AI Jargon Buster

The maximum number of tokens an LLM can read in one call (input + output combined).

A fixed-length vector of floats that represents the meaning of a piece of text.

When an LLM produces confident output that is factually wrong or fabricated.

Large Language Model — a neural network trained to predict the next token.

The full input string sent to an LLM — instructions, context, and the user question combined.

The unit an LLM reads and writes — roughly 3–4 characters of English text.

Retrieval-Augmented Generation — fetch relevant documents, then ask the LLM to answer using them.

An LLM-driven loop that plans, calls tools, observes results, and decides what to do next.

Continuing training on a specific dataset to adapt a base model to a domain or style.

A 0–1 (or 0–2) knob that controls how random the LLM's token sampling is.

Including 2–10 example input/output pairs in the prompt to teach the model the pattern.

A model feature where the LLM emits a structured request to call a function you define.

Breaking a task into a sequence of LLM calls where each output feeds the next.

Searching by meaning instead of keywords, using embedding similarity.

Forcing the LLM to return JSON (or another format) that conforms to a schema.

Model Context Protocol — an open standard for connecting LLMs to tools and data sources.

External storage that lets an agent recall information across calls or sessions.

Several specialised LLM agents that coordinate to solve a task one alone could not.

A prompting pattern where the model alternates Reasoning and Acting steps.

An agent calling external functions, APIs, or code to do work it cannot do internally.

Classifying LLM inputs and outputs for harmful, illegal, or policy-violating content.

Rules, classifiers, or validators that sit around an LLM call to block unsafe inputs or outputs.

A prompt crafted to make the model bypass its safety training and policies.

Removing personal data from prompts and outputs before they touch a third-party LLM.

An attack where untrusted input contains instructions the LLM follows instead of the developer's.

AI Bill of Materials — an inventory of every model, dataset, and component in an AI system.

A living document listing identified AI risks, owners, mitigations, and residual risk ratings.

A traceable record of where training and retrieval data came from and how it was transformed.

Repeatable tests that measure how well an LLM-powered system performs on a fixed dataset.

A standardised document describing a model's purpose, training data, limits, and known biases.