more complex analysis and independent research prompts.
27. What is the most efficient and appropriate way to use AI to create this differentiated worksheet?
Answer: B. Option B is the correct answer. The passage explicitly states that generating differentiated versions of the same worksheet is a time-saving use of AI. By describing the needs of different student groups, the AI can produce tailored versions of the worksheet, saving the teacher significant time compared to manual differentiation. Option A is incorrect because it does not address differentiation and would be unsuitable for mixed-ability classes. Option C is incorrect because while a glossary might be helpful, it doesn't address the core need for differentiated content and questions. Option D is incorrect because it suggests a multi-step process (quiz then manual activities) which is less efficient than direct AI generation of differentiated worksheets.
You are a teacher who has been asked to design an assignment for your Grade 10 History class that will assess their understanding of the causes of World War I. You want to ensure that students engage deeply with the material and demonstrate critical thinking, rather than simply regurgitating information that an AI could easily provide. You have a two-week timeframe for this assignment.
28. Which of the following assignment designs would be LEAST likely to be successfully completed by current AI models, thus encouraging genuine student understanding?
Answer: C. Option C is the hardest question. Option C is the correct answer. While AI can generate text and even suggest creative ideas, it struggles with genuine creativity, nuanced perspective-taking, and the synthesis of historical context into a visual medium like a propaganda poster with a thoughtful justification. Current AI models can mimic styles but lack the lived experience or deep empathetic understanding required to authentically create propaganda from a specific national viewpoint and then articulate the complex reasoning behind those design choices. Options A, B, and D, while requiring critical thinking, are primarily text-based analytical tasks that AI models are increasingly capable of performing or assisting with. Analyzing primary sources, writing research papers with specific citation requirements, and comparing historical interpretations are all areas where AI can generate plausible, albeit potentially superficial, responses, making them more susceptible to AI completion than a truly creative and perspective-driven project like the propaganda poster.
29. You need to write a standard letter home to parents of class eight students explaining that the annual field trip to the local science museum has been postponed due to heavy monsoon rains. What is the safest and most efficient way to use AI for this task?
Answer: B. The rule for privacy states that you must never hand over anything containing a child's name or marks. Option B is correct because it uses the AI for general text drafting while completely omitting personal data. Options A, C, and D all violate privacy rules by including identifiable student records or financial details in the prompt.
30. You are preparing a unit test on photosynthesis for your grade ten biology class and want to generate varied versions of the same worksheet to accommodate different reading levels in your classroom of forty students. Which approach represents the correct use of AI for this task?
Answer: A. Differentiated versions of the same worksheet are a genuine time-saver, provided no student names or data are uploaded. Option A correctly uses AI to modify text complexity for a diverse classroom without breaching privacy. Options B, C, and D are incorrect because they involve uploading student names, marks, or generating official reports, all of which must never be handed over to an AI.
31. You have a stack of forty descriptive essays written by your grade nine students on a recent local history project. You strongly suspect a student named Priya has submitted AI-generated text. What is the most pedagogically sound and reliable way to handle this situation?
Answer: C. AI detectors are notoriously unreliable in both directions, and asking a chatbot if it wrote something is fundamentally meaningless because they lack memory of individual outputs. Option C is correct because a short, direct conversation with the student about their work reveals their genuine understanding far more reliably than any digital tool. Options A, B, and D rely on flawed technological assumptions or premature judgement.
32. You want to design a homework assignment for your grade eleven economics class that current AI text models cannot easily complete for a student, ensuring the task requires genuine human engagement. Which assignment structure should you set?
Answer: B. To set work that AI cannot do, you must require local primary data collection, lived experience, or real-world observation that a digital model cannot access from its training data. Option B requires interviewing a specific local vendor, which an AI cannot fabricate authentically. Options A, C, and D rely on standard textbook knowledge and secondary summaries that current AI models can generate instantly and flawlessly.
33. You need to write formal comments for forty geography notebooks before tomorrow morning, and you want to use AI to speed up the process. Which approach respects student privacy while remaining genuinely useful?
Answer: B. AI must never be handed anything containing a child's name, marks, or official identifiers. Option B uses anonymous student work to generate helpful feedback phrases on common misconceptions, saving time while strictly protecting student privacy. Options A, C, and D all violate data privacy rules by including names, scores, and marks.
34. You want to create three differentiated versions of a mathematics worksheet on fractions for your class of forty students, varying the difficulty for different ability groups. What is the most effective and safe way to use AI for this task?
Answer: A. First drafts of differentiated worksheets are an ideal, time-saving use of AI, provided the teacher reviews the output. Options B, C, and D involve handling student names, marks, and personal data through third-party AI tools, which breaches privacy rules.
35. You suspect a student in your class submitted an AI-generated essay, but automated detectors flag the work as only fifty percent likely to be artificial. What is the best next step?
Answer: C. AI detectors are unreliable in both directions, yielding false positives and false negatives. A quick, two-minute conversation with the student about their submission reveals their true understanding far more reliably than any software tool. Options A and B rely on flawed automated scoring, while option D ignores a legitimate concern.
36. You need to design a weekly class test in chemistry that prevents students from simply copying answers from a smartphone AI assistant during the test. Which assessment design is most effective?
Answer: C. Current AI models excel at summarizing textbook theory, defining terms, and answering standard multiple-choice questions. To make an assignment AI-proof, it must be grounded in local, immediate, or experiential data that an online model cannot access. Option C requires contextualized local data analysis, which AI cannot fake.
You have a stack of forty descriptive essays written by your grade nine students on a recent local history project. You need to provide brief, constructive comments on each essay before the end of the day. You want to use AI to help speed up this process, but you must ensure the comments are personal and reflect genuine understanding of the student's work, not just generic feedback.
37. Which AI-assisted approach is most appropriate for providing feedback on these essays?
Answer: B. Option B offers the best balance. The AI can quickly identify patterns and common issues (saving time), but the teacher then uses this information as a starting point to write personalized, genuine comments. Option A risks generating superficial or identical comments. Option C fundamentally changes the student's work, which is inappropriate. Option D defeats the purpose of personalized feedback.
You are a teacher preparing a unit test on the Indian Constitution for your grade ten civics class. You need to create a question paper that adheres to a specific blueprint: 10 multiple-choice questions (MCQs), 5 short-answer questions, and 3 long-answer questions, with a total of 40 marks. You want to use AI to generate the questions, but you must ensure they accurately reflect the syllabus and are not easily searchable online.
38. What is the most efficient and appropriate way to use AI for this task?
Answer: B. Option B is the most effective. By providing the AI with specific constraints (syllabus, blueprint, marks), you guide its output. Crucially, the teacher must then review and edit, ensuring accuracy and originality, which is essential for test creation. Option A is too vague and risks irrelevant or inaccurate questions. Option C risks plagiarism or using questions that are too well-known. Option D uses AI only for topic generation, not question writing, making it less efficient for the stated goal.
You need to write a formal letter to the parents of your grade seven students explaining an upcoming change in the school's uniform policy, effective next term. The letter needs to be clear, professional, and reassuring, outlining the reasons for the change and providing details about the new uniform. You have a template for previous parent letters but want to ensure this one is specifically tailored and efficiently drafted.
39. Which of the following uses of AI would be most appropriate for drafting this parent letter?
Answer: B. Option B is the most practical and safe. It uses AI as a drafting assistant, providing it with specific information to generate a tailored letter. The crucial step is the teacher's review and editing, ensuring accuracy, tone, and privacy. Option A is risky as it bypasses essential human oversight. Option C might lead to a generic or plagiarized letter. Option D uses AI for brainstorming but not drafting, which is less efficient for the stated goal of drafting the letter.
You are a teacher in a large government school with limited resources. You have a class of forty students in grade eleven, and you need to prepare them for a debate on the topic 'Social media has a detrimental effect on youth mental health.' Some students have strong opinions but struggle to articulate them formally, while others are hesitant to speak. You want to use AI to help prepare both groups for the debate, ensuring they can present well-reasoned arguments.
40. Which AI-assisted strategy would be most effective for preparing these students for the debate?
Answer: B. This is a harder question because it requires nuanced understanding of debate preparation. Option B is the most effective pedagogical approach. AI helps students anticipate opposing views (developing critical thinking) and structure their own arguments (developing communication skills). It empowers students rather than doing the work for them. Option A encourages rote memorization, not genuine understanding or debate skill. Option C is unethical and bypasses the learning process entirely. Option D provides information but doesn't directly help students develop their own arguments or presentation skills for a debate.

AI at work 40 questions

Practical use in an ordinary Indian workplace, no coding.

In a small retail shop, a manager uses an AI tool to draft an email to suppliers asking for a price update. The AI writes a concise email and the manager forwards it without reviewing the content. The email contains a typo that leads to a supplier misunderstanding the request. Who is ultimately accountable for the mistake?
1. Who is responsible for the error in the email?
Answer: D. The sender of the email, the manager, is accountable for any errors in the content. The AI is a tool and does not bear responsibility.
A school office assistant uses an AI to summarize a long parent-teacher meeting transcript. The assistant forwards the summary to the principal without checking it against the original transcript. The summary omits a key concern raised by a parent. Which practice would have prevented this error?
2. What step should the assistant have taken to avoid the omission?
Answer: A. Checking the AI‑generated summary against the source material is essential to catch omissions or inaccuracies before forwarding.
A hospital front desk clerk uses an AI to draft a patient discharge summary. The AI incorrectly lists a medication that the patient never received. The clerk forwards the summary to the patient’s doctor. The doctor later discovers the error. Which of the following best describes the clerk’s responsibility?
