AI terms in plain words
Short definitions of the terms and abbreviations we use. In our articles a dotted underline marks a term: point at it, tab to it or tap it to see what it means.
A
- AGIArtificial general intelligence
- A hypothetical AI that could match people across most intellectual tasks. There is no agreed definition of it, and no agreed test.
- AI agent
- An AI system that plans and carries out several steps on its own, such as searching, running code or using apps, instead of only answering a question.
- APIApplication programming interface
- A defined way for one program to talk to another. AI companies sell access to their models through APIs, usually priced per token.
B
- Benchmark
- A standard test used to compare AI models, such as a set of coding or maths problems. A score shows how a model did on that test, not how good it is overall.
C
- Chain of thought
- A model writing out its intermediate reasoning steps before the final answer. It can improve results on problems that need several steps.
- Context window
- How much text, counted in tokens, a model can take into account at once: your prompt, any attached files and its own reply.
D
- Deepfake
- Synthetic video, audio or images that convincingly show a real person saying or doing something they did not.
- Diffusion model
- A type of generative model that makes images, audio or video by starting from random noise and refining it step by step.
E
- Embedding
- A list of numbers that represents the meaning of a piece of text or an image, so software can find items that are similar in meaning. Used in search and RAG.
F
- Fine-tuning
- Training an existing model further on a smaller, focused set of examples, so it handles a specific task, style or field better.
G
- Generative AI
- AI that creates new content, such as text, images, audio, code or video, instead of only sorting or predicting.
- GPUGraphics processing unit
- A chip built to run many calculations in parallel. GPUs do most of the work of training and running AI models.
H
- Hallucination
- When an AI model states something false or made up as if it were a fact. Check important claims against a source you trust.
I
- Inference
- Running a trained model to get an answer, as opposed to training it. Most of the day-to-day cost of using AI is inference.
J
- Jailbreak
- A prompt or trick that gets an AI model to ignore its safety rules.
K
- Knowledge cutoff
- The date after which a model has nothing in its training data. It knows nothing newer unless it can search the web or you give it the information.
L
- LLMLarge language model
- An AI model trained on very large amounts of text to predict the next word. Chatbots such as ChatGPT, Claude and Gemini are built on LLMs.
- AI models compared
M
- Machine learning
- Software that learns patterns from examples instead of following rules a person wrote by hand. Today's AI models are built with it.
- MCPModel Context Protocol
- An open standard that lets AI apps connect to outside tools and data, such as files, calendars or databases, in one consistent way.
- Multimodal
- A model that works with more than one kind of input or output, such as text, images, audio or video.
N
- Neural network
- A model made of layers of connected units that adjust the strength of their connections as they learn from data.
- NLPNatural language processing
- The field of computing that deals with understanding and producing human language.
- NPUNeural processing unit
- A chip, or part of one, built to run AI tasks efficiently on phones and laptops without draining the battery.
O
- OCROptical character recognition
- Turning text in photos, scans or PDFs into text you can edit and search.
- Open weights
- A model whose trained parameters are published, so anyone can download and run it, usually under a licence with conditions.
- Open-weight models
P
- Parameters
- The internal numbers a model learns during training. “70B” means 70 billion parameters; more parameters do not automatically mean a better model.
- Prompt
- The instruction, question or material you give an AI model. Its wording, examples and context strongly shape the answer.
- Prompt engineering
- Writing and testing prompts so a model gives more useful and more reliable answers.
Q
- Quantization
- Storing a model's parameters with fewer bits, so it needs less memory and runs faster, usually with a small loss in quality.
R
- RAGRetrieval-augmented generation
- A technique where an AI system first looks up relevant documents, then writes its answer from them, so answers can be current and cite their sources.
- Reasoning model
- A model trained to work through a problem step by step before it answers. It is usually slower and costs more per answer.
- RLHFReinforcement learning from human feedback
- A training step in which people rate a model's answers and the model learns to prefer the kind of answers rated higher.
S
- SaaSSoftware as a service
- Software you use online through a subscription, instead of installing and owning it.
- System prompt
- Instructions set by the app or developer that shape how a model behaves in every conversation, before your own prompt.
T
- Token
- The unit of text a language model reads and writes: a word, part of a word or a punctuation mark. Prices and context limits are counted in tokens.
- Training data
- The text, images and other material a model learns from. Where it comes from and how good it is shape what the model can do.
- TTSText to speech
- Turning written text into spoken audio with a synthetic voice.
V
- Vector database
- A database that stores embeddings and finds the items closest in meaning to a query. Often used for RAG.