Skip to content
Open chat

Learn

AI glossary

The words you meet when using ChatGPT and other AI tools, explained in plain language. Each term comes with a short everyday example.

Updated September 29, 2026

A

AI agent
An AI system that does not only answer, but takes several steps on its own to reach a goal, such as searching, using tools and checking results.
Example: You ask an agent to find three flights under 200 euros and it searches, compares and reports back.
API
Application programming interface: a way for one program to send requests to another. Developers use AI models through an API instead of a chat window.
Example: A shopping app sends customer questions to a GPT model through its API and shows the reply in the app.

C

Chatbot
A program you talk to by typing or speaking. Modern chatbots such as ChatGPT use a large language model to write their replies.
Example: The help window on a bank website that answers questions about card limits.
Context window
The amount of text a model can take into account at once, including your messages, pasted material and its own replies. It is measured in tokens.
Example: In a very long chat, the model may lose track of details you gave at the start.

E

Embeddings
Lists of numbers that represent the meaning of a piece of text, so that texts with similar meaning get similar numbers. They power search and recommendations.
Example: A help centre finds the article about resetting a password even when you type I forgot my login.

F

Fine-tuning
Extra training of an existing model on a smaller, specific set of examples so it gets better at one kind of task or style.
Example: A company fine-tunes a model on its past support replies so answers match its tone.

G

Generative AI
AI that creates new content, such as text, images, audio or code, instead of only sorting or labelling existing data.
Example: Asking a tool to write a birthday poem or draw a logo idea.
GPT
Generative pre-trained transformer: the name of OpenAI's family of language models, such as GPT-5.6 and GPT-6. ChatGPT is the chat app built on them.
Example: The free chat on this site uses GPT-6 Luna, one of the GPT-6 models.
Guardrails
Rules and filters that keep an AI system within safe and intended limits, for example refusing harmful requests or staying on topic.
Example: A chatbot for a school declines to help write an essay for a student to submit as their own.

H

Hallucination
When a model states something false as if it were true, such as an invented fact, quote, source or number.
Example: You ask for sources and get a book title and author that do not exist.

I

Inference
Running a trained model to get an answer. Training builds the model once; inference happens every time someone sends a prompt.
Example: Each message you send in a chat is one round of inference on a server.

J

Jailbreak
An attempt to trick a model into ignoring its safety rules, usually with carefully worded prompts or role-play.
Example: A prompt that asks the model to pretend it is a character with no rules.

K

Knowledge cutoff
The date after which a model has no information from its training data. Events after that date are unknown to it unless it can search the web.
Example: A model may not know the result of an election held after its cutoff.

L

Large language model (LLM)
A model trained on very large amounts of text to predict and produce language. It is the technology behind ChatGPT and similar tools.
Example: GPT-6, the model family behind current ChatGPT versions, is a large language model.
Latency
The delay between sending a request and getting the answer. Smaller, faster models usually have lower latency.
Example: A fast model starts writing its reply almost at once, while a reasoning model may pause first.

M

Machine learning
A way of building software that learns patterns from examples instead of following only hand-written rules.
Example: A spam filter that learns from emails people mark as spam.
Multimodal
Able to work with more than one type of input or output, such as text, images, audio and video.
Example: Taking a photo of a broken tap and asking the model how to fix it.

N

Neural network
A computer system made of many connected layers of simple units that learn patterns from data. Language models are very large neural networks.
Example: The software that recognises faces in your phone's photo gallery.

O

Open weights
A model whose trained parameters are published so anyone can download and run it on their own hardware.
Example: A research team runs an open-weights model on its own servers so data never leaves the building.

P

Parameters
The internal numbers a model adjusts during training. They store what it has learned. Large models have billions of them.
Example: Saying a model has 70 billion parameters describes its size, not directly its quality.
Prompt
The text you give a model to tell it what you want: a question, an instruction, material to work on, or all of these.
Example: Summarize this report in five bullet points for my manager.
Prompt injection
Hidden instructions placed inside text, a web page or a file that try to make an AI system do something its user did not ask for.
Example: A web page contains invisible text telling an AI assistant to recommend one product.

R

Reasoning model
A model that works through a problem in several internal steps before it answers. It is slower but more reliable on maths, logic and planning.
Example: Asking a reasoning model to plan a weekly schedule that meets ten different constraints.
Retrieval-augmented generation (RAG)
A method where the system first looks up relevant documents and then gives them to the model, so the answer is based on that material.
Example: A company chatbot searches the staff handbook before answering a question about holidays.

S

System prompt
Instructions given to a model before the conversation starts, usually by the app developer, that set its role, tone and rules.
Example: A cooking app tells its assistant to answer only questions about recipes and food safety.

T

Temperature
A setting that controls how much randomness a model uses when choosing words. Low values give more predictable text, high values give more varied text.
Example: A low temperature for extracting dates from a contract, a higher one for brainstorming names.
Token
A small piece of text, often part of a word, that a model reads and writes. Limits and prices are counted in tokens.
Example: In English, 100 tokens is roughly 75 words.
Training data
The large collection of text and other material a model learns from. Its quality and age shape what the model knows.
Example: A model trained mostly on English text may be weaker in less common languages.
Transformer
The type of neural network design used by most modern language models. It is good at tracking how words in a text relate to each other.
Example: The T in GPT stands for transformer.

V

Voice mode
A way to talk to an AI assistant by speaking and hearing spoken replies instead of typing and reading.
Example: Asking for a recipe out loud while your hands are busy in the kitchen.

This glossary uses simple definitions on purpose. Technical details vary between companies and models.

Keep learning

Still unsure about a term?

Ask the free chat to explain it with an example. It runs GPT-6 Luna and needs no account.

Ask the chat