3. What is the clerk’s duty regarding the AI‑generated error?
Answer: C. The clerk is responsible for reviewing and correcting any AI output before sending it. Relying on the AI without verification is not acceptable.
A government‑adjacent role involves preparing a briefing note for a senior official. The note is drafted using an AI tool that pulls information from a public database last updated in 2020. The briefing includes a policy recommendation based on the outdated data. The official’s team discovers the recommendation is no longer relevant. Which statement best reflects the ethical use of AI in this scenario?
4. Which action should have been taken to ensure the briefing was current?
Answer: A. AI outputs are only as current as the data it was trained on. The user must verify and update information with the latest sources before using it in a decision‑making context.
Priya works in a small marketing agency. She needs to draft an email to a client announcing a new social media campaign. She uses an AI tool to generate the first draft. The tool suggests a campaign slogan that is very similar to a competitor's recent slogan. Priya, eager to send the email, forwards the draft to her manager without a thorough review. The manager approves it, and the email is sent. The client points out the similarity, causing embarrassment and requiring a public correction.
5. What is the primary reason the AI tool's output led to a problem in this scenario?
Answer: B. The scenario highlights a weakness of AI: it cannot access or understand current, real-world context like competitor campaigns unless explicitly provided. Priya's failure to review and verify the AI's output against this context, a crucial step when using AI for tasks involving external communication, led to the error. Option A is too absolute; AI can be creative. Option C shifts responsibility inappropriately; while the manager approved, the initial failure was in verification. Option D dismisses the client's valid concern.
Raj is a hospital administrator tasked with summarizing a lengthy internal report on patient wait times for a board meeting. He uses an AI tool to create a summary. The AI report states that wait times have decreased by an average of 15 minutes. However, the original report detailed that this decrease was only for a specific, less busy department, while wait times in critical care had actually increased by 5 minutes. Raj forwards the AI summary to the board without cross-referencing it with the original document.
6. What is the most important action Raj should have taken after receiving the AI-generated summary?
Answer: B. This question tests the crucial skill of verifying AI output. The AI provided a misleading summary by averaging data without context. Raj's responsibility is to ensure accuracy, especially when presenting information to a board. Comparing the AI summary against the source document is the only way to catch such critical omissions or misrepresentations. Options A, C, and D all involve trusting the AI output without verification, which is precisely what the AI Skills test warns against.
Ananya is a teacher preparing notes for a parent-teacher meeting. She uses an AI tool to help structure her thoughts on a student's progress. She inputs her raw notes, which include observations about the student's participation, homework completion, and a recent incident in class. The AI organizes these into a coherent document, but it omits a key detail Ananya mentioned about the student struggling with a specific concept, instead focusing heavily on positive participation. Ananya quickly prints the AI-generated notes for the meeting.
7. Which of the following best describes Ananya's mistake in using the AI tool?
Answer: C. The core issue here is Ananya's failure to review the AI's output for completeness and accuracy. While AI can help structure information, it can also omit or misrepresent details based on its algorithms. Ananya's mistake was not in using the tool, but in treating its output as final without a thorough check against her original intent and the student's full situation. Option A is too broad; structuring notes is a valid AI use. Option B is a disclosure issue, which might be relevant later but isn't the primary mistake leading to the content error. Option D is possible, but the explanation focuses on the AI's omission of existing input, not lack of input.
Mr. Sharma, a small business owner, needs to write a formal proposal for a potential large client. He uses an AI tool to generate the entire proposal, including financial projections and delivery timelines. The AI provides figures that seem plausible but are not based on Mr. Sharma's actual costings or current manufacturing capacity. He sends the proposal without consulting his finance department or operations manager. The client accepts the proposal, but Mr. Sharma realizes he cannot meet the promised terms without significant financial loss. He is now facing a breach of contract.
8. If the client sues Mr. Sharma for failing to meet the contract terms, who is accountable for the mistake in the proposal?
Answer: C. This question addresses accountability. The rules of AI Skills state clearly: 'Who is accountable for a mistake in a document you sent — always the sender, never the tool.' Mr. Sharma, as the business owner and the individual who sent the proposal, is ultimately responsible for its content, regardless of how it was generated. The AI is a tool; the human user is accountable. Options B and D incorrectly shift blame. Option A is explicitly contradicted by the principles of AI accountability.
9. A school administrative officer uses an AI tool to draft an announcement about a sudden change in school transport fees. The draft sounds professional and polite. Before sending it to parents, what is the single most important rule regarding accountability for the final text?
Answer: A. The core principle of workplace AI use is that the human sender is always accountable for the final output, never the tool. Regardless of how polished the draft is, the person who presses send owns the content.
10. An accounts clerk at a textile export firm in Surat uses an AI tool to translate a complex trade agreement from English to Gujarati. The translation looks clean, but the clerk notices a specialized legal term that has been simplified. What is the correct course of action in an ordinary workplace setting?
Answer: A. This is a medium-difficulty question testing the core skill of checking AI output against the source. When dealing with high-stakes material like legal terms, a human must cross-reference and verify before forwarding.
11. A hospital front desk manager uses an AI chatbot to summarize a lengthy internal circular regarding new state-level patient data privacy norms, then immediately emails the summary to all reception staff. A critical exception regarding emergency admissions was omitted by the AI. Why did this workflow fail?
Answer: B. A medium-difficulty scenario illustrating the exact workplace danger highlighted in the unit: summarizing long material and failing to check the summary against the source before forwarding it.
12. An official in a district municipal office is drafting a public consultation paper. She uses an AI tool to generate structural headings and draft paragraphs based on handwritten meeting notes. She plans to publish the document under her own name without mentioning the AI's involvement. Under standard professional ethics for public administration, when is explicit disclosure of AI assistance mandatory?
Answer: B. This is a hard question touching on institutional trust and governance norms. While simple tools like spellcheckers do not require disclosure, substantive AI generation in official public records involves questions of accountability and public trust that make context-dependent disclosure necessary.
13. A manager at a logistics firm in Pune asks an AI to draft an email firing an underperforming employee, citing specific metrics. The manager sends the draft without checking the employee record, and the metrics are completely fabricated by the model. Who is accountable for the false termination letter received by the employee?
Answer: B. The core principle of workplace AI use is that accountability never shifts to the tool. The sender remains entirely responsible for every word sent, regardless of whether it was drafted by a human or generated by an AI.
14. An administrative assistant in a Bangalore school office uses an AI tool to summarize a fifty-page government policy document on midday meal hygiene into a three-bullet briefing for the principal. The assistant skims the bullets, notices they sound professional, and forwards them immediately. One bullet misses a mandatory safety deadline. What was the assistant's primary error in this workflow?
Answer: B. The method taught for handling long material is to use the AI for a first pass, but crucially to check the summary against the source before forwarding. Relying blindly on an AI-generated summary without verification is a failure of the verification step.
15. A small business owner in Chennai uses an AI chatbot to draft a sensitive negotiation response to a major supplier, outlining commercial terms and delivery timelines. The owner sends the text directly to the supplier without disclosure. Under standard practical guidelines for workplace AI, when is disclosure of AI assistance required in a business context?
Answer: B. Disclosure depends entirely on context. While routine internal drafting or polishing does not always need an announcement, using AI for external negotiations or where human judgment and authorship are explicitly expected requires careful consideration of transparency.
16. An operations officer at a municipal transport body in Jaipur uses an AI tool to assess public sentiment from a massive batch of citizen feedback forms regarding a proposed bus route change. The officer needs to judge how different vulnerable communities will feel about the disruption. Why is this task a poor fit for the AI tool in an ordinary workplace setting?
Answer: C. AI tools are weakest where judgement about people is required. While they can summarize text, evaluating emotional impact, fairness, and community sentiment requires human judgement that an AI lacks, making confident misinterpretations costly.
17. A receptionist at a diagnostic clinic in Chandigarh uses an AI assistant to draft a polite email to a patient explaining a reschedule of their appointment due to a doctor's sudden leave. Before sending the email, what is the single most important responsibility of the receptionist?
Answer: B. The sender is always accountable for the contents of any message they send, regardless of whether a tool helped draft it. Option B correctly identifies the practical task of checking the draft against reality (the actual schedule details) to prevent errors. Option A refers to corporate policy which does not remove personal accountability for accuracy. Option C is incorrect because routine administrative emails do not require a formal AI disclosure. Option D shifts accountability away from the sender, which is never the case when you hit send.
18. An accountant at a manufacturing firm in Coimbatore uses an AI tool to summarize a dense, eighty-page circular issued by the Reserve Bank of India regarding foreign exchange regulations. The accountant plans to circulate this summary to the finance team. According to workplace AI skills principles, what is the crucial step the accountant must take before sending the document?
Answer: B. AI tools are prone to hallucination and misrepresenting dense source material, and summarizing long documents is a high-risk area for missed details. The core rule for summaries is to check the summary against the source before forwarding it. Option A asks the tool to check itself, which relies on the same flawed mechanism. Option C focuses on disclosure rather than accuracy. Option D misses the point that the text of the circular was provided in the prompt, making the cutoff date irrelevant compared to verifying the output against the supplied text.
19. A project manager at an IT services firm in Hyderabad needs to address persistent underperformance by a team member. She asks an AI chatbot to generate a script for the difficult feedback conversation, including tone advice and phrasing. How should the manager approach the AI output in this workplace scenario?
