Chatbots feel conversational, but the core operation of a language model is simple to state: given the text so far, estimate which piece of text is likely to come next.
Tokens, not words
Text is split into small units called tokens, which may be whole words, parts of words, or punctuation. The model works with tokens because they make a manageable vocabulary.
One step at a time
For each step, the model produces a probability for every token in its vocabulary. A token is chosen, added to the text, and the process repeats. A long answer is just many such steps in a row.
Where the knowledge comes from
During training, the model reads huge amounts of text and learns to predict missing continuations. To predict well, it has to absorb grammar, facts, writing styles, and common reasoning patterns, at least in statistical form.
Why settings change the answer
The model gives probabilities, not a single fixed reply. A setting often called temperature controls how adventurous the choice is. Low values favor the most likely token and produce steadier answers. Higher values allow more variety.
What it means for you
A language model is a very capable text predictor. That explains why it is fluent and flexible, and also why it can sound confident without being correct.
Continue with the book
Explore the full treatment, worked examples, and exercises in AI Made Simple.
Adapted from AI Made Simple, published by Choral Software. Article text and original examples © 2026 Choral Software. All rights reserved.