Every AI chat assistant you’ve heard of is powered by a large language model, or LLM. You don’t need a technical background to understand the basics, and knowing them makes you much better at using these tools.

The one-sentence version

A large language model is a computer program trained on huge amounts of text to predict what words should come next, and that simple skill, done at enormous scale, lets it write, summarize, translate and answer questions.

How it learns

During training, the model reads vast amounts of text and repeatedly tries to predict the next piece of text, adjusting itself every time it’s wrong. Over time it picks up grammar, facts, writing styles and patterns of reasoning. Many models then get further training from human feedback, so they follow instructions and respond more helpfully and safely.

How it answers you

When you send a message, the model breaks it into tokens, considers everything in its context window and generates a reply one token at a time, each choice based on everything before it. It isn’t looking up a stored answer; it’s producing a new one.

Key terms, decoded

• Parameters: the internal settings the model adjusts while learning. More parameters generally means more capacity, though training quality matters just as much.
• Training data: the text the model learned from. It has a cutoff date, so the model may not know recent events unless it can search the web.
• Prompt: what you send the model, including instructions and context.
• Temperature: a setting that controls how predictable or varied the output is.
• Multimodal: a model that can also work with images, audio or files, not just text.

What LLMs are good at

• Drafting and rewriting text in different styles.
• Summarizing and explaining material you provide.
• Brainstorming, outlining and organizing ideas.
• Writing and explaining code.
• Translating between languages.

Where they struggle

• Facts they weren’t given. They can produce confident but false answers. See AI hallucinations.
• Recent events after their training cutoff, unless connected to search.
• Exact arithmetic on long calculations, unless they can use a calculator or code tool.
• Your private context: they only know what you tell them in the conversation.

Key takeaways

• LLMs generate text by predicting what comes next, learned from huge amounts of text.
• They are excellent at language tasks and unreliable as a sole source of facts.
• The more relevant context you give them, the better they perform.