Lesson 30 of 32

The Senses of the Machine

Files, Images & Voice

These assistants can now look at a photograph, read a document, work through a sheet of figures and listen to your voice - so stop labouring to describe what you could simply show. But a picture without a purpose is a ship without a port: name what the thing is, supply the context the camera could not capture, point to the exact page or column that matters, and say what you want out of it.

Reach for it whenWhen the problem is in front of you rather than in your head - a spotted leaf, a bill you cannot read, a forty-page circular handed over at five o'clock, a column of figures - or when typing is itself the obstacle.

Perplexity does take uploads - a PDF, an image, a spreadsheet - and will answer questions about them, so the scroll half of this chapter works here. Its distinct contribution is the join between your document and the live web: it can tell you what your circular says and whether a later one has replaced it, what your bill shows and what the current tariff order says it ought to show. That is a genuinely useful second opinion on a document you already half understand. But its senses are the narrowest of the four. It is weaker than the others at reading a photograph for judgement - a diseased leaf, a crack in a wall - and it has no equivalent of a sustained spoken conversation: you can dictate a question in the app and have an answer read back, but there is no talking to it the way the grandmother in this chapter talks to ChatGPT Voice or Gemini Live.

Worth knowingSo two of the three examples here belong elsewhere, and saying so is more use to you than pretending otherwise. For the leaf, come here to find what is actually being reported in Karnataka this season and what the state agriculture university advises, with links - then show the photograph to Gemini or ChatGPT, or better still to the officer at the Krishi Vigyan Kendra. For the grandmother this is simply the wrong tool: she needs a voice she can interrupt and an answer spoken slowly in Tamil. The document example is the one that fits, and even there remember that an answer engine's instinct is to go outside the file - you must say “from the attached document only” and then check that what it cites are page numbers and not web links. Crop personal details before uploading, as always; here the file is sitting in a tool whose whole business is search.
  • Ask for page numbers from your file, not web links, when it quotes.
  • Say “from the attached document only” - its instinct is to search.
  • Use it to check whether your circular has since been superseded.
  • A photograph that needs judgement belongs in another tool.
  • There is no spoken conversation here; dictation is not voice mode.
The Bare Photo vs. The Guided Eye

Instead of

What is wrong with this leaf?

Perplexity version

This is the half of the job this tool is good at. The photograph goes to another tool, or to the agriculture officer; what comes here is the question of what is actually going round Karnataka this season.

I grow tomatoes near Kolar, Karnataka. The monsoon has just begun. About 20 of 300 plants show brown spots with yellow edges on the older lower leaves, spreading over a week.

Cite everything, and prefer official sources - the Karnataka Department of Agriculture, the University of Agricultural Sciences Bengaluru, ICAR-IIHR - over blogs and seed company pages.

1. Which tomato leaf diseases are being reported or advised against in Karnataka at the start of this monsoon? Link the advisory and give its date.
2. Set early blight, septoria leaf spot and bacterial spot side by side in a table: what the lesion looks like, which leaves it begins on, which weather favours it, and the field test that separates it from the others. Cite each row.
3. What does the state advisory recommend as a smallholder's first step, and does any of it need a licence or a registered dealer? Cite it.
4. Where is the Krishi Vigyan Kendra for Kolar district, and is there a published advisory helpline number? Give the page.

Mark anything you cannot source as unverified. I would rather carry an honest gap to the officer than a confident guess to my field.
Open Perplexity 1,347 characters

Then loop it

  1. Your comparison table came off a general extension page. Find the Karnataka advisory for this season and tell me where it differs from the general one.
  2. Now compress all this into ten lines I can read out at the Krishi Vigyan Kendra, with the source beside each claim, and save the thread in a Space called “Tomato - Kolar” so next season starts from these sources.

Why it worksThe photograph still needs an eye that can judge it, so the showing happens elsewhere - and what the words were supplying here (place, season, spread) becomes a sourced question about what is actually being reported in Karnataka this month.

The Whole Document vs. The Pointed Page

Instead of

Can you summarise this PDF for me?

Perplexity version

[Upload the 40-page circular to this thread.]

Two separate jobs, and keep them apart in your answer.

**Job one - from the attached circular only. No web sources at all.**
Read Section 4 and the annexures.
1. The steps an employee must follow, in order.
2. The documents required, against each step.
3. Every deadline mentioned anywhere in it.
After each point give the PDF page - the page, not a link. Where the circular is silent or contradicts itself, say which.

**Job two - now use the web, and cite it.**
4. Is this circular still in force? Search for any later circular, office memorandum or notification on this pension procedure that amends or supersedes it, and link what you find, with its date.
5. Has the department published any clarification or FAQ on the procedure? Link that too.

If job two turns up nothing, say nothing was found rather than offering me a general article about pension rules. And do not let anything from job two alter your answer to job one: if a web source and my circular disagree, show me both and label which is which.
Open Perplexity 1,060 characters

Then loop it

  1. Quote the paragraph on page 18 about the form, exactly as printed, and give the PDF page. Nothing from the web in this answer.
  2. Your later-circular search returned a news report rather than a government page. Look again on the department's own site, and tell me honestly if there is no published amendment.
  3. Now save the thread in a Space called “Pension procedure” with the circular attached, so that when an employee asks me next month I start from this and only re-check whether it still stands.

Why it works“From the attached circular only” does the same anchoring work here, with PDF pages to verify it - and the second half uses the one thing this tool adds, which is finding out whether the circular is still the current one.

The Hesitant Typist vs. The Clear Speaker

Instead of

(spoken) Bill… why so much?

Perplexity version

Said plainly: this one does not belong here. Lakshmamma needs a voice she can interrupt and an answer spoken slowly in Tamil, and this tool has no spoken conversation to offer - a dictated question with the answer read back is not the same thing. Her bill should go to ChatGPT's Voice mode or to Gemini Live.

What a grandchild or a neighbour can usefully do here is check the bill against the published rules, once, and then explain it to her.

[Upload the photograph of the bill, with the consumer number and address covered.]

1. From the attached bill only: the amount due, the last date, the units consumed, and every charge listed separately. Quote the figures as printed, and say which ones are unclear in the photograph.
2. Now from the web, cited: what is the current TANGEDCO domestic tariff and slab structure for Tamil Nadu? Link the tariff order and give its date.
3. Do the charges on this bill match that tariff for these units? Show the arithmetic. If something does not match, say what and by how much.
4. What is the published late payment charge, and what does TANGEDCO say about how and where a pensioner can pay in person in Madurai? Cite it.

Mark as unverified anything you cannot source. I will take your answer and explain it to her myself, in Tamil, slowly.
Open Perplexity 1,283 characters

Then loop it

  1. The tariff you used is an older order. Find the one currently in force, give me its date, and redo the arithmetic in point 3.
  2. Now give me five short Tamil sentences I can read aloud to her - the amount, the date, why it rose, the late fee, and one way to use less current - with the English underneath so I can check I have it right.

Why it worksVoice is the whole point of this example and this tool has none, so the honest adaptation sends the spoken conversation elsewhere and uses the citations for what they are worth: checking the bill against the published tariff before anyone explains it to her.

Loop it

When you share a photo or file, the first answer tells you what the AI actually saw. Check that first: did it read the right page, the right column, the right figure? If it misread something, correct it - "The amount in row 7 is ₹1,250, not ₹125" - and ask it to redo the answer. If a photo was blurred, send a clearer one or a close-up. Then narrow the focus, ask for page numbers or quotes to verify, and turn the result into something you can reuse: a checklist, a table or a spoken summary.