How it works
An LLM works with tokens, which are words or pieces of words. During training it reads enormous amounts of text and adjusts a very large number of internal values, called parameters, so that it gets better at predicting the next token. Nobody writes rules for grammar or facts. The ability to answer a question or write an email comes out of that one prediction task.
When you send a prompt, the model produces a reply one token at a time. Everything it can see at once, meaning your prompt, the conversation so far and any attached files, must fit in its context window. It has no memory beyond that unless the app adds one. It also has no built-in way to check facts, which is why it can state something false in a confident tone. That failure is called a hallucination.
Creators meet LLMs through assistants such as ChatGPT, Claude and Gemini, and through an API when they build a model into their own tool or workflow. Model names, prices and limits change every few months, so compare them on the vendor's current pages, not from an old article.
A worked example
You paste a 30-minute video transcript into an assistant and ask for three short post ideas.
- The app splits your text into tokens and sends them to the model with your instruction.
- The model writes the reply token by token, each one chosen from what is likely to follow.
- You get three ideas in seconds. One of them quotes a line that is not in the transcript.
- You check each quote against the transcript before posting. The model wrote plausible text, not verified text.
An illustration, not a real account. Any numbers in it are made up to show the method, and are not benchmarks.
Common mistakes
- Treating an LLM as a search engine. It generates likely text. Unless the app runs a search and shows its sources, it is answering from memory.
- Assuming it remembers earlier chats. A model only sees what is in the current context window, plus whatever memory feature the app adds.
- Mixing up the model and the product. ChatGPT, Claude and Gemini are products. Each runs on a family of models that the vendor replaces over time.
Questions people ask
Is an LLM the same thing as AI?
No. AI is the broad field. An LLM is one kind of AI system, built for language. Image generators and recommendation systems are AI too, and they work differently.
Why do LLMs make things up?
Because they are built to produce likely text, not to look facts up. When the training data is thin or the question is unusual, the most likely sounding answer can be wrong. Giving the model source text to work from, as RAG does, reduces this.
Which LLM is best?
It depends on the task, and the ranking changes with each release. Test two or three on your own real prompts and compare quality, speed and cost.
Sources
Google defines a large language model as, at a minimum, a language model having a very high number of parameters, and a hallucination as plausible-seeming but factually incorrect output from a generative AI model.
Google for Developers: Machine Learning Glossary: Generative AI
Every link was opened on 13 October 2026. Products and policies change, so check the source before you rely on a figure.