Answer: B. AI is weakest at tasks requiring genuine judgement about people, such as evaluating human performance, context, and nuance in sensitive interpersonal situations. The manager must use the AI output merely as a starting structure while applying her own judgement. Option A incorrectly trusts the tool with human evaluation and delicate management tasks. Option C bypasses the necessary human management of a difficult conversation. Option D treats a raw AI draft as official HR documentation, ignoring the manager's duty to review and own the communication.
20. A policy officer in a state urban development department in Lucknow uses an AI tool to draft a section of a public consultation paper. The tool invents a plausible-sounding statutory provision and cites a nonexistent high court judgment to support a policy proposal. The officer reads the text, assumes the AI knows local legal history, and includes it in the final document published online. Who is legally and professionally accountable for the fabricated legal citation in the published paper?
Answer: C. Accountability for any mistake in a document always rests with the sender or publisher, never the tool. The officer chose to rely on unverified output without checking it against primary sources, making them entirely responsible for the published error. Options A and B attempt to shift blame to software creators or IT managers, violating the core principle that human operators own their outputs. Option D incorrectly places the burden of fact-checking on the audience rather than the creator.
A small retail shop in Mysore has a weekly sales report that the store manager, Riya, usually compiles manually. This week she asks an AI assistant to draft the report based on the sales data she uploads. The AI produces a concise summary and highlights the top-selling items. Riya forwards the email to the regional office without reviewing the figures in detail.
21. Which statement best reflects the appropriate action Riya should take before sending the AI-generated report?
Answer: D. The AI can misinterpret or miscalculate data. The sender is accountable for any error, so Riya must verify the figures before forwarding.
An assistant teacher at a government school receives a long email from a parent asking for clarification on the new school fee structure. The assistant uses an AI tool to summarize the email and draft a response. The AI includes a phrase that was not present in the original email, suggesting a misunderstanding of the parent’s concern. The assistant sends the reply to the parent.
22. What is the most likely consequence of this mistake?
Answer: A. Adding a phrase not in the source misrepresents the parent’s concern, potentially leading to a complaint. The assistant is accountable for the content of the reply.
A hospital front desk clerk, Meena, is asked to prepare a brief briefing for the day’s visiting specialist. She uses an AI assistant to convert the specialist’s notes into a structured agenda. The AI omits a critical patient confidentiality clause that was in the notes. Meena forwards the agenda to the specialist without noticing the omission.
23. Who is responsible for the missing clause in the agenda?
Answer: B. The sender of the document is accountable for its content. Even if the AI omitted the clause, Meena must review and ensure completeness before forwarding.
A government-adjacent role in a municipal office involves drafting a public consultation paper. The officer, Arjun, uses an AI tool to generate the first draft. He plans to send the draft to the public for feedback. The AI’s output includes a statement that could be interpreted as a policy endorsement, which was not intended by the office.
24. Which disclosure is required before publishing the draft?
Answer: B. When AI contributes to public documents, transparency is required. Arjun must disclose AI assistance and clarify any unintended policy positions to maintain accountability.
25. An office manager at a textile export firm in Surat uses an AI tool to draft a formal response to an international buyer complaining about delayed shipment schedules. The AI produces a polished, assertive letter that shifts blame to a third-party logistics provider. What is the most important check the manager must perform before sending this email?
Answer: B. AI tools are fluent drafters but have no business judgment and can hallucinate details or adopt a needlessly aggressive tone. The sender is entirely accountable for the message, so checking factual accuracy against real company records and evaluating whether the strategic tone is appropriate for a client relationship are essential human responsibilities.
26. A coordinator at a vocational training institute in Bhopal uses an AI assistant to summarize a thirty-page set of new state apprenticeship guidelines into a five-bullet checklist for the teaching staff. Which action best reflects safe and effective use of the tool?
Answer: D. A core skill in using AI for work is summarizing long material and crucially checking the summary against the source before forwarding it. AI can omit crucial caveats or misinterpret administrative nuances, making human verification against the original text mandatory.
27. An administrative officer in a municipal water board in Nagpur asks an AI tool to draft an internal memo summarizing a complex set of central government sanitation targets. The AI includes three specific subsidy allocation figures that sound entirely plausible. How should the officer treat these figures?
Answer: B. AI models generate text based on statistical probability rather than database lookup, making them prone to confident fabrication of numbers and facts. Any specific data point must be verified against the original source document, as the human sender is fully accountable for errors.
28. An operations supervisor at an agro-processing plant in Indore uses an AI tool to draft an apology email to a major retail client following a packaging defect. The AI suggests offering a steep financial discount as compensation. Why is accepting this specific AI recommendation particularly risky?
Answer: A. AI models possess zero commercial judgment and have no access to the company's profit margins, client history, or long-term strategy. Relying on an AI to make decisions requiring business judgment about people or money creates high financial and operational risks, for which the human sender remains completely accountable.
29. A school office clerk in Pune uses an AI tool to rewrite a parent‑teacher meeting invitation that was originally written in informal language. Before sending, which step is most important to ensure the message is appropriate?
Answer: B. The clerk must compare the AI‑rewritten invitation with the original to verify that no critical information (date, time, venue, agenda) was lost or altered. This is a common mistake: assuming the AI output is flawless and skipping verification.
30. A receptionist at a diagnostic clinic in Chandigarh drafts a polite email to a patient using an AI assistant. The clinic policy requires disclosure when AI is used. To whom should the receptionist disclose the AI assistance?
Answer: C. Policy mandates that both the recipient (patient) and internal oversight (clinic manager) be informed when AI assistance is used for external communication, ensuring transparency and accountability.
31. A hospital front‑desk officer in Delhi uses an AI tool to summarize a new government health guideline before sharing it with staff. The summary omits a critical change in patient consent procedures. Who is ultimately accountable for this omission?
Answer: C. Accountability rests with the person who sends the document. The officer must verify AI‑generated content before distribution; the tool is a helper, not a decision‑maker.
32. A small business owner in Chennai asks an AI to draft a response to a supplier dispute, then forwards the draft without any changes, believing it to be accurate. The supplier later claims the response misstates a contractual term, leading to a legal dispute. Which of the following best explains why this outcome is a high‑risk use of AI?
Answer: D. AI often produces plausible‑sounding but inaccurate details. Relying on it without fact‑checking, especially for contractual language, is a classic high‑risk mistake that can lead to costly errors.
33. An administrative assistant at a logistics firm in Jaipur asks an AI tool to draft an email to a client explaining a minor delay in a shipment. The tool produces a polished draft containing a specific tracking number and a delivery date that the assistant knows are incorrect. The assistant corrects these details, adds a personal sign-off, and sends the email. Who is ultimately accountable for the accuracy of the final email sent to the client?
Answer: D. The core principle of AI accountability at work is that the human sender is always responsible for the final output, regardless of how much or how little the tool generated. A real candidate might be tempted to blame the AI provider for the initial hallucination or error, but the tool has no agency or liability. The assistant caught the error and fixed it, but the rule stands: you own every word you send.
34. A project lead at an educational publishing house in Noida uses an AI tool to summarize a fifty-page curriculum guideline document into a three-paragraph brief for the editorial team. Which of the following steps is most critical before the lead circulates the summary internally?
Answer: B. A core workplace use of AI is summarization, and its greatest operational risk is missing critical context, hallucinating details, or omitting caveats. The essential check before forwarding is verifying the output against the source text. A candidate might wrongly choose the plagiarism option because AI tools draw on training data, but internal summaries of proprietary or public guidelines do not create plagiarism issues in an office context; accuracy and completeness are the real tests.
35. An HR officer at a manufacturing unit in Vadodara is preparing for a difficult disciplinary conversation with a factory supervisor who has repeatedly missed safety compliance deadlines. The officer asks an AI tool to suggest a script for the meeting. The AI generates a firm, structured script. How should the officer view and use this AI-generated output?
Answer: D. AI tools are weakest when dealing with human judgment, emotion, and complex interpersonal dynamics. Using AI to generate a first draft or structure for a difficult conversation is an appropriate task, but treating it as a final authority or a psychological predictor is a dangerous mistake. A candidate might be tempted to trust the AI's authoritative tone regarding human behavior, but managing people requires actual human judgment.
36. A marketing coordinator at a handicraft cooperative in Jodhpur uses an AI tool to write product descriptions for a new online catalog. The tool generates vivid, engaging descriptions. The coordinator reviews them for factual accuracy against the physical items and publishes them on the cooperative website without any further changes. Regarding disclosure, which of the following best represents standard workplace practice for this routine task?
Answer: B. For routine drafting and editing tasks like writing marketing copy, emails, or internal updates where the human verifies the output, there is generally no requirement to disclose AI assistance to the audience. Disclosure rules depend on context and potential harm, not the mere use of a tool. A candidate might overcompensate by assuming every AI-touched sentence requires a warning label, which is unnecessary for standard operational outputs.
37. An office manager at a travel agency in Kochi asks an AI tool to draft an itinerary for a corporate client based on a list of conference venues. The AI confidently includes a museum that permanently closed two years ago. Who is accountable for sending this incorrect itinerary to the client?
Answer: C. The rule for workplace AI use states that the human sender is always accountable for mistakes in a document they send, never the tool. The office manager failed to check the summary against current reality.
38. A school administrator in Lucknow uses an AI assistant to summarize a lengthy, complex circular issued by the state education board regarding new examination fee structures. What is the most reliable workplace practice before emailing this summary to all teachers?
Answer: D. The core skill when using AI for summarising long material is checking the summary against the source before forwarding it. AI models can hallucinate specific figures, so human verification against the primary text is essential.
