{"course":{"slug":"ai-for-students","title":"AI for Students","url":"https://addaly.com/learn/ai-for-students","level":"beginner","track":"work"},"lesson":{"slug":"which-tool-for-which-job","title":"Which tool for which job","url":"https://addaly.com/learn/ai-for-students/which-tool-for-which-job","position":7,"of":73,"minutes":9,"answer":"Match the tool to the failure you can afford: deterministic tools for arithmetic, existence and data, and the model for explanation, phrasing, and deciding which deterministic tool to reach for.","keyIdea":"Match the tool to the failure you can afford: deterministic tools for arithmetic, existence and data, and the model for explanation, phrasing, and deciding which deterministic tool to reach for."},"body":{"format":"markdown","text":"## The default is wrong more often than it is right\n\nOpening a chatbot has become the reflex for every question, the way opening a search box was ten years ago. For a fair share of schoolwork it is the wrong instrument, and the wrongness is predictable from the mechanism in the first lesson.\n\nUse this as a first pass.\n\n**Exact arithmetic or algebra.** Not a chatbot on its own. A calculator, a spreadsheet, or a computer algebra system. Free and good: **SymPy** (Python, free), **Maxima** (free, decades old, powerful), **GeoGebra** (free, browser, excellent for graphing and geometry), **Desmos** (free graphing). Job ads and universities mention MATLAB and Mathematica; **GNU Octave** runs most MATLAB code free, and **SageMath** covers much of Mathematica's ground. If your chat tool has a code interpreter, that counts too, because it writes Python and actually runs it rather than predicting the answer.\n\n**Finding out whether something exists.** Not a chatbot. A search engine or a database. Google Scholar, PubMed, arXiv, your library catalogue. A chatbot answers *what would a source about this look like*. A database answers *does this source exist*.\n\n**Current facts.** Search, or a chatbot with search switched on and the links opened. Prices, timetables, exam dates, current office-holders, this year's syllabus.\n\n**Statistics on your own data.** A spreadsheet (LibreOffice Calc, free; Google Sheets, free), or **jamovi** or **JASP**: both free, both give you SPSS-style output without SPSS's licence, both are what a lot of psychology departments now actually use. Ask the model which test and why; run the test in the tool.\n\n**Reading a scanned handout.** OCR. Free options: Google Lens on a phone, **Tesseract** on a desktop. Many chat tools now read images directly, which is fine for short text and unreliable for tables and handwriting.\n\n**Writing and rewriting prose, explaining a concept, generating practice questions, translating, summarising text you supply, arguing with you.** This is the chatbot's home ground. It is very good at all of it.\n\n**Storing anything.** Not the chat. A document, a folder, a repository.\n\n## The free stack, named\n\nCourses and job ads name paid tools because workplaces buy them. Learn the concept on the free one and the paid one takes an afternoon.\n\n- Documents: Microsoft Word → **LibreOffice Writer**, Google Docs\n- Spreadsheets: Excel → **LibreOffice Calc**, Google Sheets\n- Statistics: SPSS, Stata → **jamovi**, **JASP**, **R** with RStudio\n- Maths: Mathematica, MATLAB → **SageMath**, **GNU Octave**, **SymPy**\n- References: EndNote → **Zotero** (free, and better)\n- Diagrams: Visio → **draw.io**, **Inkscape**\n- Images for coursework: Photoshop → **Photopea** (runs in a browser, free), **GIMP**, **Krita**\n- Video for a presentation: Premiere → **DaVinci Resolve** (free tier is full-featured), **Shotcut**, **Kdenlive**\n- Audio: → **Audacity**\n- Notes: → **Obsidian**, **Joplin**, plain markdown files\n- Programming: → Python, VS Code or VSCodium, Google Colab (free GPU hours)\n\nEvery one of these runs on a modest laptop; several run on a phone browser.\n\n## Combining them is where the value is\n\nThe strong pattern is not one tool, it is a handoff. Ask the model *which statistical test suits this design and why*, then run it in jamovi. Ask it *give me the SymPy code to check this integral*, then run the code. Ask it *what would I search to find whether this claim has been tested*, then search Scholar. In each case the model does the part it is good at (knowing the shape of the field, phrasing the query, explaining the choice) and something deterministic does the part where being wrong is expensive.