Lesson 6 of 28 · Language models and chat assistants

What is a large language model (LLM)?

A large language model (LLM) is an AI model trained on huge amounts of text to predict which word fragment, or token, comes next. Repeating that prediction lets it write answers, summaries, translations and code. LLMs are the engine inside chat assistants such as ChatGPT, Claude and DeepSeek, and inside many writing, search and coding tools.

How an LLM writes an answer

When you send a message, the text is first split into tokens, small chunks such as words or parts of words. The model then calculates, for every token in its vocabulary, how likely it is to come next. One token is chosen and added to the text, and the whole process repeats until the answer is finished.

It is like the word suggestions on your phone keyboard, scaled up enormously: trained on far more text, able to take thousands of words of context into account, and good enough at the patterns of language to follow instructions, keep a consistent style and write working code.

How LLMs are trained

Almost all LLMs are built on the transformer, a neural-network design introduced in 2017 that is good at relating every word in a passage to every other word. Turning one into a chat-ready assistant usually happens in stages:

Pre-training: the model reads a very large collection of text and repeatedly predicts the next token, adjusting its parameters (the internal numbers it learns) to reduce its mistakes.
Instruction tuning: it is further trained on examples of requests paired with good answers, so it learns to follow instructions rather than just continue text.
Human feedback: people compare and rank answers, and the model is tuned towards the preferred ones. Ouyang et al. described this approach, reinforcement learning from human feedback (RLHF), for instruction-following models in 2022.

What makes it "large"

"Large" refers both to the training data and to the number of parameters. The GPT-3 paper (Brown et al., 2020), for example, described a model with 175 billion parameters and showed that scale let a model pick up new tasks from just a few examples placed in the prompt.

Bigger models tend to be more capable but slower and more expensive to run. Smaller models can run on a laptop or phone and are often good enough for focused tasks such as summarising or classifying text.

What an LLM is not

An LLM is not a database or a search engine. It does not store facts in a lookup table; it generates text that fits the patterns it learned, so it can state something false with full confidence, a failure known as hallucination. Its knowledge also stops at a training cutoff unless the product connects it to search or documents.

It also does not remember you. Between conversations the model itself stays the same; any "memory" is a feature of the app around it, which saves notes and feeds them back in. And despite its fluent language, it has no beliefs or intentions of its own.

Where you meet LLMs

LLMs power chat assistants such as ChatGPT, Claude and DeepSeek, and sit inside many products that do not call themselves chatbots: email and document editors, customer-support tools, search engines, coding assistants and AI agents. Many products use another company's model through an API, a connection that lets one piece of software send prompts to a model and receive its answers.

When choosing, look at which model a product uses, its context window, its price per token and its data-use terms. Mindber's Arena leaderboard and LLM cost calculator help you weigh capability against cost.

Example

Next-token prediction, one step at a time

Worked example

Text so far: "The capital of France is" Model's top guesses for the next token (illustrative): " Paris" 92% " a" 3% " located" 2% → picks " Paris" and appends it → predicts again: "." → the reply is complete

The percentages are made up for illustration. A real model scores every token in its vocabulary at each step, and settings such as temperature decide how often a less likely token is chosen.

Key points

An LLM predicts the next token, over and over, to produce text.
It is pre-trained on huge text collections, then tuned to follow instructions.
"Large" refers to the amount of training data and the number of parameters.
It does not look facts up unless a tool connects it to search or documents.
The model inside a product shapes its quality, speed and cost.

Try it on Mindber

AI model leaderboard (Arena)

Estimate costs with the LLM cost calculator

See DeepSeek's profile

Next lesson: What is ChatGPT?

Frequently asked questions

What does LLM stand for?

LLM stands for large language model. "Large" refers to the size of the training data and the number of parameters, "language" to the fact that it works with text, and "model" to the trained mathematical system that turns an input into an output.

Is ChatGPT an LLM?

ChatGPT is a chat product built on top of OpenAI's large language models. The LLM is the engine; ChatGPT adds the chat interface, conversation history, file uploads, web search and other features around it. The same distinction applies to most AI assistants.

Can I run an LLM on my own computer?

Yes. Smaller open-weight models can run on a capable laptop or desktop with free tools, and some run on phones. Larger models need powerful graphics cards or a cloud server. Running locally keeps your data on your machine but usually means a less capable model than the largest hosted ones.