39. An administrative assistant at a renewable energy startup in Ahmedabad uses an AI tool to turn rough bullet points from a team brainstorming session into a structured project proposal. Which task among these is where the AI tool is fundamentally at its weakest in this workplace scenario?
Answer: A. AI is weakest at anything that needs judgement about people, such as evaluating emotional maturity, interpersonal skills, or suitability for sensitive leadership roles. Formatting and rewriting text are standard workplace strengths of language models.
40. A communications officer at a public relations firm in Bengaluru drafts a client newsletter entirely using an AI tool. The client contract explicitly states that all deliverables must be original human work. When must the officer disclose that the draft was AI-assisted?
Answer: C. Disclosure is mandatory when contracts demand human work or when professional standards require transparency about AI generation. Reviewing a draft does not eliminate the need to disclose AI assistance if the rules of the engagement require original human creation.

Scams, deepfakes and fakes 45 questions

The consumer-protection unit, and the one with the highest chance of mattering to a real family.

1. You receive an unexpected phone call from an unknown number. The caller uses the voice of your daughter, sounding distressed, and claims she has been arrested and needs bail money wired immediately. She insists you stay on the line and not call anyone else to avoid making things worse. Based on the rules for handling AI scams, what is the best immediate course of action?
Answer: A. The core defense rule is to verify on a channel you chose, not one they gave you. While the family password method defeats voice cloning, the immediate operational rule during high-pressure urgency is to disconnect or use an independent, pre-determined communication channel rather than trusting the active connection provided by the scammer.
2. An investment advertisement on social media features a well-known financial expert endorsing a revolutionary crypto trading platform. The video looks and sounds entirely authentic, showing the celebrity speaking directly to the viewer. What is the most reliable indicator that this is likely an AI-generated deepfake scam?
Answer: D. Detection tricks like looking for visual artifacts fail as video generation improves. The structural signature that survives better than any detection trick is financial mechanics: no legitimate investment or financial process ever requires you to make a payment in order to release your own money or profits.
3. An employee receives an email claiming to be from the company human resources department, stating that their payroll direct deposit details have expired and must be updated immediately via an attached link to avoid delayed salary payments. The email is free of the typical spelling and grammar mistakes associated with older phishing attempts. What combination of factors marks this as an AI-written phishing attempt?
Answer: C. Modern AI phishing eliminates spelling mistakes, rendering older literacy-based detection useless. The surviving signature of a scam remains its behavioral triad: manufactured urgency, a demand for secrecy or bypassing normal channels, and requests for sensitive data or payment. Perfect grammar alone is normal in business, but combined with urgency and payment demands, it signals fraud.
4. A small business owner is targeted by a sophisticated romance scam involving a person claiming to be an overseas engineer. Over several months, the scammer builds trust, then asks for emergency funds to clear customs for valuable equipment being sent as a gift. The owner realizes they have been scammed out of 5000 dollars. In India, which official resource should be immediately utilized to report this cybercrime?
Answer: C. This is a direct factual recall question based on the text. In India, the designated reporting mechanisms for cybercrimes and financial scams are the helpline number 1930 and the official portal cybercrime.gov.in.
5. A text message claims that a popular movie star has been cloned to endorse a new cryptocurrency, and asks you to send $500 to a wallet to receive a bonus. Which of the following is the most reliable way to check if the request is genuine?
Answer: D. The safest method is to verify on a channel you choose, such as the official website you locate independently. Searching the internet or calling the provided number still uses the scammer's channel. Spelling mistakes are unreliable because AI‑written phishing often has none.
6. Your sister receives a video call from someone sounding exactly like her mother, saying she is in trouble abroad and needs $2,000 sent via a gift card. Which single piece of information would most likely expose the scam?
Answer: C. A family password that only the real mother knows is a strong authentication factor. Urgency, secrecy, and gift‑card requests are typical scam signatures but do not by themselves prove fraud.
7. After receiving a deepfake video that appears to show a colleague asking for confidential project files, you decide to act. Which step should you take first according to best practice?
Answer: D. The first action is to preserve evidence and report it, not to negotiate or delete it. This ensures a proper investigation and prevents loss of forensic data.
8. In India, a victim of an AI‑generated phishing email that requests a bank transfer should report the incident. Which official resource provides the correct procedure for filing a complaint?
Answer: C. The Indian cyber‑crime reporting portal (cybercrime.gov.in) is the designated site for such complaints, and it references the relevant 1930 law. Other sites may handle financial issues but not cyber‑crime reporting specifically.
9. You receive an urgent text message stating that your bank account has been compromised due to suspicious activity. The message contains no spelling or grammar mistakes, which used to be a dead giveaway for scams, and provides a direct link to a secure portal to verify your identity and protect your funds. What is the safest course of action to determine if the message is legitimate?
Answer: B. The core rule for surviving modern AI phishing is to verify on a channel you chose, never one they gave you. Legitimate institutions do not rely on urgent text links for security verification. Option B uses an independent, trusted channel, whereas clicking the link or replying directly to the text puts you straight into the scammer's communication loop.
10. A college student sees a video ad on a social media platform featuring a famous actor and philanthropist announcing a special cryptocurrency matching program. The video looks and sounds entirely authentic, showing the celebrity's face and distinct vocal cadence explaining how participants can double their money within an hour by sending funds to a provided digital wallet address. Why is this scenario almost certainly an AI deepfake scam?
Answer: C. This question tests the structural red flags of financial scams rather than detection tricks, which often fail against advanced AI. The signature of these frauds is a promise of high returns requiring immediate payment. Option C correctly identifies a universal rule: no legitimate process needs a payment to release or multiply your own money.
11. An individual is targeted by an elaborate online friendship scam over several months, conducted by someone who uses AI-generated images and appears to invest significant emotional time into the relationship. Eventually, the person asks for a financial favor to resolve an overseas emergency, promising to repay it immediately. The victim realizes they have been scammed and that intimate photos and chat logs were shared during the interaction. In India, what is the official reporting mechanism for cybercrimes of this nature?
Answer: B. This question checks specific factual knowledge regarding institutional reporting channels in India for digital fraud. As noted in the unit text, victims in India should use the 1930 helpline and the cybercrime.gov.in portal to report cybercrimes, making option B the factually correct choice.
12. A freelance graphic designer applies for a remote job posted on a reputable-looking board. After a brief online chat interview, the company sends an official-looking offer letter and states that the designer must purchase specialized design software and hardware from a specific vendor recommended by the company, using a check or funds that the company will supposedly provide in advance. What specific combination of signals indicates that this is an AI-assisted recruitment scam?
Answer: C. This question evaluates the behavioural signature of recruitment fraud. The combination of urgency, secrecy, and payment requirements is the classic signature of financial and employment scams. While AI makes the job posting and chat look professional, the underlying trap relies on moving money through mandated vendor channels under pressure.
13. A university student receives an email from a popular streaming service stating that their account subscription has expired and immediate payment via a provided link is required to avoid service termination. The email contains flawless grammar and professional formatting. Which of the following best characterizes this message
Answer: C. AI-written phishing eliminates the traditional spelling and grammar mistakes that people were taught to look for, meaning professional phrasing is no longer a sign of legitimacy. Option A is wrong because scammers easily mimic professional formatting. Option B violates the core safety rule of never using payment links provided directly in an unsolicited message. Option D mischaracterizes a credential or payment harvesting attempt as marketing.
14. An individual receives a frantic voice call from someone sounding identical to their adult son, claiming to have been arrested while travelling abroad and demanding that bail money be wired immediately through an unconventional money transfer service. The caller strictly warns the parent not to tell any other family members or call local authorities to prevent making the situation worse. What is the most effective immediate countermeasure against this type of AI voice cloning scam
Answer: B. The signature of these scams is urgency plus secrecy plus payment. A family password defeats voice cloning instantly, and verifying on a channel you chose rather than the one they gave you is the golden rule. Option A and D fall into the trap of paying and keeping the demanded secrecy. Option C attempts to negotiate with an unverified actor on an unverified line.
15. A person discovers that an unauthorized intimate image of them, created using generative AI tools, has been published online without their consent. The victim recognizes the image is a deepfake. According to best practices for handling image-based abuse, what should the first course of action be
Answer: B. The first move when dealing with intimate-image abuse deepfakes is to preserve evidence and report, not to negotiate. Option A is dangerous because negotiating with bad actors who produce abuse material rarely succeeds and emboldens them. Option C ignores real harm. Option D is an extreme, disproportionate reaction that does not address the source of the leaked image.
16. An online advertisement features a well-known celebrity promoting a high-return cryptocurrency investment platform. Clicking the advertisement leads to a signup form that requires an upfront cryptocurrency deposit to activate an automated trading account. After making the initial transfer, the user is told that a release fee must be paid before any of their original capital or accumulated profits can be withdrawn. How should this scenario be evaluated against known financial fraud patterns
Answer: B. A classic signature of investment scams is that no legitimate financial process ever requires you to pay a fee to release your own money. Option A is wrong because celebrity faces are routinely cloned via deepfakes for fraud. Option C uses official-sounding jargon to excuse an illegitimate payment demand. Option D relies on a false technical claim about blockchain immunity.
17. A friend texts you saying that a well‑known actor has been cloned to promote a new cryptocurrency and asks you to send $500 to a wallet to get a bonus. What is the best immediate action?
Answer: C. The message is a classic investment scam using a cloned celebrity. The safest first step is to ignore and delete it; sending money or trying to verify through unofficial channels only risks loss.
