AI Tools / AI Glossary
📖 AI Glossary: 46 Terms Explained Simply
From "token" to "transformer": the AI words you keep hearing, explained in plain English.
AGI (Artificial General Intelligence)
A hypothetical AI that could understand and learn any intellectual task a human can, across many different fields. Today's AI systems are not considered AGI; whether and when it might arrive is debated.
Agent (AI agent)
An AI system that can take actions to reach a goal — for example searching the web, using tools, filling in forms or writing and running code — usually by planning several steps rather than giving a single answer.
AI (Artificial Intelligence)
The broad field of building computer systems that perform tasks that normally need human intelligence, such as understanding language, recognising images, making decisions or solving problems.
Algorithm
A set of step-by-step instructions a computer follows to solve a problem. Machine learning algorithms learn their own rules from data instead of having every rule written by a programmer.
Alignment
Research and techniques aimed at making AI systems behave in line with human intentions and values — being helpful and honest while avoiding harmful behaviour.
API (Application Programming Interface)
A way for one piece of software to talk to another. AI companies offer APIs so developers can send a prompt to a model from their own app and receive the response.
Benchmark
A standard test used to measure and compare AI models, such as a set of maths problems, coding tasks or exam-style questions.
Bias (in AI)
When an AI system produces unfair or skewed results, often because its training data over- or under-represents certain groups or views.
Chain-of-thought
A way of getting a model to work through a problem step by step before giving its final answer, which often improves results on maths and logic tasks.
Chatbot
A program you talk to in conversation. Modern chatbots are usually powered by large language models.
Context window
The maximum amount of text (measured in tokens) a model can consider at once — including your prompt, any documents you paste and the conversation so far.
Deep learning
A type of machine learning that uses neural networks with many layers. It powers most modern AI, from speech recognition to image generation and chatbots.
Diffusion model
A type of generative model used by many image generators. It learns to turn random noise step by step into a clear image that matches a text description.
Embedding
A list of numbers that represents the meaning of a piece of text (or an image) so that similar meanings end up close together. Embeddings power semantic search and recommendations.
Few-shot prompting
Including a few examples of the input and output you want in your prompt, so the model can copy the pattern.
Fine-tuning
Further training an existing model on a smaller, specific dataset so it gets better at a particular task, style or domain.
Foundation model
A large model trained on broad data that can be adapted to many tasks — for example a general language model that is later used for chat, coding or summarising.
GAN (Generative Adversarial Network)
An older generative technique where two neural networks compete: one creates images and the other tries to spot fakes, pushing the first to improve.
Generative AI
AI that creates new content — text, images, audio, video or code — rather than only analysing or classifying existing content.
GPT
Short for Generative Pre-trained Transformer, the name of a family of language models from OpenAI. People sometimes use "GPT" loosely to mean any chatbot model.
GPU
Graphics Processing Unit — a chip that does many calculations in parallel. GPUs (and similar specialised chips) are used to train and run most large AI models.
Guardrails
Rules, filters and checks placed around an AI system to keep its outputs safe, on-topic and within policy.
Hallucination
When an AI model states something false or made-up with confidence — for example inventing a fact, quote or reference. Always verify important information.
Inference
Running a trained model to get an output — for example, when you send a prompt and the model generates a reply. Training builds the model; inference uses it.
Large language model (LLM)
A model trained on very large amounts of text to predict and generate language. LLMs power chatbots and writing assistants and can summarise, translate, answer questions and write code.
Machine learning (ML)
A branch of AI where systems learn patterns from data and improve with experience, instead of following only hand-written rules.
Multimodal
An AI model that can work with more than one type of input or output — such as text and images, or text, audio and video.
Natural language processing (NLP)
The field of AI focused on understanding and generating human language.
Neural network
A computing system loosely inspired by the brain, made of layers of connected "neurons" (mathematical functions) whose connection strengths are adjusted during training.
Open-weight model
A model whose trained parameters (weights) are published so others can download, run and adapt it, subject to its licence.
Overfitting
When a model learns its training data too closely — including noise — and performs poorly on new, unseen data.
Parameter
An internal number in a model that is adjusted during training. Large language models can have billions of parameters, though more parameters do not automatically mean a better model.
Prompt
The instruction or input you give an AI model — a question, a task, some text to work on, or a description of an image to generate.
Prompt engineering
The practice of writing and refining prompts to get better, more reliable results — for example giving context, examples, a clear format and constraints.
Prompt injection
A security problem where hidden or malicious instructions inside text (such as a web page or document) try to make an AI system ignore its original instructions.
RAG (Retrieval-Augmented Generation)
A technique where the system first searches a set of documents for relevant information, then gives that to the model so its answer is grounded in those sources.
Reinforcement learning
A training method where a system learns by trial and error, receiving rewards for good actions and penalties for bad ones.
RLHF (Reinforcement Learning from Human Feedback)
A training step where people rate or compare model responses, and the model is adjusted to produce the kinds of answers people prefer.
System prompt
Background instructions given to a chatbot by the app developer — setting its role, tone and rules — usually hidden from the end user.
Temperature
A setting that controls randomness in a model's output. Lower values give more predictable answers; higher values give more varied and creative ones.
Token
A chunk of text a model reads and writes — often a short word or part of a longer word. Context limits and API pricing are measured in tokens.
Tokenizer
The component that splits text into tokens before a model processes it. Different models use different tokenizers, so token counts vary.
Training data
The examples a model learns from — text, images, code or other data. Its quality and variety strongly shape how the model behaves.
Transformer
The neural network architecture, introduced in the 2017 paper "Attention Is All You Need", that underpins most modern language models. Its "attention" mechanism lets it weigh how words relate to each other across a passage.
Vector database
A database designed to store embeddings and quickly find the items most similar in meaning — commonly used in RAG and semantic search.
Zero-shot prompting
Asking a model to do a task without giving any examples, relying only on your instructions.
No matching term. Try a shorter word.