AI Glossary: 30 Terms Explained in Plain English

The AI world runs on jargon. Here are the terms that come up most in the news, in plain English, roughly from the basics to the more technical. Bookmark it for the next time a headline loses you.

The basics

Artificial intelligence (AI). The broad goal of getting computers to do things that normally take human intelligence: understanding language, recognizing images, making decisions.

Machine learning. The main way AI is built today. Instead of following hand-written rules, a system learns patterns from lots of examples.

Neural network. A type of machine-learning system loosely inspired by the brain, made of layers of connected "neurons" that adjust as they learn.

Deep learning. Machine learning with very large, many-layered neural networks. It's behind most of the big AI advances of the past decade.

Generative AI. AI that creates new content (text, images, audio, video, code) rather than just sorting or labeling existing data.

Language models and chatbots

Large language model (LLM). A huge neural network trained on enormous amounts of text to predict what comes next. It's the engine behind chatbots like ChatGPT, Claude and Gemini.

Chatbot. An app you talk to in a conversation. Many modern chatbots are an LLM plus a chat interface and extra tools.

Prompt. The instruction or question you give an AI. Better prompts usually mean better answers (see our prompting guide).

Token. The small chunks of text an AI reads and writes, often a word or part of a word. Usage limits and pricing are usually counted in tokens.

Context window. How much text a model can consider at once: your conversation, documents you paste, and its own reply. Bigger windows let it work with longer material.

Hallucination. When an AI confidently states something false, like a fake fact, quote or source. Always check facts that matter.

Multimodal. A model that works with more than one kind of input or output, such as text and images, or audio and video.

Reasoning model. A model designed to work through a problem step by step before answering, which tends to help with math, code and logic.

Building and training AI

Training. The process of feeding a model data so it learns patterns. Training the largest models takes enormous computing power.

Parameters. The internal numbers a model adjusts during training. Model size is often described by parameter count.

Fine-tuning. Extra training on a smaller, specific dataset so a general model gets better at a particular task or style.

RLHF (reinforcement learning from human feedback). A training method where people rate a model's answers and the model is tuned toward the ones they prefer.

Open-weight model. A model whose trained parameters (weights) are published, so anyone can download and run it themselves.

GPU. The graphics chips that do the heavy math behind training and running AI. Demand for them is why chipmakers keep showing up in AI news.

Inference. Running a trained model to get an answer. Every chatbot reply is inference.

Benchmark. A standardized test used to compare models. Useful, but a high score doesn't always mean better real-world performance.

AI that takes action

AI agent. An AI system that can take actions toward a goal, using tools like a browser, email or code, not just answer questions. Full explainer: What is an AI agent?

Tool use. When a model calls an outside tool (a search engine, calculator, calendar or app) to get information or take an action.

RAG (retrieval-augmented generation). A setup where the AI first looks up relevant documents, then answers using them. It's a common way to make answers more accurate and up to date.

Prompt injection. An attack where hidden instructions in text the AI reads (a web page, an email) try to hijack what it does. A major security concern for AI agents.

Safety, trust and policy

Deepfake. AI-generated or AI-altered video, audio or images that make it look like someone said or did something they didn't. See how to spot AI images and video.

Voice clone. An AI imitation of a specific person's voice, now a common tool in phone scams. See how to protect your family.

Content Credentials (C2PA). A standard for attaching tamper-evident information to a file about how it was made and edited, including whether AI was involved.

Alignment. The work of making AI systems reliably do what people intend, and not do harmful things.

AGI (artificial general intelligence). A hypothetical AI that could match humans across most kinds of thinking work. There's no agreed definition, and claims about when it will arrive vary widely.