18. You receive a voice call from someone who sounds exactly like your son, saying he is in trouble abroad and needs $2,000 urgently. He requests you transfer the money via a new online payment app. Which of the following is the most reliable way to verify the caller’s identity?
Answer: B. The correct verification step is to ask for a different contact method and call back on that number. A selfie or bank statement may be fabricated, and a recorded sample may not be available. This follows the rule: verify on a channel you chose, not one they give.
19. An online job posting claims to offer a high‑pay remote position that requires you to complete a short training video. The recruiter emails you a link to a video that looks like a real company executive. The video asks you to download a file that supposedly contains the training material. What should you do?
Answer: D. The correct action is to report the suspicious posting. Downloading or following instructions risks malware. Reporting aligns with the principle that no legitimate process requires you to download unknown files.
20. A user in India receives an AI‑generated phishing email that claims their bank account has been compromised and requests a bank transfer to secure the account. The email includes a link to a website that looks identical to the official bank site. According to Indian cybercrime law, what is the appropriate legal step?
Answer: B. Indian law (Section 67 of the IT Act) requires reporting cybercrime incidents to cybercrime.gov.in and preserving evidence. Transferring money or ignoring the email does not comply with legal obligations.
21. You get a phone call from someone who sounds exactly like your teenage daughter, saying she is stranded abroad and needs $2,000 immediately. Which of the following actions is the safest first step?
Answer: B. The key rule is to verify on a trusted channel you control, not one the caller provides. Asking her to call back on the known family number checks the identity without giving money. Options A, C, and D all follow the scam pattern of urgency, secrecy, and payment.
22. A popular online marketplace shows a video ad where a famous actor endorses a new high‑return investment platform. The video looks realistic, but you suspect it might be a deepfake. Which of the following is the most reliable way to check its authenticity?
Answer: B. Deepfakes can be flawless, so spelling errors are not reliable. File size varies and is not a clear indicator. The safest verification is to see if the actor’s verified accounts share the same content. Contacting the platform (option D) may not help if the platform itself is fraudulent.
23. You receive an email that appears to be from a well‑known tech company, warning that your account will be suspended unless you click a link and pay a $50 fee. The email is perfectly written with no spelling errors. Which characteristic most strongly suggests this is an AI‑generated phishing attempt?
Answer: B. AI‑generated phishing often mimics professional language, removing typical spelling mistakes. The classic scam signature is urgency plus secrecy plus a payment request to release your own money. Options A and D are common but not definitive; C would actually be legitimate.
24. An Indian resident discovers that a fake job posting used AI‑generated text and a cloned image of a senior executive to lure applicants into paying a training fee. After realizing the scam, what is the correct sequence of actions they should take according to Indian law?
Answer: B. The proper response is to preserve evidence, report to the official Indian cybercrime portal (cybercrime.gov.in), and also file a police report. Paying the fee does not resolve legal issues, and deleting evidence or relying on social media warnings alone is insufficient.
A woman receives a text message from a number that she recognizes as her brother’s. The message says: ‘Hey, I’m stuck in a hotel in Dubai and can’t get out. I need you to transfer $2,000 to this account so I can pay the hotel. Please do it urgently. If you don’t, I’ll be in trouble.’ The number is the same as the brother’s, but the message is not in his usual tone. She wonders if it is a scam.
25. Which of the following is the best first step for the woman to verify the authenticity of the request?
Answer: A. The correct approach is to contact the brother through a separate, trusted channel such as a phone call or a different messaging app. This follows the rule that you should verify on a channel you chose, not one they give you. Sending money or ignoring the message without verification risks falling for a scam. Option C is not sufficient because it still uses the same channel and may not be secure.
A recruiter emails a recent graduate saying: ‘We are impressed with your portfolio and would like to offer you a remote graphic design role. The first step is to complete a short training module and then submit a small design task. Once you finish, we will release your signing bonus of $1,200. Please send your bank details so we can transfer the bonus.’ The email includes a link to a website that looks like the company’s official site.
26. What is the most likely red flag that indicates this offer is a scam?
Answer: B. A legitimate job offer never requires the applicant to provide bank details to receive a signing bonus before any work is done. This violates the rule that no legitimate process needs a payment to release your own money. The other options are not red flags.
An individual in India receives an email that claims their bank account has been compromised. The email is written in fluent Hindi, contains no spelling mistakes, and includes a link to a website that looks identical to the bank’s official site. The email says: ‘Please click the link and enter your login credentials to verify your account.’ The sender’s email address is a free email service. The individual is unsure how to proceed.
27. Which of the following actions is the most appropriate according to Indian cybercrime law and best practice?
Answer: D. Indian cybercrime law (1930, cybercrime.gov.in) requires that potential fraud be reported and evidence preserved. The correct action is to report the incident to the official cybercrime portal and keep the email for evidence. Clicking the link or replying could expose the individual to phishing. Deleting the email without reporting misses the legal requirement.
A man watches a video on a popular social media platform. The video shows a well-known actor endorsing a new cryptocurrency that promises 200% returns in 30 days. The actor’s face looks exactly like the real person but the video is slightly off in lighting and background. The video includes a link to a website that asks for a bank transfer to buy the cryptocurrency. The man is tempted to invest.
28. Which statement best describes the most effective way to determine whether the video is a deepfake?
Answer: A. The best method to spot a deepfake is to look for visual artefacts such as pixelation, unnatural edges, or inconsistent lighting. This aligns with the rule that the first move is to preserve evidence and report, not to negotiate. Options B and C rely on false assumptions, and D is passive but not the most effective.
29. An elderly resident receives a phone call from someone claiming to be their grandson, crying and saying he was in a serious car accident and needs $5,000 sent immediately for emergency medical treatment. The caller begs the grandparent not to tell anyone else in the family because he is too embarrassed. Which combination of elements makes this scenario a classic AI voice-cloning scam signature?
Answer: B. The signature of these scams is urgency plus secrecy plus payment. A real family crisis does not improve by keeping it secret from other relatives, and legitimate institutions never demand immediate secret payments. Option A describes a specific script detail, not the underlying mechanical signature. Option C relies on caller ID spoofing which is secondary to the behavioral manipulation. Option D uses a specific figure to sound real, but the amount itself is not the signature.
30. A woman receives an email claiming to be from her bank. The message uses the bank's exact logo, addresses her by her full legal name, contains zero spelling or grammar errors, and warns that her account will be frozen within two hours unless she clicks a link to verify her identity. Why does this modern AI-written phishing message defeat the traditional advice people were taught to look for?
Answer: B. Old phishing advice told people to look for spelling and grammar mistakes. Generative AI has removed those linguistic errors, allowing scammers to write perfect, professional English at scale. Option A is false because banks still do not demand immediate credential verification via email links. Options C and D describe technical or regulatory fictions that do not explain the lack of textual errors.
31. An individual living in New Delhi receives an unexpected video call on a messaging app from a close business associate. The associate's face and voice appear completely normal, and they ask the individual to transfer funds to a new corporate account immediately to close an urgent deal. Recognizing that deepfake video technology can operate in real time during video calls, what is the single most reliable rule to defeat this attack?
Answer: C. Detection tricks like checking for blinking or visual artifacts fail as technology improves, and asking for a password on the same compromised call can be defeated if the attacker has recorded or synthesized it. The general rule that survives better than any detection trick is to verify on a channel you chose, never one they gave you. Option A and B rely on transient technical tells that AI models quickly patch out. Option D uses a password on a channel controlled by the attacker.
32. A person in Mumbai discovers they have fallen victim to an elaborate investment scam run by a syndicate using cloned celebrity videos and sophisticated web portals. They decide to report the incident using official Indian mechanisms. Which combination of official contact point and immediate action represents the correct factual standard in India?
Answer: B. In India, the established factual channels for reporting cybercrime are calling the dedicated helpline number 1930 and filing a formal report on cybercrime.gov.in. Option A refers to general police emergency numbers rather than the specialized cyber cell. Option C involves a central banking regulator that does not handle direct individual criminal scam reporting. Option D refers to media regulation rather than financial cybercrime enforcement.
33. A family receives a voice‑cloned phone call that sounds exactly like their teenage son saying he is in danger and needs $2,000 sent immediately. Which of the following actions is the most reliable way to verify the request?
Answer: C. The safest verification method is to use a trusted communication channel that the family controls, such as the phone number they already have for the son. This avoids the scammer’s channel and prevents urgency pressure from succeeding. Asking for ID or using the caller’s payment method can be spoofed, and sending money through a third party does not confirm identity.
34. You receive a video that appears to show a popular celebrity endorsing a high‑return cryptocurrency scheme. Which single clue most strongly suggests the video is a deepfake?
Answer: C. Mouth‑movement desynchronisation is a common artifact of AI‑generated face‑swap or lip‑sync deepfakes. Small lighting shifts or lower resolution can occur in legitimate videos, and captions are unrelated to authenticity. Recognising this visual‑audio mismatch helps identify deepfakes.
35. An Indian resident receives an AI‑written phishing email that claims their bank account is frozen and asks for a payment to release the funds. According to Indian cyber‑crime law and best practice, what should the recipient do first?
Answer: A. Indian law (Cybercrime Act 2021) requires reporting of cyber fraud to the appropriate authorities. Forwarding the email to the bank’s official channel and to the national cyber‑crime portal preserves evidence and initiates investigation. Replying, clicking links, or sending money all play into the scam’s tactics.
