LLM stands for large language model: the kind of AI that reads and writes text, and the engine inside chatbots like ChatGPT, Claude and Gemini. It learned the patterns of language from a huge amount of writing, and it answers by predicting what text should come next.
Think of the word suggestions above your phone keyboard, which guess your next word. An LLM is that idea grown enormously: trained on so much writing that it can keep guessing well enough to produce whole emails, essays and computer code.
In a sentence: “The help chat on the website is powered by an LLM, which is why it can answer questions in full sentences.”
What the word means
The plain point: each part of the name tells you something.
- Large: it was trained on a vast amount of text, and the model itself is huge. IBM describes training data running to “billions or trillions of words”.1
- Language: its main job is words, though many now handle pictures and sound too.
- Model: a file of learned patterns that a computer runs. It is not a database of facts.
Google’s training course for developers describes a language model as something that estimates which piece of text is likely to come next.2 Those pieces of text are called tokens.
A simple example
The plain point: it builds the answer piece by piece; it does not look up a finished one.
Ask “What is the capital of France?” The model picks the most fitting next word, then the next, until the reply is done. Anthropic, which makes Claude, says this plainly: “Claude writes text one word at a time.”3
Nobody typed in the answers. The model learned them as patterns during training.3
An LLM is a pattern expert, not a fact database. That explains both why it writes so well and why it can be confidently wrong.
How is an LLM trained?
The plain point: it reads a huge amount of text, practises predicting the next piece over and over, and is then tuned with help from people.
- Read huge amounts of text. Books, articles, websites and code, running to billions or trillions of words.1
- Predict the next token, again and again. The model guesses the next piece of text, checks the real one, and its internal numbers (weights) are adjusted a little each time. IBM calls LLMs giant prediction machines that repeatedly predict the next word.4 Almost all of today’s LLMs are built on a design called the transformer.5
- Tune it with people. Extra training on smaller, hand-picked examples (fine-tuning), then people rank its answers so it learns to prefer the better ones. That ranking step is called RLHF.4
The whole build, from chips to apps, is in how AI is made, step by step.
What LLMs can’t do
The plain point: an LLM is a powerful writer with three built-in blind spots.
- It can make things up. It can state something false in a fluent, confident way, called a hallucination.4
- Its knowledge stops at a date. It learned from text collected up to a point, its knowledge cutoff, and knows little after that unless the app gives it a search tool. Anthropic, for one, publishes a “reliable knowledge cutoff” for each Claude model.6
- It can only hold so much at once. Everything in a conversation has to fit in its context window, and accuracy drops as that fills up.7
Examples of LLMs
- GPT, from OpenAI: the models behind ChatGPT. OpenAI’s researchers say ChatGPT is based on an LLM.8
- Claude, from Anthropic.6
- Gemini, from Google DeepMind: built to handle pictures, audio and video as well as text.9
- Llama, from Meta: Meta publishes guides for setting it up and hosting it yourself.1011
Names and versions change often; follow the links for each maker’s current line-up.
Why it matters to you
The plain point: knowing what an LLM is tells you how to use it.
- Great for: drafting, rewriting, summarising, explaining, brainstorming and code.
- Check it on: names, numbers, dates and quotes. Pattern-based answers can be wrong in a convincing way. See hallucination meaning.
- It may not know recent events unless the app gives it a search tool, because its training stopped at a certain date.
Often confused with
- Chatbot: the app you talk to. The LLM is the engine inside it; the chatbot adds the chat window, memory, search and safety rules.
- Generative AI: the wider family of AI that creates things, including pictures, video and music. LLMs are the text members of that family. See generative AI meaning.
- AI agent: an LLM given tools and allowed to take actions. See AI agent meaning.
- Foundation model: a big general-purpose model that other products are built on. Many LLMs are described this way: Meta calls Llama 3 “a new set of foundation models” (source in the examples above). See foundation model meaning.
- Small language model: the same idea built at a much smaller size. See small language model meaning.
Questions people ask
Is ChatGPT an LLM?
Not exactly: ChatGPT is the chatbot app, and an LLM is the engine inside it. OpenAI’s researchers put it this way: ChatGPT is based on an LLM, and has used several different ones since launch.8
What does “large” mean in large language model?
Two things: the amount of text it learned from, running to billions or trillions of words, and the size of the model itself, with billions of adjustable numbers called parameters.4 A much smaller cousin is a small language model.
What is the difference between an LLM and AI?
AI is the whole field; an LLM is one kind of AI, a deep learning model trained on huge amounts of text to understand and write language. Every LLM is AI, but most AI (spam filters, maps, face unlock) is not an LLM.4
Learn more on this site
- How one is made: How AI is made, step by step.
- Where LLMs sit among other AI: Understanding AI model types.
- The big question: Does AI understand anything?
- Running one on your own computer: Run AI on your own computer.
- The basics: the AI glossary and AI explained for complete beginners.
