A PREREQUISITE

ChatGPT, Claude, Perplexity. What's actually happening?

Three different apps. Same magic. By the end of this article, you'll explain it to a kid, or to that uncle who keeps asking how Karthik next door cleared TNPSC so fast.

01 · WHY THIS MATTERS

Why this matters

A young Tamil woman at a sun-lit study desk reading her phone, with Kani, a small Tamil baby robot with a cream body, sky-blue accent panels, a kolam pattern on its chest, mustard-yellow antenna tip, and large rectangular sky-blue screen-eyes showing curious question-mark shapes, sitting on the corner of her desk and watching her phone. A cream stainless-steel tumbler of filter coffee sits beside the open notebook. Warm afternoon light through the window.
"What's actually inside the app she's holding? That's the question this article answers."

Across Tamil Nadu right now, four different people are typing into the same kind of app on their phones.

A 22-year-old in Coimbatore is preparing for a remote software interview, asking ChatGPT to mock-interview her in Python, even though she dropped out after second year of engineering college. She'll get the offer next month.

A mother in Madurai tonight is helping her daughter solve a Class 11 physics problem about projectile motion. ChatGPT walked her through the concept first so she could explain it to her daughter in their kitchen.

A small business owner in T.Nagar is writing a legal-tone reply to a vendor dispute. His English is functional, not fluent. ChatGPT drafts the letter in a formal corporate register, saving him the cost of a legal consultant.

A college student in Chennai is preparing for the TNPSC Group 2 exam. She types a prompt for a summary of Indian Polity. Three seconds later, she has structured notes that would usually require hours of tuition.

How is this possible?

Whether they open ChatGPT, Claude, or Perplexity, they are seeing the same result. Different companies, different names, same magic.

But what is happening inside? Is it thinking? Does it understand the law or the physics? In the next ten minutes, we will look under the hood. By the end, you'll be able to explain it to a ten-year-old.

THE ONE-SENTENCE ANSWER

AI is like a "Sivagami Computer Josiyam" that memorized the entire internet, stitching together answers that sound right, one word at a time.

02 · THE ANCHOR

Meet the Sivagami Computer

The Sivagami Computer Josiyam, a tall maroon-red painted Tamil festival robot mannequin with antenna and a phone receiver to its ear, stands on Marina Beach at golden hour. An elderly Tamil sadhu in white veshti and rudraksha mala stands beside it wearing headphones, listening intently. In the background, kite-flyers, families on the sand, and a distant lighthouse silhouette.
"Drop a coin. Hear your future. Walk away thinking it knew you."

If you have ever been to a trade fair or a temple festival, you've seen it: Sivagami Computer Josiyam.

It's a red robot mannequin with glowing eyes, sitting behind a small keyboard. You give it your name or date of birth, it whirrs and hums, and then a small printer spits out a slip of paper with your "future" on it.

Does that robot know your future? No. It doesn't know you at all.

Inside that box is a memory chip filled with thousands of generic sentences about life, wealth, and health. When you provide an input, it simply selects the sentences that fit the pattern of your category.

AI works the same way.

ChatGPT is like that Sivagami Computer, but with three massive upgrades:

03 · INTERNET SCALE

Same trick. Internet scale.

The maroon-red Sivagami Computer robot mannequin stands centered with a black phone receiver held to its ear. Above its head, a vertical waterfall of handwritten manuscripts, photographs, book pages, and notes pours down INTO the top of its head and antenna, the fragments fanning outward at the source. Vintage-toned cream paper background, deep maroon and aged paper palette.
"The Sivagami Computer has read everything ever written. It still doesn't understand any of it."

Now picture the Sivagami Computer again, but at internet scale.

This bigger Sivagami Computer was fed every Wikipedia article, every Tamil poem, every TNPSC question paper, and every coding tutorial ever written. Not a coin slot, not one card. Trillions of words, in every language, on every topic.

It memorized the patterns. After the words "Indian Constitution is..." the most likely next word is "the." After "the," the most likely word is "supreme." After "supreme," the word "law." It learned this by reading the same sentence written a hundred thousand different ways.

When you ask for a summary, the Sivagami Computer doesn't know what a Constitution is. It looks at its massive internal pile of text and asks: "Based on everything I have seen, what is the most likely word a human would write next?"

It prints that word. Then it predicts the next one. Thousands of times per second. Because it has seen so much data, the "guess" reads back like a real answer. Same trick as the festival robot. Just a much, much bigger pile of memorized text.

04 · FOUR MYTHS TO DROP

What AI is NOT

Once you see the trick, you can bust four big myths about AI.

One. AI is not magic.

It is a massive math equation. Your calculator knows 2+2=4 through math. AI knows that "Deepavali" is usually followed by "Wishes" through probability.

Two. AI is not thinking.

When it writes a summary, it isn't picturing the Parliament. It is just matching patterns. Thinking requires a brain; AI only requires a fast processor.

Three. AI is not conscious.

It doesn't want to help you. It has no feelings. It is a program that runs when you call it and stops when you don't.

Four. AI is not always right.

The maroon-red Sivagami Computer Josiyam robot mannequin (with the brand label சிவகாமி கம்ப்யூட்டர் ஜோதிடம் visible on its chest plate, antenna on its head, black phone receiver held to its ear) extends a small printed fortune card toward Kani, a small Tamil baby robot with a cream body, sky-blue accents, kolam pattern on chest, and mustard-yellow antenna tip. Kani is in confused register: rectangular sky-blue screen-eyes filled with Tamil-script-like text (the prophecy taking up Kani's mental space). Behind them, a Tamil temple gopuram glows in golden festival light, with kites in the sky, festival string lights overhead, and people in saris in the courtyard background.
"The card sounds wise. The boy doesn't recognize his own life in it. Neither does ChatGPT."

