AI Dictionary

When a machine or computer does a job automatically so a person doesn't have to do it manually every time.

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

Like a recipe — it tells the computer exactly what to do, step by step, to get a result.

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

AI is when computers are trained to do things that usually only humans can do — like understanding what you say or recognising your face.

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

If you teach an AI using examples that aren't fair or balanced, it learns to be unfair too — just like a person can learn bad habits from bad examples.

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

A computer program you can have a text conversation with, like messaging a very smart robot.

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

Teaching computers to "see" and understand pictures and videos.

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

A way of teaching AI using many layers of learning — like practising a skill over and over until you get really good at it.

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

Information that computers use to learn. The more good data an AI has, the smarter it can become.

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

Starting with a clever AI and then teaching it to be even better at one specific thing — like a chef who already knows cooking but then specialises in Italian food.

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 very powerful type of AI that has read enormous amounts of text and learned to write in a human-like way.

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

AI that can make new things — like writing a story, drawing a picture, or composing music — instead of just looking at things that already exist.

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

When an AI makes something up and presents it as true — like a student who doesn't know the answer but guesses confidently anyway.

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

An AI that has read an enormous amount of text — books, websites, articles — and learned to understand and write language like a human.

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

Teaching a computer to learn from examples rather than giving it exact instructions for every situation — like how you learn from experience rather than following a rulebook.

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

An AI that can understand and work with different types of information at the same time — reading text AND looking at pictures, for example.

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

Think of a model as the AI's "brain" — the thing that was trained on lots of information and now knows how to do a specific job.

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

Teaching computers to understand the way humans actually speak and write — including slang, context, and meaning.

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

A way of organising a computer's "thinking" that is loosely inspired by how brain cells connect and communicate with each other.

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

AI tools that are shared freely with everyone — like a recipe made public so anyone can use and improve it.

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

Learning how to ask AI the right questions in the right way to get the best results — it's a genuine skill that takes practice.

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

What you type or say to an AI to tell it what you want it to do. A better question gets a better answer.

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

The adjustable settings inside an AI's brain. More parameters usually means the AI can learn more complex things — like having more dials to fine-tune.

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

Instead of only using what it already knows, the AI looks up relevant information first — like checking your notes before answering a test question.

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.

Teaching the AI by showing it millions of examples until it gets good at recognising patterns — like studying for an exam but at an enormous scale.

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

The small chunks that AI breaks text into before reading it — a bit like how you might read a sentence word by word.

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

The examples the AI learned from. If you teach an AI using bad or unfair examples, it learns bad or unfair things.

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

When an AI can do something new it has never seen before by using what it already knows — like a student who has never seen a question type but works it out from related knowledge.

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.