AI chatbots and writing assistants are now built into email, office software, search engines and phones. Almost all of them are powered by a large language model, or LLM. You don't need a maths degree to understand the basics — and understanding them helps you use these tools well and avoid their pitfalls.
The one-sentence version
A large language model is a program trained on enormous amounts of text to predict the next piece of text, which it does so well that it can answer questions, summarise documents, write code and hold a conversation.
Tokens: how a model reads text
Models do not read words the way we do. Text is split into tokens — whole words, parts of words or punctuation. A common word like network may be one token; an unusual word like deduplication may be split into several.
Everything about a model is measured in tokens: how much it can read at once, how fast it responds and, for paid services, how much it costs.
Training: learning from examples
Building a model happens in stages:
- Pre-training. The model reads a vast collection of text — books, websites, code and more — and repeatedly tries to predict the next token. Each time it is wrong, its internal settings (called parameters) are adjusted slightly. Over billions of examples, it learns grammar, facts, reasoning patterns and writing styles.
- Fine-tuning. The model is trained further on examples of helpful conversations, so it learns to follow instructions rather than just continue text.
- Feedback. People rate the model's responses, and the model is adjusted to prefer the better ones. This is often called reinforcement learning from human feedback.
The "large" in large language model refers to the number of parameters and the amount of training data, both of which are huge.
Generating an answer
When you ask a question, the model:
- Converts your message into tokens.
- Predicts the most suitable next token.
- Adds that token to the text and predicts the next one.
- Repeats until the answer is complete.
That is why responses appear word by word. The model is not looking up a stored answer — it is generating new text each time, which is also why the same question can produce different answers.
The context window
A model can only consider a limited amount of text at once: your messages, its replies and any documents you provide. This is the context window. In a long conversation, early details can fall outside it and be "forgotten". Modern models have large context windows, but if an answer ignores something you said earlier, restating it usually helps.
Hallucinations: confident and wrong
Because a model generates plausible text rather than retrieving verified facts, it can produce statements that sound authoritative but are false. These are called hallucinations. Common examples include invented references, incorrect command syntax and made-up product features.
Many AI tools now reduce this by searching the web or company documents and citing sources, but the risk never disappears entirely.
Using LLMs well
- Be specific. Say who the answer is for, what format you want and what to include or avoid.
- Provide context. Paste the error message, the relevant config or the document you are asking about.
- Ask for reasoning or sources when accuracy matters.
- Iterate. Treat the first answer as a draft and refine it.
Privacy and data
What you type into an AI tool may be stored, reviewed or — depending on the service and your settings — used to improve future models. Before pasting in customer data, passwords, internal documents or source code, check your organisation's policy and the provider's data terms. Business and enterprise plans usually offer stronger guarantees than free consumer versions.
The bottom line
Large language models are powerful pattern-matching systems for language. They are excellent at drafting, summarising, explaining and transforming text — and they still need a knowledgeable human to check their work.