Just like the Sivagami Computer might give you a generic prediction that doesn't apply to your life, AI can make things up. This is called hallucination. It predicts the next word based on probability, even if that word is factually wrong.

05 · THE THREE FAMILIES

The three families of AI

Three small Kani baby robots stand side by side, each with a cream body, sky-blue accent panels, kolam pattern on the chest, and mustard-yellow antenna tip. Each Kani's rectangular sky-blue screen-eyes carry a context-specific shape mirroring the phone it holds. The leftmost Kani holds a phone showing a face-unlock screen with a green checkmark and a smiling-face icon; its eyes show closed-arc satisfied shapes (the phone recognized me). The middle Kani holds a phone showing a translation interface with up-and-down arrows between two text blocks; its eyes show tiny up-and-down arrow icons. The rightmost Kani holds a phone with warm-amber chat bubbles emerging from the screen; its eyes show tiny chat-bubble shapes. All three rendered in soft watercolor and ink on warm cream paper.
"Three different jobs. Same trick underneath. Memorize, match, predict."

You've seen ChatGPT, the kind that talks.

But the same Sivagami Computer trick (memorize patterns, predict result) shows up in three different forms in your life right now.

The first family: AI that SEES

When you unlock your phone with your face, that's AI.

This Sivagami Computer was shown millions of human faces during training. It learned the specific pattern of your eyes, nose, and the geometry of your face. Now when it sees you, it predicts: "99% probability this is the owner. Yes." Phone opens.

It doesn't know you. It matches the pattern. Google Photos auto-tagging your relatives in old pictures, same family.

The second family: AI that TRANSLATES

Google Translate is AI. It memorized millions of English sentences alongside their Tamil translations.

When you type something in English, it predicts the Tamil version that most likely matches. Sometimes it's perfect. Sometimes it's funny-wrong, because the exact sentence you typed wasn't in its memory, so the machine stitches together pieces from similar sentences it does have.

The third family: AI that TALKS

This is ChatGPT. This is Claude. This is Perplexity. This is the family taking the world by storm right now.

It didn't just memorize faces or simple translation pairs. It memorized almost all the text humans have ever written. When you ask it something, it predicts what text most likely comes next, one word at a time.

Every word it writes is a tiny prediction. The predictions are so good, you can't tell them apart from a human. But it's still a Sivagami Computer. Just a much, much bigger one.

06 · THE MENTAL MODEL

The one mental model to remember

A close-up of a smartphone displaying a chat with 'அம்மா' (Amma). Her message at 4:17 PM reads 'சாப்ட்டியா?' (Have you eaten?). The user has just typed 'ம்' in reply. Tamil autocomplete pills appear above the keyboard: 'ம் சாப்ட்டேன்', 'இல்ல', 'இப்போ சாப்டுறேன்'. A cream stainless-steel tumbler of filter coffee sits on the wooden table. To the right, Kani the small Tamil baby robot stands on the table holding its own filter coffee tumbler with steam, watching the autocomplete; its rectangular sky-blue screen-eyes show two horizontal dashes inside (typing-pause indicator, specific to the autocomplete moment).
"Your phone is already pattern-matching for you. ChatGPT just does it on a billion-paragraph scale."

Think about when you type a message on WhatsApp. Your phone suggests the next word. That is Autocomplete.

ChatGPT is autocomplete with superpowers.

Your phone only knows your messages. ChatGPT knows everything ever written. Your phone predicts one word; ChatGPT predicts entire essays. But the logic is identical. It is a machine asking: "What word comes next?"

Six words to keep

Carry these mental tools. The next time anyone tries to sell you on "AI magic," check it against this list.

01AI = pattern matching, not magic.
02It memorized the internet.
03It does not understand.
04It predicts the next word.
05It is sometimes confidently wrong.
06Autocomplete with superpowers.

Five questions to try this week

Don't believe me. Test it yourself. Open the app on your phone right now and run these.

01
The News Test

Ask it about a local event from two hours ago. Does it know? Why or why not?

02
The Comparison

Ask ChatGPT and Claude the same question about your local MLA. Are the answers different?

03
The Autocomplete Race

Start a sentence on your phone keyboard and keep clicking the suggestions. Now do it in ChatGPT. What's the difference in scale?

04
The Local Test

Ask it to write a poem about your specific village. Does it sound generic or specific?

05
The Hallucination Hunt

Ask it for a list of books written by you. See if it invents any titles. Save the screenshot.

Quiz · Did you understand this article?

0 / 5

5 questions · 2 minutes · See where you stand.

Q1
What is AI most accurately compared to in this article?
The article anchors AI to the Sivagami Computer Josiyam: a robot mannequin with a stored-pattern fortune-telling mechanism. Coin in, card out. Same trick at internet scale.
Q2
When ChatGPT writes a Tamil poem, what is it actually doing?
It's pattern-matching one word at a time. No thought, no understanding. Just very, very good prediction.
Q3
What's the technical term for when AI confidently states wrong facts?
Hallucination is when AI produces confident-sounding but factually wrong output, because it's matching patterns without understanding.
Q4
Which of these is NOT one of the three families of AI mentioned in this article?
AI doesn't feel anything. The three families are sees, translates, and talks. There is no "feels" family: AI has no consciousness.
Q5
What's the simplest mental model for ChatGPT?
Your phone keyboard predicts one word. ChatGPT predicts entire essays. Same trick. Different scale.
🎉
Unnal Mudiyum!

You got it. Now go explain this to someone.

What was unclear, or what surprised you?

Next article

How does ChatGPT know today's date if it just memorized old text?

If AI doesn't have a brain, how does it seem to reason through a math problem? That's the next article.