AI glossary from A to Z
Short explanations of key terms, from agent to world model.
Each technical term comes with a one-sentence meaning. Every entry links to a longer answer.
| Term | What it means | Explanation |
|---|---|---|
| Agent | A system that chooses steps towards a goal and can use tools. | Read |
| AGI | Short for artificial general intelligence; a concept with no single yardstick. | Read |
| Algorithm | A sequence of steps for carrying out a task. | Read |
| API | An agreed way for programs to send each other requests and data. | Read |
| Attention | A mathematical mechanism for connecting parts of the data. | Read |
| Automation | A process that carries out a task automatically. | Read |
| Benchmark | A standardised test for comparing systems. | Read |
| C2PA | A set of standards for the origin and history of digital content. | Read |
| Causal analysis | Studying what a change would cause. | Read |
| Context window | How much content a model can process at one time. | Read |
| Deep learning | Machine learning with neural networks that have many layers. | Read |
| Deepfake | Synthetic or altered content that can mislead people about whether it is genuine. | Read |
| Diffusion model | A family of generative models that learn to create an image from noisy, scrambled data. | Read |
| Embeddings | Numerical representations used to find similarities and other relationships. | Read |
| Embodied AI | AI tied to a body and sensors acting in an environment. | Read |
| Fine-tuning | Further training of a model that has already been trained. | Read |
| Generative AI | AI that creates text, images, sound, video or other outputs. | Read |
| GPU | A type of processor suited to doing many calculations in parallel. | Read |
| Hallucination | A wrong or made-up answer from a model that sounds convincing. | Read |
| Inference | Using a trained model to produce output. | Read |
| LLM | A large language model for working with text and related data. | Read |
| Machine learning | Methods that learn patterns from examples. | Read |
| MCP | An example of a protocol for connecting AI apps to tools and data sources. | Read |
| Model | A computational system that turns input into a prediction, decision or content. | Read |
| Multi-agent system | A system in which several agents cooperate or compete. | Read |
| Multimodality | Handling different kinds of data, such as text and images. | Read |
| Neural network | A model made of layers of connected computing units. | Read |
| NLP | The field of processing natural language by computer. | Read |
| Open weights | A model’s trained weights (parameters) made available for download; not necessarily open data. | Read |
| Parameters | Numerical values adjusted while a model is trained. | Read |
| Prompt | The instruction or question you give an AI. | Read |
| Prompt injection | An attack using instructions hidden in untrusted content. | Read |
| RAG | An approach in which the system looks up extra sources before answering. | Read |
| RL / reinforcement learning | Learning from feedback on the consequences of actions. | Read |
| Self-supervised learning | The learning task is created from the data itself. | Read |
| Token | A unit of text or other data that a model processes. | Read |
| Transformer | A model design that uses, among other things, the attention mechanism. | Read |
Last reviewed 1 November 2026. An explainer, not professional, legal or medical advice. For rules and figures that change, check the official version.