AI Dictionary

Using technology to perform tasks with minimal human involvement. AI-powered automation can handle complex, variable tasks that older automation systems could not.

Example:AI automation can read incoming emails and sort them into folders without anyone clicking a button.

A set of step-by-step instructions that a computer follows to complete a task or solve a problem. Algorithms are the foundation of all software, including AI systems.

Example:A recommendation algorithm decides which videos to show you next on YouTube based on what you've watched.

Computer systems designed to perform tasks that normally require human intelligence, such as understanding language, recognising images, making decisions, and solving problems.

Example:The voice assistant on your phone uses AI to understand your questions and give you answers.

When an AI system produces unfair or skewed results because the data it was trained on contained imbalances or prejudices. AI bias can reflect and amplify human biases present in historical data.

Example:A hiring AI trained mostly on male applicants' data might unfairly score female applicants lower.

A computer program designed to simulate conversation with humans, typically via text. Modern chatbots use large language models to generate human-like responses.

Example:Customer service chatbots on websites answer common questions without needing a human agent.

A branch of AI that enables computers to interpret and understand visual information from the world — such as images and video — in the same way humans use their eyes and brain.

Example:Computer vision lets a self-driving car identify pedestrians, traffic lights, and road signs.

A type of machine learning that uses neural networks with many layers to learn complex patterns from large datasets. Deep learning powers most modern AI breakthroughs including image recognition and language models.

Example:Deep learning enables AI to translate between languages, generate images from text descriptions, and detect diseases in medical scans.

Raw facts and information — numbers, text, images, audio — that computers process and learn from. AI systems require large amounts of high-quality data to train effectively.

Example:An AI that recognises cats in photos learned by processing millions of labelled cat images.

The process of further training a pre-built AI model on a specific, smaller dataset to make it better suited for a particular task or domain.

Example:A company might fine-tune a general language model on their own customer service documents to make it more helpful for their specific business.

A family of large language models developed by OpenAI. GPT models are trained on vast amounts of text data and can generate human-like text, answer questions, write code, and more.

Example:ChatGPT is built on GPT technology and can hold conversations, write essays, and solve problems.

AI systems that can create new content — including text, images, audio, video, and code — rather than simply analysing or classifying existing data.

Example:ChatGPT generates written responses, DALL-E generates images, and Suno generates music — all are examples of generative AI.

When an AI model generates information that sounds confident and plausible but is factually incorrect or entirely made up. A key limitation of current large language models.

Example:An AI might hallucinate by inventing a fake quote from a real person, or stating an incorrect historical date as fact.

A type of AI model trained on massive amounts of text data that can understand, generate, and manipulate human language. Examples include GPT-4, Claude, Gemini, and Llama.

Example:When you ask ChatGPT a question, a large language model processes your words and generates a relevant answer.

A branch of AI where systems improve their performance on a task by learning from data, without being explicitly programmed for every possible scenario.

Example:A spam filter uses machine learning to recognise unwanted emails by studying thousands of examples of spam and non-spam messages.

AI systems that can process and generate multiple types of data — such as text, images, audio, and video — rather than being limited to one format.

Example:GPT-4o is multimodal — you can show it a photo and ask a question about it in text.

In AI, a model is the result of training an algorithm on data. It is the system that makes predictions, generates content, or performs tasks.

Example:The model behind a translation app has learned patterns across many languages to convert text from one language to another.

A branch of AI focused on enabling computers to understand, interpret, and generate human language in a meaningful way.

Example:NLP powers autocomplete suggestions when you type a text message, as well as voice assistants and translation tools.

A computing architecture loosely inspired by the human brain, consisting of layers of connected nodes (neurons) that process information and learn patterns from data.

Example:Neural networks are used in facial recognition systems, language translation, and medical image analysis.

AI models and tools whose underlying code and sometimes training data are made publicly available, allowing anyone to inspect, use, or build upon them.

Example:Meta's Llama models are open source — developers can download and run them on their own computers.

The skill of designing, refining, and optimising prompts to get better, more accurate, or more useful outputs from AI systems.

Example:A prompt engineer might test dozens of phrasings of the same question to find the version that produces the most accurate answer.

The instruction, question, or input you give to an AI tool. The quality and clarity of your prompt directly affects the quality of the AI's output.

Example:"Summarise this article in three bullet points for a ten-year-old" is a more effective prompt than just "summarise this."

The numerical values within an AI model that are adjusted during training. The number of parameters is often used as a measure of a model's size and complexity. GPT-4 is estimated to have over a trillion parameters.

Example:A model with more parameters can generally understand more nuance but also requires more computing power to run.

A technique where an AI model retrieves relevant information from an external knowledge base before generating a response, reducing hallucinations and improving accuracy on specific topics.

Example:A customer service AI using RAG searches a company's knowledge base before responding, so its answers are based on up-to-date, accurate information.

The process by which an AI model learns from data — adjusting its internal parameters to improve performance on a task. Training large models requires enormous amounts of data and computing power.

Example:Training GPT-4 involved processing hundreds of billions of words of text across months of computation on thousands of specialised chips.

The basic unit of text that AI language models process. A token is roughly equivalent to a word or part of a word. Models have a "context window" that limits how many tokens they can process at once.

Example:The sentence "AI is fascinating" contains four tokens — though exact tokenisation varies by model.

The dataset used to teach an AI model. The quality, diversity, and size of training data significantly affects the model's capabilities and any biases it may exhibit.

Example:An image recognition AI is trained on millions of labelled photographs so it learns to identify objects correctly.

The ability of an AI model to perform a task it was never explicitly trained on, by applying knowledge learned from related tasks.

Example:A language model that was never specifically trained to write poetry can still write poems by applying its understanding of language structure and patterns.