AI Glossary

An AI system trained on massive amounts of text to understand and generate human language, such as ChatGPT, Claude, and Gemini.

The foundational architecture behind most modern language models, which uses an "attention" mechanism to understand relationships between words.

The numerical values a model learns during training. More parameters generally mean more capability — large models have billions of them.

A way of converting text or images into numerical vectors that represent their meaning, letting computers compare similarity between them.

The maximum amount of text a model can process at once. Larger windows allow the model to analyze entire long documents.

A model that can handle multiple types of data at once — for example text, images, and audio together.

A large general-purpose model used as a base for building specialized applications through fine-tuning or prompting.

A design where a model is split into several specialized "experts", activating only a subset for each input to save compute.

A model whose weights are publicly available to download, study, and run freely — such as Llama, Mistral, and DeepSeek.

The newest and most capable models available at a given moment, typically from top labs like OpenAI, Anthropic, and Google DeepMind.

A model that runs on your own hardware instead of a vendor's cloud — used when data must never leave the machine, often at the cost of capability.

A model dedicated to one organisation, isolated from other tenants — your data is never shared or used for training. Different from a local model.

The principle that a country or region should control its own AI infrastructure, data, and models rather than depend entirely on foreign providers.