\n\n## The one-sentence test\n\nBefore opening a chat window, ask: **what would it cost me if this answer were confidently wrong?** If the answer is a wasted five minutes, use the chatbot and move on. If the answer is a wrong number in a lab report, a fabricated source in a bibliography, or a program that runs and produces rubbish, use the deterministic tool and let the model advise you about which one.\n\n> Figure: What would it cost me if this answer were confidently wrong?. A two by two grid, rows Costs minutes and Costs marks, columns Explain or phrase it and Compute or verify it: Costs minutes and Explain or phrase it: Chatbot; Costs minutes and Compute or verify it: Chatbot, checked; Costs marks and Explain or phrase it: Chatbot; Costs marks and Compute or verify it: The real tool. One question sorts most of schoolwork and needs no list to remember. The bottom right is where the free deterministic tools belong: SymPy or Maxima for the arithmetic, jamovi or JASP for the test, Scholar and the DOI resolver for whether a source exists. The model is still useful in that cell. It is good at saying which tool to open and why.\n\nThis single question sorts most of schoolwork correctly, and it does not require you to remember any list.\n\n## When you have only a phone\n\nA great deal of this works on a phone browser: GeoGebra, Desmos, Photopea, Google Sheets, Colab, Zotero's web library, Scholar. What does not work well on a phone is anything requiring a large screen for comparison: reading a paper alongside your notes, or checking code output against a spec. If a phone is your only machine, use your institution's computer room for those tasks specifically, and do everything else where you are."},"check":{"question":"A student needs to check whether the integral they computed by hand is correct before submitting. Which approach is structurally sound?","options":[{"text":"Ask the model to differentiate its own answer and confirm it matches the integrand, since that is a self-check.","correct":false,"tempting":"The check happens inside the same predictive process that produced the answer, so a consistent-looking chain can be wrong at both ends."},{"text":"Ask the model, then ask a second model, and accept the answer if both agree.","correct":false,"tempting":"Two agreeing predictions raise confidence slightly but share the same failure mode on unusual integrals; a symbolic engine settles it outright."},{"text":"Ask the model to show every step of the integration, then read the steps carefully, since a visible chain of reasoning cannot hide an error.","correct":false,"tempting":"Steps are worth asking for and do catch many errors, but a plausible invalid step reads exactly like a valid one; visibility is not verification."},{"text":"Ask the model for SymPy or Maxima code, run it, and compare with the hand result.","correct":true}]},"prev":{"slug":"context-limits-and-forgetting","title":"Context, limits, and why it forgets","url":"https://addaly.com/learn/ai-for-students/context-limits-and-forgetting"},"next":{"slug":"getting-set-up-free","title":"Getting set up for nothing","url":"https://addaly.com/learn/ai-for-students/getting-set-up-free"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Which tool for which job', in AI for Students. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-students/which-tool-for-which-job (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Which tool for which job","courseTitle":"AI for Students","module":{"index":1,"slug":"what-the-machine-is-doing","title":"What the machine is doing when it helps you"},"answer":null,"keyIdea":"Match the tool to the failure you can afford: deterministic tools for arithmetic, existence and data, and the model for explanation, phrasing, and deciding which deterministic tool to reach for.