What AI and models are

"Artificial intelligence" is an old and broad term. It has meant chess programs, spam filters, self-driving cars, and much more. When people say "AI" in 2026, they almost always mean one specific thing: a model that learned patterns from a large amount of data and uses them to produce useful output, such as text, images, or code.

That is a narrower idea than the science-fiction version. A model like this is not conscious, does not have goals of its own, and does not "understand" in the human sense. It is a very capable pattern machine. Keeping that plain picture in mind will make everything else easier.

Two families of AI

It helps to split the field in two. Traditional (rule-based) AI follows logic a person wrote by hand: if this, then that. Machine learning flips that around. Instead of writing the rules, you show the system many examples and let it learn the patterns itself. Almost everything called "AI" today is machine learning.

The rest of this course is about the kind of model behind chatbots and coding assistants: the large language model. But the same core ideas apply to models that generate images or audio.

Under the hood, a model is a large collection of numbers. These numbers are called parameters, or weights. During training, the system adjusts these numbers over and over until its output on the training examples gets good. Once training stops, the weights are frozen and that fixed set of numbers is the model.

You will see models described by parameter count: "7B" means seven billion parameters, "70B" means seventy billion. Bigger models can capture more patterns, but size is not the whole story. A well-trained smaller model often beats a poorly trained larger one, and a larger model costs more memory and compute to run.

Why parameter count matters to you

• Memory: parameters have to fit in RAM or video memory to run. More parameters means more memory.
• Speed: larger models are slower to respond on the same hardware.
• Quality: more parameters can mean better answers, but training quality and data matter just as much.

A large language model (LLM) is a model trained on a very large amount of text, from books, code, and the public web, with one core objective: given some text, predict what comes next. That sounds almost too simple to be useful. It turns out that doing it extremely well requires the model to pick up grammar, facts, reasoning patterns, writing styles, and a great deal more.

Everything a chatbot appears to do, answering questions, drafting an email, writing code, translating, is built on that one skill: continue the text in a plausible, helpful way. When you ask a question, the model is continuing the conversation with the most likely helpful response.

Why "large"

The "large" refers to two things: the number of parameters (billions, from the last lesson) and the size of the training data (a large slice of human writing). Scale is what took these models from producing gibberish to producing text that is often genuinely useful.

A model does not see text the way you do. Before anything happens, your text is broken into tokens: small chunks that are often a word, part of a word, or a piece of punctuation. The model reads and writes in tokens, not letters or whole words.

A rough rule of thumb for English: one token is about four characters, and 100 tokens is roughly 75 words. Common words are usually a single token; rare or long words get split into several.

Why tokens matter in practice

• Length limits are measured in tokens, not words. The "context window" you will meet later is a token budget.
• Cloud APIs bill per token, for both what you send and what you get back. Longer prompts and longer answers cost more.
• Different models split text differently, so the same sentence can be a different number of tokens on different models.