START HERE · HOW AI WORKS
Chatbots like ChatGPT, Gemini, and Claude run on something called a large language model. Click each step to see what happens between your question and its answer.
HOW IT WORKS · CLICK A STEP TO LEARN MORE
BEFORE YOU ASK
Long before your question, the model was trained on an enormous amount of text: books, websites, articles, and more. It didn’t store it all like a library. It learned patterns: how words, ideas, and facts tend to fit together.
It’s a bit like: Someone who has read millions of books and absorbed how language works, without keeping a copy of every page.
STEP 1 OF 5
You type or say a question, a request, or an idea. Everything you include, like details, examples, or the tone you want, is what the AI has to work with.
It’s a bit like: Giving instructions to a helpful new coworker. The clearer the ask, the better the help.
Learn the prompt formula →STEP 2 OF 5
The AI doesn’t read the way we do. It splits your message into small chunks called tokens: whole words, pieces of longer words, even punctuation. Each token is turned into numbers the model can work with.
How␣do␣I␣water␣a␣cactus?
An example of how one question might be split up.
It’s a bit like: Scrabble tiles. The AI works with the tiles, not the whole sentence at once.
STEP 3 OF 5
Before answering, it takes in your newest message plus everything earlier in the chat, along with any files or photos you shared. There’s a limit to how much it can hold at once.
It’s a bit like: A desk. Whatever fits on the desk, it can see. In a very long chat, the oldest pages can slide off the edge.
STEP 4 OF 5
Now the large language model gets to work. Using the patterns from training, it predicts the most likely next token, adds it, then predicts the next one, over and over, until the answer is done. That’s why answers appear word by word.
It’s a bit like: Your phone’s autocomplete, but trained on far more text and able to keep going for pages.
STEP 5 OF 5
The finished answer lands in your chat, and it’s often genuinely helpful. Because it’s built from patterns rather than looked up, it can occasionally sound confident and still be wrong. Your judgment is the last step.
It’s a bit like: A first draft from a smart assistant. You decide what to keep, what to change, and what to double-check.
KEY TERMS
Click a term to see what it means in plain English.
Computer systems that handle tasks we usually think of as needing human thinking, like understanding language, recognizing images, or making recommendations.
AI that creates something new, like text, images, music, or video, instead of only sorting or labeling what already exists.
The app you talk to, like ChatGPT, Gemini, Copilot, or Claude. If the model is the engine, the chatbot is the car.
The trained system behind an AI tool. Companies release new models over time, and one app may let you choose between several.
A model trained on huge amounts of text to understand and write language. It powers most AI chatbots.
The message you send to an AI: a question, a task, or background information.
A small chunk of text, often about three-quarters of a word. It’s how AI breaks language into pieces.
How much of the conversation the AI can consider at once. Very long chats can push older parts out.
The learning phase where a model studies huge amounts of data to pick up patterns, before anyone uses it.
When an AI states something false as if it were true. It isn’t lying; it produced something that fit the pattern but not the facts.
A feature in some apps that remembers details about you across chats. You can usually view, edit, or turn it off in settings.
Keeping a person involved to review, guide, and decide what to do with an AI’s output.
MORE FROM START HERE