AI Glossary

This page explains common AI terms in plain English. If you encounter unfamiliar words while learning about AI, you'll find explanations here.

Attention

A mechanism that helps AI focus on the most important parts of information. Think of it like paying close attention to certain words in a sentence while skimming over less important ones. Learn more about attention

Backpropagation

The process AI systems use to learn from mistakes. When an AI gives a wrong answer during training, backpropagation traces back through the system to find which settings (weights) caused the error, then adjusts them to do better next time. It's like learning from feedback.

Context Window

The amount of text an AI can "read" at one time. Think of it like how much you can hold in your working memory. If your document is too long, it exceeds the context window and the AI can't see all of it. Learn more about context windows

Embeddings

A way to represent words or ideas as numbers so AI can understand them. Imagine converting the meaning of a word into coordinates on a map—similar words would be close together on the map. This helps AI compare and understand how words relate to each other. Learn more about embeddings

Fine-tuning

Taking an already-trained AI system and adjusting it for a specific task or style. Like a musician taking their instrument and making small adjustments to get the exact sound they want. Learn more about fine-tuning

Hallucination

When AI generates false or made-up information as though it were real. The AI isn't literally hallucinating or seeing things—it's a term we use to describe when AI confidently invents answers. For example, it might invent a quotation someone never said, a book that was never published, or statistics that aren't real. Learn more about why AI gets things wrong

Large Language Model (LLM)

A type of AI trained on enormous amounts of text. It learns patterns from all that text to predict what should come next in a sentence. ChatGPT, Claude, and Gemini are all large language models. Learn more about language models

Neural Network

A system of connected nodes (like neurons) inspired by how brains work, but much simpler. Information flows through these connections, and the system learns by adjusting the strength of those connections. Learn more about neural networks

Parameters

The settings inside an AI system that determine how it works. Think of parameters like the thousands of tiny dials and switches inside a complex machine. For a large language model, there might be billions of parameters, each contributing to the overall behavior. Learn more about parameters

Tokens

Small pieces of text that AI systems break your input into. A token might be a whole word, part of a word, or punctuation. It's like how a cashier breaks a £50 note into smaller denominations—AI breaks text into smaller units so it can process it step by step. Learn more about tokens

Training (AI Training)

The process of teaching an AI system by showing it enormous amounts of data. During training, the system learns patterns and relationships. Think of it like learning a skill by practicing repeatedly. Learn more about training

Transformer

A type of AI architecture (design) that's particularly good at understanding language. Most modern AI chatbots like ChatGPT and Claude use transformer architecture. The name comes from the "attention" mechanism that transforms how the system focuses on important information. Learn more about transformers

Weights

The numbers that represent the strength of connections in a neural network. Weights determine how strongly one part of the AI system influences another. Think of weights like the volume knob on a speaker—they control how "loud" or important each connection is. During training, the system learns the best weights to produce accurate answers. Learn more about weights