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"cols":["Explain or phrase it","Compute or verify it"],"rows":["Costs minutes","Costs marks"],"type":"matrix","cells":[["Chatbot","Chatbot, checked"],["Chatbot","The real tool"]],"title":"What would it cost me if this answer were confidently wrong?","caption":"One question sorts most of schoolwork and needs no list to remember. The bottom right is where the free deterministic tools belong: SymPy or Maxima for the arithmetic, jamovi or JASP for the test, Scholar and the DOI resolver for whether a source exists. The model is still useful in that cell. It is good at saying which tool to open and why.","cellLabels":[["ask and move on","confirm the one fact"],["then say it unaided","SymPy, jamovi, DOI"]]},"alt":"Figure: What would it cost me if this answer were confidently wrong?. A two by two grid, rows Costs minutes and Costs marks, columns Explain or phrase it and Compute or verify it: Costs minutes and Explain or phrase it: Chatbot; Costs minutes and Compute or verify it: Chatbot, checked; Costs marks and Explain or phrase it: Chatbot; Costs marks and Compute or verify it: The real tool. One question sorts most of schoolwork and needs no list to remember. The bottom right is where the free deterministic tools belong: SymPy or Maxima for the arithmetic, jamovi or JASP for the test, Scholar and the DOI resolver for whether a source exists. The model is still useful in that cell. It is good at saying which tool to open and why."}},"blocks":[{"t":"h2","id":"the-default-is-wrong-more-often-than-it-is-right","runs":[{"text":"The default is wrong more often than it is right"}]},{"t":"p","runs":[{"text":"Opening a chatbot has become the reflex for every question, the way opening a search box was ten years ago. For a fair share of schoolwork it is the wrong instrument, and the wrongness is predictable from the mechanism in the first lesson."}]},{"t":"p","runs":[{"text":"Use this as a first pass."}]},{"t":"p","runs":[{"text":"Exact arithmetic or algebra.","b":true},{"text":" Not a chatbot on its own. A calculator, a spreadsheet, or a computer algebra system. Free and good: "},{"text":"SymPy","b":true},{"text":" (Python, free), "},{"text":"Maxima","b":true},{"text":" (free, decades old, powerful), "},{"text":"GeoGebra","b":true},{"text":" (free, browser, excellent for graphing and geometry), "},{"text":"Desmos","b":true},{"text":" (free graphing). Job ads and universities mention MATLAB and Mathematica; "},{"text":"GNU Octave","b":true},{"text":" runs most MATLAB code free, and "},{"text":"SageMath","b":true},{"text":" covers much of Mathematica’s ground. If your chat tool has a code interpreter, that counts too, because it writes Python and actually runs it rather than predicting the answer."}]},{"t":"p","runs":[{"text":"Finding out whether something exists.","b":true},{"text":" Not a chatbot. A search engine or a database. Google Scholar, PubMed, arXiv, your library catalogue. A chatbot answers "},{"text":"what would a source about this look like","i":true},{"text":". A database answers "},{"text":"does this source exist","i":true},{"text":"."}]},{"t":"p","runs":[{"text":"Current facts.","b":true},{"text":" Search, or a chatbot with search switched on and the links opened. Prices, timetables, exam dates, current office-holders, this year’s syllabus."}]},{"t":"p","runs":[{"text":"Statistics on your own data.","b":true},{"text":" A spreadsheet (LibreOffice Calc, free; Google Sheets, free), or "},{"text":"jamovi","b":true},{"text":" or "},{"text":"JASP","b":true},{"text":": both free, both give you SPSS-style output without SPSS’s licence, both are what a lot of psychology departments now actually use. Ask the model which test and why; run the test in the tool."}]},{"t":"p","runs":[{"text":"Reading a scanned handout.","b":true},{"text":" OCR. Free options: Google Lens on a phone, "},{"text":"Tesseract","b":true},{"text":" on a desktop. Many chat tools now read images directly, which is fine for short text and unreliable for tables and handwriting."}]},{"t":"p","runs":[{"text":"Writing and rewriting prose, explaining a concept, generating practice questions, translating, summarising text you supply, arguing with you.","b":true},{"text":" This is the chatbot’s home ground. It is very good at all of it."}]},{"t":"p","runs":[{"text":"Storing anything.","b":true},{"text":" Not the chat. A document, a folder, a repository."}]},{"t":"h2","id":"the-free-stack-named","runs":[{"text":"The free stack, named"}]},{"t":"p","runs":[{"text":"Courses and job ads name paid tools because workplaces buy them. Learn the concept on the free one and the paid one takes an afternoon."