It almost feels as if something on the other side truly understood what you meant.

But ChatGPT doesn't understand language the way humans do. It doesn't read words, recognize sentences, or interpret meaning the way we naturally can.

So what really happens between the moment you press Enter and the moment your answer appears?

In this article, we'll trace that entire journey—from how your text is converted into numbers to how a transformer model predicts one token at a time to generate the response you see.

1.What is an LLM?

LLM stands for Large Language Model.

It is a model trained on lots of text so that it can understand and generate responses in human language.

It bridges the gap between human language and machine understanding, allowing us to communicate with AI using everyday language.

Examples:

What problems do LLMs solve?

Talk about how computers traditionally struggled with understanding human language. LLMs make it possible to communicate with computers naturally instead of using rigid commands.

Popular examples of LLMs

ChatGPT, Claude ,Gemini ,LLaMA.

LLMs in Everyday Life

You may not realize it, but you're probably using LLMs every day.

• Smart Reply and autocomplete in emails and messaging apps

• Customer support chatbots on shopping and banking websites

• Code assistants like GitHub Copilot

• Language translation tools.

2. What Happens When You Send a Message to ChatGPT?

Here's the deep dive into the journey behind the prompt-to-response.

Let's understand this in steps:

Step 1: Your prompt

Whatever you type is prompt. Prompt is simply the input the model gets from the user. It can be an instruction , a paragraph of a story, can be anything.

Step 2 : Processing your message

Ofcourse, the model doesn't get your text in its true form instead it is broken down into tokens and is converted into numbers. The model runs these numbers thorugh several computaions and predicts what comes next.

Also,

No. of Words in a Prompt ≠ No. of Tokens

Step 3: A Response is Generated

Once the model understands your prompt, it begins generating a response.

Interestingly, it doesn't write the entire answer in one go. Instead, it generates one token at a time. After producing each token, it asks itself:

"Based on everything I've seen and generated so far, what's the most likely next token?"

It then adds that token to the response and repeats the process until the answer is complete.

Step 4: Why Responses aren't copied from Internet

The biggest misconception is that ChatGPT searches the internet and copies answers. But that's not how it works.

Instead, ChatGPT has learned patterns from a huge amount of text during its training. When you ask a question, it creates a new response by predicting what should come next based on those patterns.

That's why it can:

• Explain the same topic in different ways.

• Change its answer based on your prompt.

3. Why Computers Don't Understand Human Language

Computers don't understand language. They deal with numbers.

Text vs Numbers

Imagine showing the word "Hello" to a computer. It doesn't recognize it as a greeting or even as a word. To the computer, it's just a series of characters. Before it can make sense of the text, it has to convert it into numbers—the only form it can actually work with.

Every calculation inside an AI model happens with numbers, not words. Before the model can analyze your prompt, the text must first be broken down and represented as numbers.

Converting text into numbers is the foundation of how every LLM works.

Meet the Token

The bridge between words and numbers is the token. They break text into smaller pieces that can be converted into numbers, making it possible for the model to process and understand your input.

4 . Tokenization

Tokenization is the process of transforming text into tokens—the units an AI model can process and understand.

What is a token?

A token is a chunk of text. It can be:

• A complete word → Hello

• Part of a word → (token + ization)

• Punctuation → !

Why is tokenization needed?

Let's say , you are given a 500-page book and asked to read it without any spaces, punctuation, or paragraphs.

Sounds impossible, right? So is for AI .

Breaking text into smaller pieces makes processing much easier. That’s exactly why Tokenization exists. The model first divides everything into manageable pieces before doing any calculations.

Words vs tokens

A word and a token are not always the same. While some words become a single token, others are split into multiple tokens.

Remember:

• The common words are mostly treated as a single token.

• A long or unfamiliar word may become multiple tokens.

• Tokens can be words, parts of words, punctuation, or even emojis.

5. Transformers : The Brain Behind Modern AI

So far, we've converted text into tokens and then into numbers. But how does AI actually understand the relationships between those words and generate meaningful responses?

That's where Transformers come in.

What actually is a Transformer ?

The Transformer is the foundation of modern AI. Since its introduction in 2017, it has become the architecture behind models like ChatGPT, Claude, Gemini, and Llama, enabling them to understand context and generate human-like text.

Ever thought what does GPT in ChatGPT is ?

It is Generative Pre-trained Transformer .

That's right—ChatGPT is built on the Transformer architecture.

Why it changed AI?

Earlier models processed text one word at a time, making it difficult to understand long sentences and relationships between words. Transformers changed this by looking at the entire sentence at once, making it much easier to capture context and meaning. It introduced Self-Attention mechanism.

How it helps understand language?

Self-Attention allows every token in a sentence to look at every other token all at once and decides which one is more relevant.

Example:

"The animal didn't cross the road because it was tired."

When processing the word "it", the model doesn't guess. It looks at all the other words in the sentence and asks:

"Which word is 'it' referring to?"

Since "animal" has the strongest relationship with "tired", the model understands that "it" refers to "the animal", not "the road."

Why almost every modern LLM uses Transformers

• Understands the entire context, not just one word at a time.

• Connects related words, even if they're far apart in a sentence.

• Processes text in parallel, making training much faster.

• Scales to billions of parameters and massive datasets.

• Produces more natural, coherent, and context-aware responses.

Connecting the Pieces

The next time you ask ChatGPT a question, remember what's happening behind the scenes.

• You type a prompt.

• It is broken into tokens.

• The tokens are converted into numbers.

• The Transformer understands the context using self-attention.

• The model predicts one token at a time until the response is complete.

• Finally, the tokens are converted back into text for you to read.

The next time you use ChatGPT, you'll know there's an incredible journey happening behind every response.