The AI glossary, in plain words

An AI system that can take actions, not just answer. It uses tools, checks the result and decides what to do next.

Coding agents edit files and run tests until a task is done.

The work of making AI systems pursue the goals people actually intend, safely and honestly.

Why assistants decline some requests and ask before risky actions.

A way for software to talk to other software. AI companies offer APIs so developers can build models into their own products.

Most AI features inside apps you use are powered through an API.

A standard test used to compare models on tasks like coding, maths or reasoning.

Useful for a rough comparison, but your own tasks are the real test.

A program you talk to in natural language. Modern chatbots are built on large language models.

ChatGPT, Claude and Gemini all started as chatbots.

A link that lets an assistant read from or act in another app, such as your email, calendar or files.

Connect only what a task needs, and review access regularly.

How much text a model can consider at once, including your messages, files and its replies.

A bigger window means you can paste a whole report and ask about it.

A mode where an assistant searches many sources over several minutes and writes a structured report.

Good for a first map of a topic. Always check the key sources yourself.

A type of model that creates images or video by gradually turning random noise into a picture that matches your prompt.

Most image and video generators work this way.

A list of numbers that represents the meaning of a piece of text, so similar ideas end up close together.

Powers "search by meaning" instead of by exact words.

Training an existing model further on your own examples so it gets better at a specific task or style.

Often unnecessary today. Good instructions and examples go a long way.

One of the most capable models available at a given time, usually from a major lab.

New frontier models tend to arrive every few months.

AI that creates new content, such as text, images, audio, video or code, rather than only classifying or predicting.

Everything in our tools directory is generative AI.

Tying a model’s answer to specific sources, such as documents or search results, so it can be checked.

NotebookLM and Perplexity show exactly where each answer came from.

Rules and checks that limit what an AI system will do, to keep it safe and on task.

Why a support bot will not discuss topics outside its job.

When a model states something false with confidence, such as an invented quote or citation.

Ask for sources, and verify anything that matters.

Running a trained model to get an answer. Each message you send is an inference.

Inference cost is why heavy use is often on paid plans.

A model trained on vast amounts of text to predict and generate language. The engine behind most assistants.

Claude, GPT and Gemini models are LLMs.

The Model Context Protocol, an open standard for connecting AI assistants to tools and data sources.

Lets one connector work across several assistants.

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

Snap a photo of a chart and ask what it shows.

A model whose trained parameters are published, so anyone can download and run it.

Popular with developers who need to run AI on their own servers.

The internal numbers a model learns during training. More parameters often, but not always, means more capability.

Model sizes are usually quoted in billions of parameters.

The instructions and context you give a model. Clear prompts get better results.

See our prompt library for patterns that work.

Hidden instructions in content an AI reads, such as a web page, that try to make it do something you did not ask.

A key reason agents should ask before sensitive actions.

Retrieval-augmented generation. The model looks up relevant documents first, then answers using them.

How "chat with your documents" tools stay accurate.

A model that works through a problem step by step before answering, which helps with maths, code and planning.

Slower, but noticeably better on hard problems.

Background instructions set by the app or developer that shape how an assistant behaves in every conversation.

Custom GPTs and Claude Projects are built on system prompts.

A setting that controls how predictable or varied a model’s output is.

Low for facts and code, higher for brainstorming.

A chunk of text, roughly three-quarters of a word, that models read and write. Limits and prices are counted in tokens.

A 1,000-word article is about 1,300 tokens.

The text, images and other material a model learned from.

Explains why models know some things and not others.

Building software mainly by describing what you want to an AI and steering the result, rather than writing code line by line.

App builders like Lovable made this mainstream.