}]},{"t":"ul","items":[[{"text":"Documents: Microsoft Word → "},{"text":"LibreOffice Writer","b":true},{"text":", Google Docs"}],[{"text":"Spreadsheets: Excel → "},{"text":"LibreOffice Calc","b":true},{"text":", Google Sheets"}],[{"text":"Statistics: SPSS, Stata → "},{"text":"jamovi","b":true},{"text":", "},{"text":"JASP","b":true},{"text":", "},{"text":"R","b":true},{"text":" with RStudio"}],[{"text":"Maths: Mathematica, MATLAB → "},{"text":"SageMath","b":true},{"text":", "},{"text":"GNU Octave","b":true},{"text":", "},{"text":"SymPy","b":true}],[{"text":"References: EndNote → "},{"text":"Zotero","b":true},{"text":" (free, and better)"}],[{"text":"Diagrams: Visio → "},{"text":"draw.io","b":true},{"text":", "},{"text":"Inkscape","b":true}],[{"text":"Images for coursework: Photoshop → "},{"text":"Photopea","b":true},{"text":" (runs in a browser, free), "},{"text":"GIMP","b":true},{"text":", "},{"text":"Krita","b":true}],[{"text":"Video for a presentation: Premiere → "},{"text":"DaVinci Resolve","b":true},{"text":" (free tier is full-featured), "},{"text":"Shotcut","b":true},{"text":", "},{"text":"Kdenlive","b":true}],[{"text":"Audio: → "},{"text":"Audacity","b":true}],[{"text":"Notes: → "},{"text":"Obsidian","b":true},{"text":", "},{"text":"Joplin","b":true},{"text":", plain markdown files"}],[{"text":"Programming: → Python, VS Code or VSCodium, Google Colab (free GPU hours)"}]]},{"t":"p","runs":[{"text":"Every one of these runs on a modest laptop; several run on a phone browser."}]},{"t":"h2","id":"combining-them-is-where-the-value-is","runs":[{"text":"Combining them is where the value is"}]},{"t":"p","runs":[{"text":"The strong pattern is not one tool, it is a handoff. Ask the model "},{"text":"which statistical test suits this design and why","i":true},{"text":", then run it in jamovi. Ask it "},{"text":"give me the SymPy code to check this integral","i":true},{"text":", then run the code. Ask it "},{"text":"what would I search to find whether this claim has been tested","i":true},{"text":", then search Scholar. In each case the model does the part it is good at (knowing the shape of the field, phrasing the query, explaining the choice) and something deterministic does the part where being wrong is expensive."}]},{"t":"h2","id":"the-one-sentence-test","runs":[{"text":"The one-sentence test"}]},{"t":"p","runs":[{"text":"Before opening a chat window, ask: "},{"text":"what would it cost me if this answer were confidently wrong?","b":true},{"text":" If the answer is a wasted five minutes, use the chatbot and move on. If the answer is a wrong number in a lab report, a fabricated source in a bibliography, or a program that runs and produces rubbish, use the deterministic tool and let the model advise you about which one."}]},{"t":"p","runs":[{"text":"This single question sorts most of schoolwork correctly, and it does not require you to remember any list."}]},{"t":"h2","id":"when-you-have-only-a-phone","runs":[{"text":"When you have only a phone"}]},{"t":"p","runs":[{"text":"A great deal of this works on a phone browser: GeoGebra, Desmos, Photopea, Google Sheets, Colab, Zotero’s web library, Scholar. What does not work well on a phone is anything requiring a large screen for comparison: reading a paper alongside your notes, or checking code output against a spec. If a phone is your only machine, use your institution’s computer room for those tasks specifically, and do everything else where you are."}]}],"check":{"question":[{"text":"A student needs to check whether the integral they computed by hand is correct before submitting. Which approach is structurally sound?"}],"options":[{"text":[{"text":"Ask the model to differentiate its own answer and confirm it matches the integrand, since that is a self-check."}],"correct":false,"tempting":[{"text":"The check happens inside the same predictive process that produced the answer, so a consistent-looking chain can be wrong at both ends."}]},{"text":[{"text":"Ask the model, then ask a second model, and accept the answer if both agree."}],"correct":false,"tempting":[{"text":"Two agreeing predictions raise confidence slightly but share the same failure mode on unusual integrals; a symbolic engine settles it outright."}]},{"text":[{"text":"Ask the model to show every step of the integration, then read the steps carefully, since a visible chain of reasoning cannot hide an error."}],"correct":false,"tempting":[{"text":"Steps are worth asking for and do catch many errors, but a plausible invalid step reads exactly like a valid one; visibility is not verification."}]},{"text":[{"text":"Ask the model for SymPy or Maxima code, run it, and compare with the hand result."}],"correct":true,"tempting":null}],"order":[2,0,3,1],"why":null},"prev":{"slug":"context-limits-and-forgetting","title":"Context, limits, and why it forgets","minutes":8},"next":{"slug":"getting-set-up-free","title":"Getting set up for nothing","minutes":8}}}