Glossary

Artificial intelligence (AI) is the field of computing that builds programs able to do tasks that used to need human intelligence: recognizing a face, translating, recommending a song or holding a conversation.

Machine learning is the part of artificial intelligence where a program learns patterns from examples instead of following hand written rules.

A neural network is a kind of machine learning model made of layers of simple units connected to each other.

Deep learning is the part of machine learning that uses neural networks with many layers.

Artificial general intelligence (AGI) is the idea of a system that can learn and handle almost any intellectual task a person can, and move from one to another without being trained again.

Generative AI is artificial intelligence that creates new content (text, images, audio or code) from a request called a prompt.

Multimodal AI is AI that works with more than one kind of data: text, images, audio or video.

Computer vision is the field of AI that teaches a machine to understand images and video: recognize a face, read a license plate or find a dog in your photos.

A deepfake is a fake video, photo or audio made with AI to look real: someone's face on another person's body, or their voice saying something they never said.

A prompt is all the text an AI receives to do its job: your request, the instructions, the context and any examples you add.

Prompt engineering is the practice of writing, testing and adjusting the instructions an AI model receives until it answers well consistently.

An AI model is a computational system trained on lots of data to do a task, like writing text, recognizing images or transcribing speech.

An LLM (large language model) is an AI model trained on huge amounts of text to generate text.

A chatbot is a program that talks with you by text or voice.

A token is the unit of text a language model like ChatGPT works with.

A tokenizer is the program that turns text into tokens before a language model reads it, and turns the tokens the model answers with back into text.

The context window is the maximum amount of text, measured in tokens, that an AI model can take into account at once: your messages, its answers, the instructions and any documents you give it.

Temperature is a setting that changes how much randomness an AI model uses when picking the next token, the piece of text that comes next.

An AI hallucination is an answer that sounds confident and convincing but is false or made up: a fact, a quote, a law, a link or a book that does not exist.

An embedding is a list of numbers that represents the meaning of a piece of text.

A vector database stores data as lists of numbers, called embeddings, that represent its meaning, and quickly finds the ones closest to a query.

RAG (retrieval-augmented generation) is a technique where, before answering, the system searches your documents for the chunks that best match the question and passes them to the model along with it.

Fine-tuning is training an existing AI model further on your own examples so it specializes in a task.

An AI API is the door a program uses to send text to a model and get the answer back, without going through a chat screen.

Natural language processing (NLP) is the branch of artificial intelligence that processes human language, written or spoken.

A local AI model is a model you download and run on your own computer, instead of using it on a company's servers.

Quantization means storing an AI model's weights with fewer bits, for example 8 or 4 instead of 16.

An open weights model is an AI model whose weights, the numbers adjusted during training, are available to download.

On-device AI is AI that does its computing inside your own phone or laptop, not on a server.

"System One model" is the name TypeSafe AI uses for AI models built for quick, bounded judgments: picking an option, scoring or answering yes or no, with probabilities, instead of writing text.

Jev is an AI model by TypeSafe AI that does not write text: you give it a context and typed questions, and it returns structured answers with probabilities.

The Decisions API is an OpenAI endpoint that does not write text: you give it a text, or text and images, plus typed questions, and it returns decisions with probabilities.

Structured outputs is a feature of AI APIs that forces an LLM to answer with JSON that follows a schema you define: which fields go in, of what type and which are required.

An AI classifier is a model that assigns a category to what it receives: an email, a text or a photo.

A probability is calibrated when it matches what happens across many comparable cases: across many comparable predictions with probability 0.8, that option matches the actual outcome about 80% of the time.

An API key is a long password that identifies your account when a program uses an AI model through its API.

An AI agent is a program that uses a language model to reach a goal in several steps.

An AI browser is a web browser with AI built in.

A coding agent is an AI tool that works inside your project: it reads files, edits them, runs commands like the tests and checks what happened.

A subagent is an agent the main agent launches for a narrow task, like finding where a function is used.

Vibe coding is programming by describing what you want to an AI and accepting the code it generates without reading it: if it works when you try it, you move on.

CLAUDE.md is a text file in your repository with the project rules that a coding agent like Claude Code reads at the start of every session.

AGENTS.md is a text file in your repository with instructions for coding agents: how to install, how to run the tests and which conventions to follow.

A diff is the list of differences between two versions of a file or a project: which lines were added, which were removed and where.

Plan mode means asking an AI agent to first propose what it will change, where and how it will test it, without editing anything yet.

Context engineering is deciding what goes into the model's context window at each step and what stays out: instructions, rules files, tool results, history, retrieved documents and summaries.

MCP (Model Context Protocol) is an open standard, created by Anthropic in 2024, for connecting AI apps like Claude or ChatGPT to tools and data.

An MCP server is the program that offers tools and data to an AI app using MCP.

An API (application programming interface) is the set of rules one program uses to ask another for data or actions: what you can ask for, how to ask and what format the answer comes in.

A skill is a reusable recipe that teaches an AI agent how to do one specific task: packaged instructions and, sometimes, supporting files or scripts.

Prompt injection is when text the model reads (a web page, an email, a PDF, a comment) carries instructions and the model follows them as if they came from you.

AI guardrails are controls that surround a model or an agent: they check what goes in, filter what comes out and limit what it can do.

Using AI through an API is usually billed by tokens: you pay for what you send (input) and what the model answers (output), each with its own rate per million.

They are two ways to pay for AI.

An AI benchmark is a fixed exam, with questions or tasks whose answers are known, given to several models to compare them by the same yardstick.

Prompt caching is a feature of some AI APIs that stores the work done on the start of an input so it can be reused.

Effort is a setting in some AI models that steers how much work they do before answering: how much they reason and, depending on the provider, also how long the answer is and how many tools they call.

Reasoning tokens are the intermediate text a reasoning model generates before it gives its final answer.

A reasoning model is a language model trained to write intermediate steps before giving its final answer.

Chain of thought means getting a language model to write out intermediate steps before its final answer.

Batch processing, or the batch API, is an option in some AI APIs where you send many requests together and the provider runs them when it has spare capacity.

The Jevons paradox says that using a resource more efficiently can raise its total consumption instead of lowering it.