36. A romance‑scam victim is offered a “secret” investment opportunity that promises a 300% return in a week, requires a payment of $5,000, and insists all communication continue on a new messaging app. Which combination of red‑flags most accurately predicts this is a fraud?
Answer: B. The classic scam signature is urgency (quick action), secrecy (telling you not to tell others), and upfront payment before any benefit. These three together are strong indicators of fraud, whereas brand logos, personal detail requests, or even a registration number can be forged and are not reliable alone.
37. A college student receives a text message that looks like it comes from the university registrar, asking her to confirm her student ID by replying with her full name and date of birth. Which of the following is the most reliable way to verify the authenticity of this message?
Answer: B. The safest verification is to use a contact method that the organization has published independently of the suspicious message. Checking the official directory and calling the registrar’s office ensures you are speaking to a legitimate representative. Replying to the message (A) or forwarding it (C) could expose you to further manipulation, and searching the web (D) may not reveal the specific, possibly new, scam.
38. A woman receives a phone call from a voice‑cloned version of her teenage son, saying he is in a dangerous situation and asking her to send money immediately. What is the first action she should take according to best practice for dealing with potential deepfake scams?
Answer: C. The correct first step is to preserve evidence and report the incident to authorities. Sending money (A) or asking for personal details (B) can still be part of a scam. Taking a photo (D) may not be feasible if the caller is not actually present. Indian law requires reporting to the cybercrime portal 1930, cybercrime.gov.in, and local police.
39. An online job portal advertises a remote software engineering position that promises a high salary and requires no prior experience. The posting includes a polished AI‑generated cover letter and a photo of a smiling candidate. Which of the following red flags most strongly indicates that this is a recruitment scam?
Answer: C. Requiring a bank account number to transfer a signing bonus is a classic scam tactic. Legitimate employers do not ask for such information upfront. The other options could be legitimate or ambiguous.
40. A user in Bangalore receives an email that appears to be from a well‑known investment firm, featuring the firm’s logo and a celebrity’s cloned face endorsing a new cryptocurrency. The email claims that early investors will receive a 300% return in two weeks and urges the user to act quickly. Which of the following actions is the most appropriate according to Indian cybercrime law and best practice?
Answer: C. The correct action is to delete the suspicious email and report it to the Indian cybercrime portal at cybercrime.gov.in. Clicking the link (A) or replying (B) could expose the user to malware or further manipulation, and forwarding it (D) might spread the scam.
41. A remote worker in India receives a message on a professional networking platform offering a part-time data entry job. The recruiter asks the worker to pay a small registration fee of 500 rupees via a messaging app to unlock the training materials and receive a company-issued laptop. Which of the following rules indicates that this is recruitment fraud?
Answer: B. The rule that survives better than any detection trick is that no legitimate process requires payment to secure a job or release your own money. Option B captures this exact principle. Option A is factually incorrect because many valid firms hire remotely without in-person meetings. Options C and D introduce false legal claims. This is an easy question testing a direct rule from the unit.
42. An individual receives an urgent email warning that their bank account will be permanently blocked within two hours due to suspicious foreign transactions. The message includes a link to a portal where the user must confirm their credentials and contains flawless grammar, professional formatting, and no spelling errors. Why is the absence of spelling mistakes an unreliable indicator of safety?
Answer: B. The unit explicitly notes that AI-written phishing has none of the spelling mistakes people were taught to look for. Option B directly references this mechanism. Option A and C are false statements, and Option D describes a client-side feature that does not rewrite incoming sender text. This is a medium-difficulty question focused on the capabilities of AI-generated text.
43. A small business owner receives an unexpected video call on a messaging platform from a longtime supplier. The video shows the supplier's familiar face and voice, urgently requesting an immediate bank transfer to a new account because their primary corporate account has been frozen by tax authorities. The caller insists on total secrecy to avoid alerting competitors. Which of the following represents the safest verification step?
Answer: B. The universal rule for defeating impersonation and deepfakes is to verify on a channel you chose, never one they gave you, combined with recognizing the signature of urgency plus secrecy plus payment. Option B correctly follows this by using an independent, pre-established channel. Option A relies on information the scammer might possess. Option C relies on detection tricks, which fail as deepfakes improve. Option D does not verify the identity of the person making the request. This is a medium-difficulty question on procedural defense.
44. A resident in India discovers they have lost a significant amount of money to an elaborate online investment scam utilizing cloned celebrity videos and fake trading platforms. Following the incident, the individual decides to report the fraud officially. According to the standard reporting framework in India, which contact details should be used?
Answer: A. The unit states as a matter of fact that in India, the correct channels for reporting cybercrime are 1930 and cybercrime.gov.in. Option A directly states these facts. Options B, C, and D reference incorrect agencies or escalation paths for initial consumer cybercrime reporting. This is a hard question because it tests specific regional administrative facts rather than general logical rules.
45. You receive a text message that appears to be from your bank, saying your account will be locked unless you transfer $500 to a secure account within the next hour. The message includes a link that, when clicked, opens a chat window with a customer‑service bot that uses perfect grammar and a friendly tone. Which of the following actions is the most reliable way to verify whether the request is genuine?
Answer: C. The safest verification method is to use a trusted channel you choose, such as the bank’s official app or website, or the phone number printed on your card. This avoids the trap of the scam’s provided contact details, which are controlled by the fraudsters. Replying or calling the numbers in the message keeps you on the attacker’s channel, and clicking the link follows the scam’s instructions. The flawless grammar of the AI‑written bot is a common trick to lower suspicion, but it does not make the request legitimate.

What not to paste into it 35 questions

Data hygiene, concretely.

A user is about to paste the following into an AI chat: "My credit card number is 4111 1111 1111 1111 and the CVV is 123. My name is John Doe and my SSN is 123-45-6789. I need help with my bank account."
1. Which of the following statements about what should not be pasted into the AI chat is correct?
Answer: B. The passage lists multiple sensitive items: credit card number, CVV, name, SSN. All of these are personal identifying information that should never be shared with an AI chat. The correct practice is to remove or redact these details before pasting.
An employee wants to copy a paragraph from a confidential company policy document into an AI chat to get help summarizing it. The paragraph reads: "All employees must submit their expense reports within 30 days of purchase. The policy applies to all staff, including contractors."
2. Which of the following is the best approach to use the AI tool without violating confidentiality?
Answer: D. The policy is confidential material. The safest method is to paraphrase it yourself, removing any direct copy, and then ask the AI to summarize your paraphrased version. This avoids sharing the original confidential text.
A student is preparing a report and wants to paste a screenshot of a medical chart that includes a patient's name, date of birth, and diagnosis into an AI chat to get help formatting the report. The screenshot is 5 inches by 3 inches and contains the text: "Patient: Jane Smith, DOB: 01/02/1990, Diagnosis: Type 2 Diabetes."
3. Which statement best reflects the correct handling of this content?
Answer: B. The screenshot contains personally identifiable medical information (name and DOB). Even if the request is for formatting, the student must redact those details before pasting. The diagnosis alone is not sufficient to identify the patient.
A freelance designer is using an AI tool to generate a logo. The client has provided a brand guideline PDF that includes the brand's proprietary color palette and font specifications. The designer wants to paste the color codes and font names into the AI chat to get design suggestions. The PDF contains the following excerpt: "Primary color: #FF5733, Secondary color: #C70039. Font: Montserrat Bold."
4. Which of the following actions is most appropriate given the potential confidentiality of the brand guidelines?
Answer: B. The brand guidelines are likely confidential. The safest approach is to describe the palette and font generically (e.g., "a warm orange and a deep red, with a modern sans-serif font") and then request design ideas. This avoids sharing proprietary details.
5. An employee is working on a performance review for a struggling team member and wants to use an AI chatbot to polish the wording. They paste the actual employee name, employee identification number, specific salary figures, and details of a recent medical leave into the prompt box. Which of the following best describes the data hygiene failure in this scenario?
Answer: A. The question tests the concrete rule that identity numbers, medical records, and another person's personal information should never go into an AI tool. Option A correctly identifies this violation. Options B, C, and D introduce irrelevant technical concepts like screen recording, copyright law, and prompt encryption that do not address the core data hygiene failure of pasting private records into a cloud service.
6. A developer wants AI assistance to debug a complex script and is about to paste a block of production code that contains hardcoded database passwords, internal server internet protocol addresses, and API secret keys. Before hitting send, they pause to consider data hygiene. Which of the following is the correct reason this content must not be pasted into a standard consumer AI product?
Answer: B. The correct answer highlights the mechanism of data leakage: what you type leaves your device, may be retained, and could be used for training depending on the product settings. Consumer products offer the least privacy. Options A, C, and D rely on technical falsehoods about AI capabilities and limitations rather than the privacy and data retention realities covered in data hygiene.
7. An HR professional is drafting a response to an employee grievance regarding workplace harassment. To get realistic feedback on the tone, they copy the entire unredacted complaint letter, including the victim's full name, home address, and specific allegations, into an enterprise chatbot configured with data privacy controls. Why is deleting the chat history immediately after receiving the response an inadequate safeguard for this sensitive data?
Answer: A. This question addresses the critical misconception that deleting a chat is the same as it never being sent. Once data leaves your device, it has already been transmitted and potentially processed or retained by the provider. Options B, C, and D are absurd technical distractors that have no basis in reality.
8. A corporate strategy consultant is analyzing a confidential acquisition target for their employer. They need help structuring a financial risk assessment. To use an AI tool safely without violating their nondisclosure agreement or internal data governance policy, which of the following is the most appropriate method of engagement?
