AI Glossary
Plain-English definitions of the artificial intelligence terms you meet in the news, from agents to tokens.
Agent. An AI system that carries out multi-step tasks using tools such as a browser, files or other apps, instead of only answering. See what are AI agents.
AGI (artificial general intelligence). A loosely defined idea of AI that can do most intellectual work as well as a person. There is no agreed test for it.
Alignment. Work aimed at making AI systems behave as their developers and users intend.
API. The interface developers use to build a model into their own software, usually billed by the token.
Benchmark. A standard test used to compare models. Scores are useful but do not always predict real-world performance.
Chatbot. An app for talking to a language model, such as ChatGPT, Claude or Gemini.
Context window. The most text a model can consider at once, measured in tokens. See tokens and context windows.
Deepfake. Synthetic audio, image or video made to look or sound like a real person. See how to spot AI-generated images.
Diffusion model. The type of model behind most AI image generators, which builds a picture by gradually removing noise.
Fine-tuning. Further training of an existing model on specific data to specialise it.
Frontier model. One of the most capable models available at a given time.
Generative AI. AI that produces new text, images, audio, video or code.
Guardrails. Safeguards that stop a model producing harmful or disallowed output.
Hallucination. A confident answer that is false or made up. See AI hallucinations explained.
Inference. Running a trained model to get an answer, as opposed to training it.
Large language model (LLM). A model trained on large amounts of text to predict and generate language.
Multimodal. Able to work with more than one kind of input or output, such as text, images and audio.
Open-weight model. A model whose trained parameters are published so anyone can run it. See running AI models locally.
Parameters. The numerical values a model learns in training. More parameters generally means a larger, more capable and more expensive model.
Prompt. The instruction or question you give a model. See how to write AI prompts.
Prompt injection. An attack in which hidden instructions in a web page, file or message try to hijack an AI system.
RAG (retrieval-augmented generation). A setup in which a model looks up relevant documents and answers from them, which reduces made-up answers.
Reasoning model. A model or mode that works through a problem in steps before answering, trading speed for accuracy.
System prompt. Background instructions, set by the app's developer, that shape how a model behaves.
Token. The unit of text a model reads and writes: roughly four characters in English. Try the token calculator.
Training data. The text, images or other material a model learned from.
Watermark. A hidden or visible marker showing that content was AI-generated.