Agent
A program that does more than answer a single question: it breaks a task into steps and carries it out using tools such as search, files or a calculator.
Glossary
Artificial intelligence terms with simple explanations, in alphabetical order.
A
AI ethics
Principles for using systems fairly, transparently and responsibly: questions of bias, privacy and accountability.
Algorithm
A set of precise, ordered steps for solving a particular problem.
API
A defined way for programs to talk to each other. It is how AI services are connected to your own application.
Artificial intelligence
The field of computer systems that perform tasks that normally require human intelligence, such as understanding text, recognizing images and translating.
B
Bias
A systematic skew in results against certain groups, caused by the data or the model.
C
Chatbot
A program that converses with a user through text or voice.
Context window
The amount of text (your request plus the conversation history) a model can take into account at once. Anything that falls outside is “forgotten”.
D
Dataset
A collection of data gathered for training or testing a model.
Debugging
The process of finding the cause of an error in a program and fixing it.
Deep learning
A kind of machine learning that uses multi-layer neural networks.
Deepfake
A fake image, audio or video created with artificial intelligence that imitates the face or voice of a real person.
Dictionary
A Python structure that stores key and value pairs: a value is read through its key.
E
Embedding (vector representation)
A way of expressing the meaning of a word, sentence or image as a list of numbers. Things with similar meaning get similar numbers.
F
f-string
A Python way to put a variable’s value into text through {}, with an f before the quotation marks.
Few-shot prompting
Showing the model a few examples before giving it the task. Without examples, it is called zero-shot.
Fine-tuning
Training a ready-made model further on a smaller, specialized dataset to adapt it to a particular task.
Function
A named piece of code: written once, called many times; it can take data (parameters) and return a result.
G
Generative AI
Artificial intelligence that creates new text, images, audio or code.
H
Hallucination
Incorrect or invented information that a model states in a confident tone. This is why important facts need checking.
I
Inference
The process in which a trained model answers a new request.
J
JSON
A format that writes data as text: key and value pairs and lists; widely used to exchange data between programs.
L
Large language model (LLM)
A model trained on a very large amount of text that can understand and generate text. Most chat tools are built on it.
List
A Python structure that stores several values in order; items are numbered from 0.
Loop
A construct that repeats the same action several times; in Python these are for and while.
M
Machine learning
An approach in which a program learns patterns from data instead of following rules written by hand.
Model
The mathematical structure produced by training that answers requests, together with its tuned parameters.
Multimodal
A model that can work with several types of data at once, such as text, images and audio.
N
Neural network
A type of model made of many simple computing units, inspired by the connections in the human brain.
O
Open model
A model whose weights (parameters) are published for public use. License terms differ from model to model.
P
Parameter
An internal value of a model that is tuned during training. A model’s “size” is often expressed as its number of parameters.
Prompt
The request or instruction text you give to a model.
Prompt template
A ready-made request text with blanks for the changing parts: each time only the needed values are filled in.
Python
A programming language known for its simple syntax; widely used in data analysis and artificial intelligence.
R
RAG (retrieval-augmented generation)
A method where, before answering, the model finds relevant text in an external source (documents, databases) and builds its answer on it.
Role prompting
Telling the model in the prompt which role to answer in, for example “explain this as a teacher”.
S
Spec
A precise description of what a program should do: input, output, examples and edge cases.
System prompt
A standing instruction given to the model at the start of a conversation that sets its behavior and style.
T
Temperature
A setting that controls how random or predictable the model’s answer is. A low value gives more consistent answers.
Token
The unit a model processes: a word, part of a word or a character. Text length is measured in tokens.
Tool call (function calling)
When a model, instead of answer text, names a function (a tool) and its arguments; the program runs it and returns the result to the model.
Training
The process of tuning a model on data so that it learns to perform a task.
V
Variable
A name attached to a value: a program stores data under this name and then reuses it.
Vibe coding
A way of building a program by talking with artificial intelligence: the tool proposes code, while checking and testing stay with the programmer.
W
WebGPU
A modern web technology that lets a page use the power of the graphics card (GPU) through the browser; it is used, among other things, to run some models right in the browser.