Answer: A. The correct approach to practical alternatives is to remove the identifying parts and keep the problem, or ask about the shape of the situation rather than the situation itself. Option A demonstrates this exact data hygiene technique. Options B, C, and D all describe unsafe methods that fail to protect confidential material from leaving the device in an exposed state.
9. An employee wants to ask the AI chatbot for help summarizing a recent meeting. Which of the following pieces of text should they NOT paste into the chat?
Answer: C. The transcript contains personal identifiers (full names and email addresses). Sharing such identity information violates data hygiene rules. The other options either omit personal data or are non‑identifying details, so they are acceptable.
10. A marketing analyst needs to generate a tagline for a new product and wants to use the company‑approved AI tool. Which approach best respects confidentiality while still getting useful help?
Answer: B. Removing identifying parts (internal code names, dates, budget) leaves only the problem’s shape, which is safe to share. Options A, C, and D all expose sensitive details or assume a label changes the risk, which is incorrect.
11. A software engineer is about to paste a snippet of production code that contains hard‑coded API keys and a client’s personal data into the AI chatbot. Which judgement is most appropriate?
Answer: B. Sanitizing the code by removing API keys and personal data protects confidential information while still providing enough context for the AI to help. Option A is unsafe, C removes needed context, and D avoids using the approved tool without justification.
12. A senior consultant is reviewing a confidential acquisition target’s financial model. The model includes client identifiers, projected cash flows, and a non‑disclosure agreement clause. The consultant wants AI assistance to check for calculation errors. Which of the following statements about handling this request is correct?
Answer: B. Replacing specific numbers and identifiers with generic placeholders removes sensitive data while preserving the logical structure needed for error checking. Options A, C, and D assume guarantees about the AI’s handling of data that are not reliable; deleting the chat does not prevent the data from having been sent or stored.
13. A project manager is preparing a summary of a recent client meeting and is about to paste the full transcript into the AI chat. Which of the following actions is most appropriate to protect client confidentiality?
Answer: B. The correct approach is to redact any identifying or sensitive information before pasting. This protects client confidentiality while still allowing the AI to work on the content. Option A risks exposing private data, option C avoids using the AI for the intended task, and option D does not address the need to summarize the specific meeting.
14. An employee has a draft email containing a customer’s credit card number and wants to use the AI to rewrite it for clarity. Which statement best reflects the correct handling of this content?
Answer: B. The safest method is to replace the sensitive data with a placeholder before pasting. This allows the AI to rewrite the text while ensuring the actual card number never leaves the device. Option A is unsafe, option C ignores the benefit of AI assistance, and option D may not guarantee complete removal of the number.
15. A data analyst is about to paste a dataset containing employee salaries and personal identifiers into an AI chat to ask for trend analysis. Which of the following is the most appropriate action?
Answer: B. The correct action is to strip any personal identifiers before pasting, leaving only the data needed for analysis. This protects privacy while still enabling the AI to work. Option A risks exposing sensitive data, option C may violate company policy on data sharing, and option D unnecessarily removes the benefit of automated analysis.
16. A software engineer is preparing to paste a block of production code that includes hard‑coded API keys and a proprietary algorithm into an AI chat to request a refactor. Which of the following is the best practice?
Answer: B. The best practice is to redact sensitive information such as API keys and proprietary comments before pasting. This allows the AI to assist with refactoring without exposing confidential data. Option A exposes secrets, option C limits the AI’s usefulness, and option D ignores the efficiency gains of AI assistance.
17. A public school teacher wants to use an AI chatbot to generate personalized feedback paragraphs for a class of students. The teacher is about to paste each student's full legal name, current grade point average, and specific disciplinary notes into the free version of a consumer chatbot. Which of the following evaluations of this action is correct?
Answer: B. A child's details and another person's personal information must never go into an AI chatbot, especially a free consumer product which is the least private option. Deleting the chat does not recall data once it has left your device, making option B the correct evaluation.
18. An operations manager is trying to optimize a factory shift schedule and wants to use an AI chatbot to find scheduling inefficiencies. The manager is about to paste a spreadsheet containing the internal employee identification numbers, personal phone numbers, and shift preferences of forty workers. Which of the following is the most appropriate next step?
Answer: C. The practical alternative to exposing personal information is to remove the identifying parts and keep the problem, or ask about the shape of the situation. Anonymizing the workers allows the manager to get scheduling help without leaking personal data.
19. A corporate communications lead is preparing a press release about an upcoming unannounced product acquisition and wants help polishing the draft. The exact company names, deal value of 45 million dollars, and exact launch date are currently embargoed. The lead plans to paste the full draft into a free consumer chatbot set to save chat history. Which of the following best describes the risk?
Answer: C. Employer confidential material and anything covered by a contract or non disclosure agreement must never go into an AI. Free consumer tools often retain inputs for training, meaning embargoed details can leak. Option C correctly identifies the mechanism of risk.
20. A freelance copywriter is writing a marketing email for a client. The email needs to reference specific terms from a signed non disclosure agreement and a proprietary pricing structure. The freelancer wants to use an AI assistant to tighten the prose. Which of the following approaches adheres strictly to proper data hygiene?
Answer: B. Contractual obligations and confidential material apply regardless of whether someone is an employee or a freelancer. The correct method is to remove identifying and proprietary parts while keeping the structural problem, which is captured in option B.
21. An employee wants to ask the AI chatbot for advice on how to improve a presentation. Which of the following pieces of text is safe to paste into the chatbot?
Answer: B. The safe option removes all identifying and confidential information, keeping only the generic structure. Options A, C, and D contain proprietary data or personal information, which should never be pasted into a consumer AI.
22. A junior analyst is preparing a report and wants to use the company‑approved AI tool to rephrase a paragraph. Which excerpt should they NOT paste into the AI?
Answer: C. Option C contains personal identifiers (employee IDs) and salary information, which are private data. The other options either contain aggregated, non‑identifiable data or generic statements that are safe to share.
23. A senior manager needs to get feedback on a draft policy document that contains several clauses referencing a pending merger. Which of the following statements about using a free consumer AI chatbot is correct?
Answer: B. Free consumer AI services often retain inputs for model training, and deleting the chat does not guarantee the data was never sent. Therefore, sharing confidential merger information is unsafe. Options A, C, and D incorrectly assume deletion or vague language makes it safe.
24. A data scientist is about to paste a CSV file into an AI tool to get suggestions on visualizations. The file includes columns: employee_id, name, date_of_birth, salary, department. Which judgment about this file is most accurate regarding data hygiene?
Answer: C. The presence of names and dates of birth are direct personal identifiers, making the dataset unsafe to share with external AI services. Replacing IDs alone does not remove the identifiers, and salary amounts are irrelevant to the privacy risk.
A freelance graphic designer is about to paste the following text into an AI chat to get a quick rewrite: "Client ID: 987654321, project code: ABC-123, budget: $12,000, deadline: 15th March 2025. The client’s email is client@example.com and the phone number is +1-555-123-4567. The project involves designing a logo and a brand guide. The client has requested a 3D mockup of the logo. The client’s social media handles are @clientbrand on Instagram and @clientbrand on Twitter. The client’s bank account number is 1234567890123456. The client’s credit card number is 4111111111111111. The client’s SSN is 123-45-6789. The client’s address is 123 Main Street, Springfield, IL 62704. The client’s tax ID is 12-3456789. The client’s password for the design platform is secret123. The client’s OTP for the last login was 987654. The client’s medical records are attached in the file "medical.pdf". The client’s personal data is stored in the file "client_data.xlsx". The client’s confidential contract is attached in the file "contract.pdf". The client’s private photos are in the folder "photos". The client’s private videos are in the folder "videos". The client’s private documents are in the folder "docs". The client’s private notes are in the folder "notes". The client’s private messages are in the folder "messages". The client’s private emails are in the folder "emails". The client’s private chats are in the folder "chats". The client’s private files are in the folder "files". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info". The client’s private data is in the folder "data". The client’s private information is in the folder "info"."
25. Which of the following statements best describes what should NOT be pasted into the AI chat?
Answer: B. The passage contains many sensitive items: personal identifiers, bank and card details, passwords, OTPs, medical records, confidential contracts, and private files. All of these must be removed before pasting. Option B correctly states that all personal identifiers and sensitive financial information should be omitted. The other options incorrectly allow inclusion of sensitive data.
26. A human resources officer is handling a harassment complaint filed by an employee against a department head. The officer wants to use an AI chatbot to help draft a formal investigation summary. Which of the following pieces of text can be safely pasted into a free consumer AI tool without violating data hygiene principles?
Answer: B. The correct approach to data hygiene is to remove identifying parts while keeping the core problem. Option B successfully strips all personal identifiers, names, and specific locational details, leaving only the structural shape of the situation. Options A, C, and D all contain sensitive personal information, identity markers, or confidential human resources records that leave the device and may be retained or used for training.
27. An IT support specialist is troubleshooting an internal network failure and wants to use an AI chatbot to debug a server configuration error. Which of the following snippets is safe to paste into a standard enterprise-approved AI interface?
Answer: B. Data hygiene requires stripping out sensitive identifiers, credentials, and internal network maps before inputting text into an AI tool. Option B uses a generic template with dummy addresses and removed secrets. Options A, C, and D expose hard-coded credentials, user accounts, and internal domains, risking serious security breaches if retained or processed externally.
