Agent: An AI tool that can carry out tasks or take decisions to achieve a particular goal. It can use other software or the internet, rather than just replying to a single prompt.

Agentic AI: AI designed or used to perform tasks to achieve a goal, involving multiple steps. This may involve a single AI agent, or a complex system of multiple agents.

Context length: The amount of information an AI model can process at one time, usually measured in tokens. This can include your instructions, the conversation so far, attached documents, and the model’s response. If the amount of information exceeds the model’s limit, some information may be left out or handled less reliably.

Hallucination: When an AI states something that sounds plausible, but is inaccurate, unsupported, or made up. Always verify the output against reliable sources.

Inference: The process in which a trained AI model uses its learned patterns to produce an output, such as an answer, prediction, translation, or image, in response to input (a prompt).

Large language model (LLM): A type of AI model trained on very large collections of text. It generates text by predicting likely sequences of tokens based on the input and the patterns it learned during training. 

Multimodality: Describes an AI tool that can work with more than one type of information, such as text, images, audio, or video.

Prompt: The instruction or question you give an AI tool, typically written in ordinary language. The clearer and more specific it is, the better the response (though this is no guarantee for accuracy).

Prompt engineering: The practice of designing and refining prompts to improve the quality of the output. This can involve writing techniques, providing more information sources, or setting boundaries for a tool.

RAG (retrieval-augmented generation): A method where the AI tool looks up information from specific sources such as documents, a database or the web, before answering. This grounds the output of a tool in those sources, rather than relying on the tool’s training data alone.

Token: The small pieces of text a model processes. This can be a whole word, parts of a word, or punctuation signs.  A tool's limits (see context length) are usually counted in tokens.