Generative AI is one of the fastest-moving areas of technology, and it comes with its own set of terms that can be confusing at first. This generative AI glossary explains the most important concepts in plain English — no technical background required. Whether you’re exploring AI tools for your work or trying to understand what you’re reading in the news, this guide has you covered.
For a broader overview of AI terminology, see our main AI Glossary: 100+ Artificial Intelligence Terms Explained Simply.
What Is Generative AI?
Generative AI refers to artificial intelligence systems that can create new content — text, images, audio, video, or code — based on patterns learned from large amounts of existing data. Unlike older AI that only classified or analysed data, generative AI produces something new. Tools like ChatGPT, Claude, Midjourney, and Sora are all examples of generative AI in action.
Core Generative AI Terms
Generative AI — AI that creates new content by learning patterns from existing data. It can write articles, generate images, compose music, and write code based on a text prompt or other input.
Large Language Model (LLM) — A type of generative AI trained on massive amounts of text to understand and produce human language. GPT-4, Claude, and Gemini are LLMs. They power most of today’s AI chatbots and writing assistants.
Foundation Model — A large AI model trained on broad, general data that can be adapted to many different tasks. Foundation models like GPT-4 or Gemini serve as the base for hundreds of applications without being retrained from scratch each time.
Diffusion Model — The type of AI architecture most commonly used for generating images. It learns to gradually remove noise from a random image until a coherent picture emerges. Stable Diffusion and DALL-E use diffusion models.
Multimodal AI — A generative AI model that can work with multiple types of content — such as text and images together. GPT-4o can take an image as input and respond in text. Multimodal AI is increasingly the standard for frontier models.
Inference — The process of running a trained generative AI model on new input to produce output. Every time you send a message to an AI chatbot and receive a reply, that is an inference.
Context Window — The maximum amount of text (measured in tokens) a language model can read and consider at once. A larger context window allows the model to process longer documents or maintain longer conversations before “forgetting” earlier content.
Token — The basic unit a language model processes. A token is roughly equivalent to a word or part of a word — “running” is one token, while “artificial intelligence” is about three. Token limits directly affect how much text a model can handle at once.
Hallucination — When a generative AI model produces information that sounds plausible and confident but is factually incorrect or entirely made up. Hallucinations are a known limitation of LLMs. Always verify important facts from AI outputs with reliable sources.
Synthetic Media — AI-generated content designed to look or sound like real content — photos, voices, videos, or text. Deepfake videos are a well-known example of synthetic media used in misleading ways.
AI Workflow — A structured sequence of AI-powered steps designed to complete a task automatically — for example, receiving a customer email, summarising it, drafting a reply, and routing it for review.
AI Agent — A generative AI system that can take autonomous multi-step actions — browsing the web, running code, writing files — to complete a goal. Agents go beyond simply responding to a single prompt.
Prompting & Output Control Terms
Prompt Engineering — The practice of carefully writing the instructions you give an AI model to improve the quality, accuracy, or format of its output. A well-engineered prompt is specific, provides context, and clearly states the desired outcome.
Zero-Shot Prompting — Asking an AI to complete a task without giving it any examples. The model relies entirely on its training to understand and respond. Example: “Summarise this article in three bullet points.”
Few-Shot Prompting — Including a small number of examples in the prompt to guide the model toward the format or style you want. Example: showing the model two sample Q&A pairs before asking it to generate more in the same format.
Temperature — A parameter that controls how creative or predictable the model’s output is. A temperature near 0 produces consistent, focused answers; a higher temperature produces more varied and imaginative responses.
Top-p Sampling — An alternative to temperature for controlling output randomness. The model only considers tokens whose combined probability reaches a threshold (the “p” value), cutting off unlikely choices. Often used alongside temperature.
Training & Customisation Terms
Fine-Tuning — Training an existing foundation model on a smaller, specific dataset to make it better at a particular task or domain. A legal tech company might fine-tune an LLM on thousands of contracts to improve its legal accuracy.
Retrieval-Augmented Generation (RAG) — A technique where an AI model retrieves relevant information from an external knowledge source before generating its response. RAG reduces hallucinations and makes answers more current without retraining the model.
Knowledge Base — A collection of documents, FAQs, or data that an AI system can search and draw from to answer questions. Used in RAG systems and AI-powered support tools.
Embeddings — Numerical vector representations of text that capture meaning and relationships. Embeddings allow AI systems to find semantically similar content — retrieving the most relevant documents for a RAG query, for example.
Safety & Quality Terms
Guardrails — Rules, filters, and constraints that limit what a generative AI will produce — preventing harmful, offensive, or off-topic outputs. Guardrails can be applied at the model level or added separately by the application developer.
Bias (AI) — When a model produces outputs that reflect unfair or skewed patterns from its training data. Generative AI can inherit and amplify biases present in the text or images it was trained on.
Alignment — The challenge of ensuring a generative AI system does what its creators and users actually intend — safely and reliably. Poor alignment can cause a model to produce outputs that are technically responsive but harmful or unhelpful in practice.
Frequently Asked Questions
What is the difference between generative AI and regular AI?
Traditional AI is mostly used to classify, analyse, or make predictions from existing data — like detecting spam or recommending products. Generative AI creates new content — text, images, audio, or video — based on what it has learned. The key difference is output: analyse vs. create.
What is a foundation model?
A foundation model is a large AI model trained on broad, general data that can be adapted to many tasks. GPT-4 and Gemini are foundation models. Rather than building a new AI for every task, developers start from a foundation model and fine-tune or prompt it for their specific needs.
How does RAG reduce hallucinations?
RAG (Retrieval-Augmented Generation) gives the AI model access to relevant, factual documents before it responds. Instead of relying solely on what it memorised during training, the model retrieves the correct information first and bases its answer on that. This significantly reduces the chance of the model making things up.
What is the difference between fine-tuning and RAG?
Fine-tuning changes the model itself by training it on new data. RAG leaves the model unchanged but gives it access to an external knowledge source at runtime. Fine-tuning is better for teaching a model a style or domain-specific language; RAG is better for keeping answers current and grounded in specific documents.
Final Thoughts
Generative AI terminology can look intimidating at first, but most of these concepts follow a clear logic once you see how they connect. The technology creates content by learning from patterns, and the terms around it describe how that learning happens, how outputs are controlled, and how the systems are kept safe and accurate.
For a complete overview of AI terms across all categories, visit our main AI Glossary. You can also explore our ChatGPT & AI Chatbot Glossary for practical terms you’ll use every day.
