What it means
An LLM, or large language model, is a computer program trained on an enormous amount of text so that it can read and write language.
It’s the engine inside chatbots such as ChatGPT, Claude and Gemini. You type a message, and the model builds a reply one small piece at a time. Each piece is a prediction of what should come next. Those pieces are called tokens, and they’re roughly word-sized.
The closest everyday comparison is the predictive text on your phone. Your keyboard suggests the next word based on what people usually type. An LLM does a similar job, but it has learned from a vast mix of books, websites and code, so it can keep going for pages and follow instructions. At scale, that’s enough to draft emails, summarise reports and write code.
“Large” refers to size in two senses. The model has a huge number of internal settings, often billions, called parameters. And the text it learns from is vast. Nobody writes the rules by hand. The model works them out from examples, which is also why its mistakes can be hard to predict.
Where it came from
“Language model” is an old phrase in computing. For decades it described any system that estimated how likely a word was to follow the words before it. Speech recognition and machine translation relied on these long before chatbots made them famous.
The big shift came in 2017, when researchers at Google published a new neural network design called the transformer. It handled long stretches of text far better than earlier methods. Models built on it, including OpenAI’s GPT series and Google’s BERT, appeared from 2018. The usual account is that “large language model” settled into research vocabulary over the next few years, as each model was trained on more data than the last.
The term went mainstream after OpenAI released ChatGPT in late 2022. Journalists suddenly needed a name for the category rather than the brand, and “LLM” filled the gap. By 2023 it was turning up in job adverts and LinkedIn posts. It now anchors a whole vocabulary of newer terms, collected in our AI jargon glossary.
Here’s how it compares with nearby terms.
| Term | What it refers to |
|---|---|
| AI | The broad field of getting computers to do tasks that seem to need intelligence |
| Machine learning | The part of AI where systems learn from data instead of fixed rules |
| LLM | A machine learning model trained on text to read and produce language |
| Chatbot | The app or window you actually talk to, which may run on an LLM |
| GPT | OpenAI’s family of models, short for generative pre-trained transformer |
How people actually use it
- “Which LLM is this running on?” The speaker wants to know whose model sits under the branding.
- “We’re adding LLM features to the dashboard next quarter.” Work chat. Usually means a chat box or automatic summaries.
- “Don’t paste client data into a public LLM.” A standard line in company IT policies.
- “New LLM tops rivals on coding tests.” A typical headline after a model launch.
- “LLMs are just fancy autocomplete.” A sceptic’s view that the hype has outrun the technology.
- “I asked three LLMs and got three different answers.” A common complaint, and a fair reminder to check things.
In a sentence
Maya: is ChatGPT an LLM or is that the company
Leo: ChatGPT is the app. the LLM is the model inside it
Maya: and OpenAI is the company
Leo: yes. three different things
Dad: does the LLM know about the match last night
Jess: only if it can search the web. otherwise it only knows what it was trained on
Common misconceptions
- An LLM isn’t the same thing as AI. It’s one kind. Spam filters and face recognition are AI too, but not language models.
- It doesn’t look things up by default. A plain LLM answers from patterns learned in training. Some products add web search or document lookup, often through a technique called RAG.
- Fluent doesn’t mean correct. An LLM can produce smooth, confident text that is simply false. This is called hallucination, and it’s the main reason to check anything important.
- It isn’t learning from your chat as you type. The model’s knowledge is fixed when training ends. Some apps store notes about you separately, but the model itself isn’t rewiring mid-conversation.
Related terms
- Token: the small chunk of text an LLM reads and writes, often part of a word.
- Parameters: the internal numbers adjusted during training. More usually means a bigger model.
- Context window: how much text a model can take into account at once.
- Prompt: the instruction or question you give the model.
- Foundation model: a large general-purpose model that other products are built on.
Questions people ask
What does LLM stand for?
LLM stands for large language model, a type of AI trained on vast amounts of text so it can process and produce language.
Is ChatGPT an LLM?
ChatGPT is an app that runs on large language models made by OpenAI. Calling ChatGPT itself an LLM is loose but widely understood.
How is an LLM different from a search engine?
A search engine finds existing pages and sends you to them. An LLM writes a new answer from learned patterns, which is often more convenient but less reliable on facts.
Do LLMs actually understand what they say?
Researchers genuinely disagree about whether LLMs understand anything. What’s clear is that they work by predicting text, and they can be impressive at some tasks while oddly unreliable at others.
Why do people say LLMs are bad at maths?
Early LLMs were shaky at arithmetic because they treat numbers as chunks of text rather than calculating. Newer models are much better, and many AI tools hand sums to separate software.
The short version
An LLM is a large language model: software trained on huge amounts of text to predict what comes next, which lets it write, summarise and chat. It’s the technology inside ChatGPT, Claude, Gemini and most AI assistants. It’s useful, but it isn’t a fact database, so check anything that matters.