AI Glossary: Plain-English Terms for Normal People

AI jargon makes the whole field feel harder than it is. This glossary explains the words you keep seeing without assuming you are a researcher or software engineer.

Use it as a reference while reading the rest of Power of AI.

How to use the glossary

When a tool announcement mentions a new model, ask: is this a new engine, a new app feature, or a new workflow? When someone says "agent," ask what tools it can actually use. When someone says "context," ask how much source material it can really see.

Terms

Model

The underlying AI system that produces answers. ChatGPT, Claude, and Gemini are apps; GPT, Claude Opus/Sonnet, and Gemini models are the engines inside them.

Prompt

The instruction you give the AI. Good prompts include context, goal, constraints, and the format you want back.

Context window

How much information the AI can consider at once: your prompt, prior messages, uploaded text, tool results, and sometimes files.

Token

A chunk of text the model reads or writes. Pricing and context limits are usually counted in tokens, not words.

Agent

An AI system that can take steps toward a goal, often by using tools, reading files, searching, writing code, or asking follow-up questions.

MCP

Model Context Protocol. A standard way to connect AI tools to outside systems like files, databases, APIs, and services.

Hallucination

A confident wrong answer. It can be a fake fact, fake citation, wrong date, or made-up explanation.

RAG

Retrieval-augmented generation. The AI retrieves relevant source material first, then answers using that material.

Embedding

A numerical representation of meaning. It helps systems search by similarity instead of exact keyword match.

Fine-tuning

Training a model further on specific examples so it behaves better for a narrower task.

System prompt

Higher-priority instructions that shape how an AI behaves before the user prompt is considered.

Structured data

Machine-readable metadata on a web page, often JSON-LD, that helps search engines understand the page.

JSON-LD

A structured data format placed in a page head or body so machines can read facts about the page without guessing from layout.

Temperature

A setting that changes how varied or conservative some model outputs are. Lower usually means more predictable; higher usually means more varied.

Tool use

When an AI can call outside tools such as search, code execution, file access, calendars, browsers, or databases.

Grounding

Tying an answer to provided sources, search results, files, or database records instead of relying only on model memory.

Multimodal

Able to work with more than one kind of input or output, such as text, images, audio, video, or files.

Latency

How long the model takes to respond. A slightly weaker but faster model can be better for everyday work.

Evaluation

A repeatable way to test whether an AI system is doing the job well, often using examples, expected outputs, and human review.

Guardrail

A rule, filter, permission setting, or workflow step designed to reduce unsafe, private, wrong, or low-quality output.

Prompt injection

A trick where outside text tries to override the user or system instructions, often by hiding commands inside documents or websites.

Model router

A system that sends different tasks to different models depending on cost, speed, difficulty, or available tools.

Computer use

A capability where an AI can interact with software through a browser, desktop, terminal, or app interface.

AI crawler

A bot that reads public web pages for AI search, citation, answer generation, or model-related indexing.

Related Power of AI pages

Sources and official references

Related Power of AI pages

Keep reading with AI Finder, Prompt Studio, ChatGPT vs Claude vs Gemini, the AI glossary, and Which AI Should You Use?.