AI in Development

Artificial Intelligence / Machine Learning

AI is software that performs tasks normally needing human intelligence; ML is the main technique, where systems learn patterns from data.

Large Language Model

The kind of AI, trained on huge amounts of text, that powers chatbots and coding assistants such as ChatGPT and Claude.

The instruction you give an AI

The text you send to an AI model. 'Prompt engineering' is the craft of wording and structuring that instruction to get reliable results.

How AI text is measured

The small chunks of text an AI model reads and writes, roughly three-quarters of a word each. Usage and API pricing are counted in tokens.

How much the model can hold

The maximum amount of text a model can consider at once, covering your input and its reply. Exceed it and the earliest material drops out.

Confident but wrong output

When an AI model produces something plausible-sounding but factually incorrect. The main reason AI output needs checking before it is relied on.

Retrieval-Augmented Generation

A technique that feeds an AI model your own up-to-date documents at the moment of a question, so its answers stay accurate and grounded.

Adapting a model

Further training an existing AI model on your own data so it better suits a specific task, tone or domain.

AI that takes actions

An AI system that does not just answer, but carries out multi-step tasks using tools, such as reading files, calling APIs or sending messages. This is why approval controls on what an agent may do matter.

Model Context Protocol

An open standard, introduced in November 2024, for connecting AI assistants securely to external tools, apps and data.

Numerical semantic representations

Arrays of numbers that represent the meaning or features of text, images or other data so similar items can be compared mathematically.

Embedding search database

A database designed to store, index and search embeddings, commonly used for semantic search, recommendations and RAG retrieval.

Running a trained model

Using a trained machine-learning model to make a prediction or, for an LLM, generate a response to new input.

Model evaluations

Repeatable tests and scoring criteria used to measure an AI system's quality, safety or task performance and compare versions.

AI safety controls

Software, policies and review steps intended to prevent an AI system from causing harm, leaking data or taking unauthorised actions.

LLM application vulnerability

A crafted input that tries to override an AI application's intended instructions, expose data or cause unauthorised behaviour.

Multiple input or output modes

An AI model or system that works with more than one kind of information, such as text, images, audio or video.

Machine Learning Operations

The practices and automation used to build, deploy, monitor, govern and update machine-learning models reliably in production.