Large language models (LLMs) power today's AI chat assistants. You don't need a computer science degree to understand the basic idea.
Learning from text
An LLM is trained on enormous amounts of text. During training, it repeatedly tries to predict the next piece of text and adjusts itself when it's wrong. Over time it learns patterns of language, facts, reasoning styles and how ideas connect.
Predicting the next word
When you ask a question, the model generates an answer one token (a word piece) at a time, each time choosing a likely continuation based on everything before it. That's why it can write fluently in many styles.
Fine-tuning to be helpful
After initial training, models are further trained with examples and human feedback to follow instructions, be helpful and avoid harmful outputs.
Why it sometimes gets things wrong
The model generates what sounds plausible, which usually overlaps with what's true, but not always. It can produce confident errors, especially about obscure or recent facts.
Tools and search
Many assistants can now search the web, run code or read your files, which helps them give more accurate, up-to-date answers.
What this means for you
Give clear context, ask for sources, and verify important facts.
See also: AI glossary.