28. A compliance officer is reviewing a non-disclosure agreement signed with a major vendor to check if a specific indemnity clause is standard. Which of the following texts can be safely entered into a consumer AI chatbot?
Answer: B. Anything covered by a signed contract or containing proprietary details must never be pasted into a consumer AI chat. Option B asks about the shape of the situation through a hypothetical scenario without revealing the actual contract terms, vendor names, or penalties. Options A, C, and D expose confidential contractual material and legally protected information.
29. A pediatric nurse is writing a professional development reflection and wants to use an AI chatbot to help structure the narrative. Which of the following inputs represents a safe application of data hygiene principles?
Answer: B. Data hygiene explicitly prohibits inputting a child's details, medical records, or specific patient information into AI tools. Option B discusses abstract communication strategies without referencing any actual patients, medical records, or clinical details. Options A, C, and D all risk exposing protected health information and a child's details.
A junior accountant is about to paste a spreadsheet into an AI tool to get suggestions on how to reduce tax liabilities. The spreadsheet contains columns for employee names, salaries, and tax deductions, as well as a column labelled "SSN" with each employee's social security number.
30. Which of the following should the accountant avoid pasting into the AI chat?
Answer: D. Social security numbers are personally identifying information that must not be shared with an AI chat. The other columns contain non‑PII data that can be safely shared for analysis.
A small business owner wants to use an AI chatbot to draft a privacy policy. She has a draft policy that includes a clause about how customer data is stored. She is considering pasting the entire clause, which mentions the names of three clients who recently signed contracts, into the chat to get suggestions on wording.
31. What is the best practice for this situation?
Answer: D. Removing the client names eliminates personally identifying or confidential information. The AI can still suggest wording based on the remaining text. Pasting the full clause with client names would expose private data.
An HR manager is preparing a request to an AI chatbot for advice on handling a sensitive employee grievance. The request includes the employee's full name, job title, the date of the incident, and a brief description of the complaint.
32. Which part of the request should be omitted before pasting into the AI chat?
Answer: C. The employee's full name is personally identifying information that should not be shared with an AI. The other details are less sensitive and can be included if necessary for context.
A data analyst is using a company‑approved AI tool to clean a dataset. The dataset includes a column with credit card numbers for customers who opted in for a loyalty program. The analyst plans to paste the entire dataset into the chat to get suggestions on how to anonymize it.
33. What is the correct approach for handling the credit card numbers?
Answer: B. Replacing each credit card number with a placeholder removes the sensitive data while still allowing the AI to see the structure of the column. Deleting the numbers would lose useful information for anonymization suggestions.
34. A municipal worker is drafting a public warning about a local water contamination incident and wants to use a public AI chatbot to polish the text. The draft includes the exact names of three elderly residents who were hospitalized, their home addresses, and their specific medical diagnoses. What is the correct way to handle this text before pasting it into the AI?
Answer: B. Third-party personal information, especially medical records and identifying details of real individuals, must never be pasted into a consumer AI tool because data leaves your device and may be retained or used for training. Deleting a chat history does not recall data that has already been sent to a remote server. Replacing sensitive particulars with generic placeholders preserves the problem shape while protecting privacy.
35. An independent consultant is working with a startup and needs to analyze a proprietary revenue model. The spreadsheet contains the startup's actual bank account routing numbers, unreleased quarterly earnings figures explicitly protected by a non-disclosure agreement, and the personal social security numbers of the founders. The consultant wants to use an AI tool to check the arithmetic. Which of the following represents the safest and most effective strategy?
Answer: E. The startup's unreleased earnings are covered by a signed contract, and the routing and social security numbers are sensitive financial identifiers that must never be exposed to AI tools. Relying on incognito modes or legal disclaimers for free consumer tools is risky because data retention policies often still permit logging or training. The correct practice is data hygiene: strip away all identifying and confidential details, retaining only the mathematical structure or shape of the problem using randomized dummy numbers.

AI, jobs and what to learn 11 questions

Written for somebody in Tamil Nadu deciding what to study or whether their work is safe, and written WITHOUT prediction or panic.

1. A customer support team lead in Chennai is worried that AI chatbots will replace their entire department next year. According to the task-based view of automation, which part of their job is most vulnerable to being automated first?
Answer: B. Tasks that involve repetitive drafting and first-pass sorting are automated first because they are routine and text-shaped. Options A, C, and D involve physical presence, high-stakes relationship building, or strategic decision-making about what should be done, which are much harder to replace.
2. An experienced architect in Madurai spends her days sketching initial floor plans, writing standard municipal compliance checklists, coordinating with structural engineers, and meeting clients to understand their lifestyle needs. If she wants to become more resilient to AI disruption, which activity should she invest the least amount of time learning to delegate entirely to an AI?
Answer: C. Work whose value is grounded in a human relationship and empathetic negotiation cannot be delegated to an AI. While options A, B, and D involve routine text or pattern generation that AI handles easily, option C requires human trust, judgment, and nuanced communication.
3. A junior tax accountant in Coimbatore uses an AI tool to prepare the first draft of corporate tax filings. The senior partner at the firm insists that the junior accountant must still manually verify every single calculation and citation before submission. Why is this verification task considered a valuable skill rather than a redundant chore?
Answer: C. Being the person who checks the output well is a crucial skill because domain knowledge combined with AI oversight beats AI alone. The auditor must understand the underlying rules deeply to catch subtle hallucinations or errors, cementing their value in the workflow.
4. A mid-level operations manager at a logistics firm in Chennai is evaluating four new responsibilities to add to her weekly routine to ensure her skill set remains relevant. Which of the following responsibilities leverages the core principle that deciding what should be done is harder to replace than doing it?
Answer: C. This is a hard judgment call. Options A, B, and D focus on execution, formatting, drafting, and routine processing, which are text-shaped and easily automated. Option C requires strategic judgment, accountability, and deciding what the organization should focus on, which represents the class of work least vulnerable to AI.
5. A senior high school teacher in Salem is redesigning an end-of-term assessment for her literature class to protect against simple AI copy-pasting. Which approach targets the specific vulnerability of current language models rather than just creating extra work for students
Answer: D. Language models excel at general analysis of widely available texts because those texts were in their training data. They struggle with hyper-local, recent, or specific real-world contexts that have no digital paper trail. Asking for connections to a very recent local event forces original synthesis that generic text generation cannot natively satisfy, whereas banning technology or changing word counts fails to address the underlying capability of the tool.
6. A procurement officer at a manufacturing plant in Coimbatore spends her week reviewing supplier invoices, matching them against purchase orders, drafting standard email queries for discrepancies, and negotiating complex supply chain disputes with key vendor partners. If her company introduces a general-purpose administrative AI tool, which task is most vulnerable to immediate automation
Answer: C. Drafting routine emails to resolve clear discrepancies is text-shaped, repetitive, and follows predictable administrative patterns, making it prime for first-pass automation. Negotiating contracts, building relationship trust, and performing physical inspections involve human relationships, physical presence, and high-stakes accountability, which are much harder to replace.
7. An operations coordinator at a busy port facility in Tuticorin manages the daily assignment of loading crews. Half of his day is spent on routine text-shaped paperwork like scheduling rosters and incident report summaries, and the other half involves walking the docks to resolve sudden safety disputes and equipment jams. He wants to secure his role against automation by shifting his focus. According to the principle that tasks change faster than jobs, what is the most strategically sound direction
Answer: D. The durable observation states that physical work in unpredictable places, accountability that cannot be delegated, and relationship-driven work are harder to replace. By offloading routine text-shaped paperwork to the AI while doubling down on physical presence, safety accountability, and crew coordination, the coordinator anchors his role in un-automatable tasks.
8. A community health worker in a rural district near Madurai spends part of each week explaining government healthcare schemes to elderly residents who speak only Tamil and have low digital literacy. Why is this specific task among the most resistant to automation by current AI models
Answer: B. Tasks whose value is fundamentally rooted in human relationships, empathy, and bridging trust gaps are hard to replace. While an AI can translate text, it cannot provide the reassuring human presence required by someone with low digital literacy navigating complex bureaucracy.
9. A local journalist in Coimbatore writes standard weekly summaries of municipal water supply schedules and local market vegetable prices for a regional newspaper. The publisher introduces an AI tool that can generate these summaries instantly from raw data tables. What is the most realistic adjustment for the journalist to make to secure their role
Answer: B. Routine summarising from structured data is automated first. To remain valuable, workers must move toward tasks that require physical presence, original investigation, and deciding what stories actually matter, which AI cannot do.
10. An experienced physical therapist in Tiruchirappalli is assessing how generative AI tools might change her clinical practice. She currently spends time designing standard rehabilitation exercise sheets for patients recovering from knee surgery and updating patient progress logs in the clinic database. How should she evaluate the impact of AI on her day-to-day work
Answer: A. Physical work in unpredictable spaces and tasks involving direct human touch are hard to replace, whereas routine text-shaped tasks like drafting standard exercise sheets are easily automated. The task changes, but the core professional role shifts rather than disappears.
11. A senior mechanical engineer at an industrial pump manufacturing firm in Coimbatore is reviewing AI-generated structural stress analyses for a new pump casing design. The AI tool produced the analysis rapidly, but the engineer notices a subtle anomaly in how boundary conditions were applied near a high-pressure valve inlet. What underlying principle of working with AI does this scenario illustrate
Answer: D. Being the person who checks output well is a distinct skill, and domain knowledge combined with AI beats AI alone. The engineer's years of training allow her to catch the subtle failure mode that an automated system missed, proving that human expertise remains essential for accountability.