- Related: How Does ChatGPT Work? What Happens When You Press Send.
- Related: What Is a Vision-Language Model? VLM Meaning in Plain English.
- Related: What Is a World Model in AI? World Model Meaning in Plain English.
- Related: LLM vs Reasoning Model.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. How does an LLM build its answer?
- It looks up a finished answer in a database
- It copies the answer from a website
- It predicts the next piece of text, again and again
Show the answer
Answer: It predicts the next piece of text, again and again. An LLM picks the most fitting next word, then the next, until the reply is done; it does not look up a finished one.
2. Is ChatGPT itself an LLM?
- Not exactly: ChatGPT is the chatbot app, and an LLM is the engine inside it
- Yes, the app and the model are the same thing
- No, ChatGPT does not use any language model
Show the answer
Answer: Not exactly: ChatGPT is the chatbot app, and an LLM is the engine inside it. The chatbot adds the chat window, memory, search and safety rules around the LLM.
3. Why should you check names, numbers and dates an LLM gives you?
- LLMs are not allowed to use numbers
- It is a pattern expert, not a fact database, so it can be confidently wrong
- LLMs always round numbers up
Show the answer
Answer: It is a pattern expert, not a fact database, so it can be confidently wrong. Pattern-based answers can be wrong in a convincing way, which is called a hallucination.
Related terms: Generative AI · GPT · Hallucination · Machine learning · Narrow AI (weak AI) · Natural language processing (NLP)
- 1IBM, What Are Large Language Models (LLMs)?, 6 Oct 2021, read at source 1 Oct 2026“billions or trillions of words from books, articles, websites, code and other text sources.” ↑a“billions or trillions of words from books, articles, websites, code and other text sources.” ↑b
- 2Google, Introduction to Large Language Models, updated 9 Jan 2026, read at source 1 Oct 2026“A language model estimates the probability of a token or sequence of tokens occurring within a longer sequence of tokens.” ↑
- 3Anthropic, Tracing the thoughts of a large language model, 27 Mar 2025, read at source 1 Oct 2026“Claude writes text one word at a time.” ↑a“Language models like Claude aren’t programmed directly by humans—instead, they’re trained on large amounts of data.” ↑b
- 4IBM, What Are Large Language Models (LLMs)?, 6 Oct 2021, read at source 5 Oct 2026“LLMs work as giant statistical prediction machines that repeatedly predict the next word in a sequence.” … “Through an iterative cycle of making predictions and then updating model weights through backpropagation and gradient descent, the model ‘learns’”. ↑a“Fine-tuning most often happens in a supervised context with a much smaller, labelled dataset.” … “RLHF is a form of fine-tuning where humans rank model outputs and the model is trained to prefer outputs that humans rank higher.” ↑b“During hallucinations, the model generates information that is false or misleading while sounding plausible.” ↑c“The number of parameters refers to how many of these variables exist in a model, with some LLMs containing billions of parameters.” ↑d“Large language models (LLMs) are a category of deep learning models trained on immense amounts of data, making them capable of understanding and generating natural language.” ↑e
- 5IBM, What is a transformer model?, published 28 Mar 2025, as cited on this site’s transformer page, read at source 1 Oct 2026“The transformer model is a type of neural network architecture that excels at processing sequential data, most prominently associated with large language models (LLMs).” ↑
- 6Anthropic, Models overview, undated documentation, read at source 5 Oct 2026“Reliable knowledge cutoff: The date through which the model’s knowledge is most extensive and reliable.” ↑a“Claude is a family of state-of-the-art large language models developed by Anthropic.” ↑b
- 7Anthropic, Context windows, undated documentation, as cited on this site’s context window page, read at source 1 Oct 2026“The ‘context window’ refers to all the text a language model can reference when generating a response, including the response itself.” … “As token count grows, accuracy and recall degrade, a phenomenon known as context rot.” ↑
- 8OpenAI (Chatterji et al.), How People Use ChatGPT, 15 Sep 2025, read at source 5 Oct 2026“ChatGPT is based on a Large Language Model (LLM), a type of Artificial Intelligence (AI)”. ↑a“ChatGPT is based on a Large Language Model (LLM)” … “Since its launch ChatGPT has used a variety of different underlying LLMs”. ↑b
- 9Gemini Team (Google), Gemini: A Family of Highly Capable Multimodal Models, arXiv:2312.11805, 19 Dec 2023, read at source 5 Oct 2026“This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding.” ↑
- 10Grattafiori et al. (Meta), The Llama 3 Herd of Models, arXiv:2407.21783, 31 Jul 2024, read at source 5 Oct 2026“This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage.” ↑
- 11Meta, Get started with Llama, undated documentation, read at source 5 Oct 2026“This guide provides information and resources to help you set up Llama including how to access the model, hosting, how-to and integration guides.” ↑
The comparison on this page (phone keyboard suggestions) is this site’s own way of explaining, not a claim from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK