AI
AI is software that can do tasks that usually need a person's thinking, like reading words, spotting patterns or making plans.
Start with AIAI 101
AI 101
A picture dictionary of the words this site uses, for people who are not coders. The model is the engine. The harness is the vehicle. You decide who drives. 124 terms, 41 with their own page.
Start here: the first 10 words
New to AI? Read these in order. Each one is short, plain and comes with a picture.
AI model
An AI model is a program that learned patterns from lots of examples and uses them to turn an input into an output.
LLM
An LLM is an AI model trained on huge amounts of text to predict what comes next, which lets it read and write language.
Chatbot
A chatbot is an app you talk to by typing or speaking, where an AI model replies in a back-and-forth conversation.
Prompt
A prompt is the text, files or images you give an AI model to tell it what you want.
Token
A token is a small chunk of text, like a word, part of a word or a punctuation mark, that a model reads and writes.
Context window
The context window is how much text, counted in tokens, a model can look at in one go, including your prompt and its answer.
Hallucination
A hallucination is when an AI model states something false or made up as if it were true.
Harness
A harness is the software built around an AI model that gives it instructions, tools, memory and a screen, so it can do real work.
Agent
An agent is an AI model inside a harness that works toward a goal in steps, choosing actions, using tools and checking results.
A
Access token
An access token is a digital pass an app gets through OAuth that allows only the actions you approved, often for a limited time.
Agent
An agent is an AI model inside a harness that works toward a goal in steps, choosing actions, using tools and checking results.
Agent file (AGENTS.md)
An agent file is a plain text file, like AGENTS.md, with project instructions that a coding agent reads at the start of each session.
Agent loop
The agent loop is the cycle an agent repeats: look at the situation, decide, act with a tool, check the result, and go again.
Agent skill (SKILL.md)
An agent skill is a folder with a SKILL.md file of instructions, and sometimes scripts, that teaches an agent one specific task.
Agentic workflow
An agentic workflow is a multi-step process where AI plans and carries out actions toward a goal, sometimes correcting itself.
AGI
AGI is a hoped-for kind of AI that could handle almost any thinking task a person can; there is no agreed test for it.
AI
AI is software that can do tasks that usually need a person's thinking, like reading words, spotting patterns or making plans.
AI model
An AI model is a program that learned patterns from lots of examples and uses them to turn an input into an output.
Alignment
Alignment is the work of making AI systems do what people actually intend, helpfully and safely.
API
An API is a set of rules that lets one program ask another program for something, like an answer from a model.
Automation bias
Automation bias is when people trust an automated system's answer too much, even when it is wrong.
B
Benchmark
A benchmark is a standard test used to compare how well different AI models do the same tasks.
Bias
Bias in AI is when a system treats some people or groups unfairly, often because of patterns in its training data.
Branch
A branch is a separate line of work inside a repo where you can make changes without touching the main version.
C
Chain of thought
Chain of thought is a model writing out its steps on the way to an answer, rather than jumping straight to the end.
Chatbot
A chatbot is an app you talk to by typing or speaking, where an AI model replies in a back-and-forth conversation.
CLI
A CLI is a program you use by typing text commands instead of clicking buttons.
Closed model
A closed model is one whose weights stay private, so you can use it only over the internet through the maker's app or API.
Coding agent
A coding agent is an AI agent that reads a project's files, writes and changes code, runs commands and can open pull requests.
Commit
A commit is one saved change to files in git, with a note about what changed, who made it and when.
Computer use
Computer use is a tool that lets an AI agent see a screen and move the mouse and keyboard, so it can operate ordinary apps.
Content credentials (C2PA)
Content credentials are tamper-evident labels on an image or video that record how it was made or edited, including with AI.
Context engineering
Context engineering is choosing what goes into a model's context window at each step, so an agent has what it needs and no clutter.
Context window
The context window is how much text, counted in tokens, a model can look at in one go, including your prompt and its answer.
D
Deep learning
Deep learning is machine learning that uses neural networks with many layers, and it powers most modern AI.
Diffusion model
A diffusion model makes images, video or audio by starting from random noise and cleaning it up step by step to match a prompt.
Distillation
Distillation is training a smaller model to copy a bigger model's answers, so it runs faster and cheaper.
E
Embedding
An embedding is a list of numbers that captures the meaning of text or an image, so similar things end up close together.
Environment variable
An environment variable is a named setting stored on a computer, often used to keep secrets like an API key out of the code.
EU AI Act
The EU AI Act is a European Union law that sets rules for AI systems based on how risky their use is.
Evals
Evals are tests you run on an AI system to check whether its answers are good enough for your own job.
F
Few-shot prompting
Few-shot prompting means putting a few examples of good answers in your prompt so the model copies the pattern.
Fine-tuning
Fine-tuning is extra training that teaches an existing model a focused job, using a smaller set of examples.
Forecasting model
A forecasting model predicts the next numbers in a series over time, such as sales, energy use or traffic.
Foundation model
A foundation model is a very large model trained on broad data that can handle many tasks and be adapted for new ones.
Frontier model
A frontier model is one of the most capable models available at a given time; which models count changes as new ones come out.
Function calling
Function calling is the technical name for tool use: the model replies with a structured request naming a tool and its inputs.
G
Generative AI
Generative AI is AI that creates new content, such as text, images, sound or video, instead of only sorting or scoring things.
Git
Git is free software that tracks every change to a set of files, so you can see the history, undo mistakes and combine work.
GitHub
GitHub is a website where people store code projects, track changes with git and work on them together.
GPU
A GPU is a computer chip that does many calculations at once, which makes it the main hardware for training and running AI models.
Grounding
Grounding means basing an AI answer on trusted sources it was given or looked up, so it can point to where facts came from.
Guardrails
Guardrails are checks and limits around an AI system that block harmful actions, data leaks or unsafe answers.
H
Hallucination
A hallucination is when an AI model states something false or made up as if it were true.
Harness
A harness is the software built around an AI model that gives it instructions, tools, memory and a screen, so it can do real work.
Hugging Face
Hugging Face is a website where people share and download AI models, datasets and demo apps, many of them open.
Human in the loop
Human in the loop means a person checks and approves each important AI action before it happens.
Human on the loop
Human on the loop means the AI acts on its own while a person watches and can step in or stop it at any time.
I
IDE
An IDE is an app for writing code that combines an editor, tools to run and test code, and often built-in AI help.
Image model
An image model creates or understands pictures, for example drawing an image from a text description.
Inference
Inference is when a trained model is actually used: you send it a prompt and it produces an answer.
Issue
An issue is a note on a GitHub project for a bug, task or question, with its own discussion thread.
J
Jailbreak
A jailbreak is a prompt written to trick an AI model into breaking its own safety rules.
JEPA
JEPA is a world-model design from Meta research that predicts the meaning of what comes next instead of every pixel.
JSON
JSON is a simple text format for data, using labels and values inside curly brackets, that programs use to swap information.
K
Knowledge cutoff
A knowledge cutoff is the date after which a model has not learned anything from its training data.
L
Least privilege
Least privilege is the safety rule of giving a person, app or agent only the access it needs for its task, nothing more.
License
A license is the legal permission that says what you may and may not do with software, code or model weights.
LLM
An LLM is an AI model trained on huge amounts of text to predict what comes next, which lets it read and write language.
Local model
A local model is an open-weight model running on your own device, so it can work offline and your data stays with you.
M
Machine learning
Machine learning is a way to build software that learns from examples, instead of following rules a person wrote line by line.
Markdown
Markdown is a simple way to format plain text with symbols, like # for headings and * for lists.
MCP
MCP is an open standard that gives AI apps one common way to connect to outside tools and data.
MCP client
An MCP client is the part inside an AI app that holds one connection to one MCP server and passes its information back.
MCP host
An MCP host is the AI app, such as a chat or coding app, that manages connections to one or more MCP servers.
MCP server
An MCP server is a program that offers tools, data or ready-made prompts to AI apps through MCP.
Memory
Memory is notes an AI app saves and shows the model later, because models don't remember past chats on their own.
Metered API
A metered API charges you for each use, usually per million tokens in and out, instead of a flat monthly fee.
Mixture of experts
A mixture-of-experts model has many specialist parts and switches on only a few of them for each piece of text.
Model card
A model card is a document published with a model that explains what it is for, how it was made and its known limits.
Model router
A model router is a small decision layer that picks which AI model should answer each request, balancing quality, speed and cost.
Multi-agent system
A multi-agent system is several AI agents with different jobs that pass work to each other to solve a bigger task.
Multimodal
A multimodal model can take in or produce more than one kind of data, such as text, images, audio or video.
N
Neural network
A neural network is a kind of model made of many small connected math units that adjust their connections as they learn.
NIST AI RMF
The NIST AI RMF is a free US government guide that helps organizations spot and manage the risks of AI systems.
No human in the loop
No human in the loop means an AI agent finishes a task with nobody watching, which suits only low-risk jobs you chose to hand off.
O
Open source
Open source means software whose code anyone may read, change and share, under a license that meets set rules.
Open weights
An open-weight model is one whose trained weights you can download and run on your own computer or server, within its license.
OpenID Connect
OpenID Connect is a standard built on top of OAuth that handles signing in, so a site can confirm who you are.
OpenRouter
OpenRouter is a service that gives you one API and one bill for hundreds of AI models from many makers.
P
Parameters
Parameters are the numbers inside a model that it learned during training; big models have billions of them.
Password
A password is a secret word or phrase that proves who you are when you sign in; never paste one into an AI chat.
Permissions
Permissions are settings that decide what an AI agent may do on its own and what needs your OK, like editing files or sending email.
Personal data (PII)
PII is information that can identify a specific person, such as a name, address, ID number or face.
PKCE
PKCE is an extra OAuth safety step that stops anyone who intercepts your sign-in from using it, used by phone apps and command-line tools.
Pretraining
Pretraining is the first, biggest round of training, where a model learns general patterns from a huge pile of data.
Prompt
A prompt is the text, files or images you give an AI model to tell it what you want.
Prompt engineering
Prompt engineering is the skill of writing prompts that get the answers you want, by being clear, giving context and showing examples.
Prompt injection
Prompt injection is when hidden or sneaky text in a webpage, email or file tricks an AI into ignoring your instructions.
Pull request (PR)
A pull request is a proposal to add changes to a project, where others can review and discuss them before accepting.
Q
Quantization
Quantization stores a model's numbers with less detail so it needs less memory and can run on smaller computers.
R
RAG
RAG is a method where an app first looks up relevant documents, then gives them to the model so its answer uses those facts.
Rate limit
A rate limit caps how many requests or tokens you can use in a set time, such as per minute or per day.
Reasoning model
A reasoning model is trained to work through steps before giving its final answer, which helps on hard, multi-step problems.
Recommendation model
A recommendation model picks a few things you are likely to want, like videos, songs or products, from a huge catalog.
Red teaming
Red teaming is when a team deliberately attacks or tricks a system, like an AI model, to find weak spots before others do.
Repository (repo)
A repository, or repo, is a project's folder on GitHub or a computer that holds all its files and their full change history.
RLHF
RLHF is a training step where people rate a model's answers and the model is adjusted to give more of the highly rated ones.
Robotics model
A robotics model looks through a camera, reads an instruction and outputs movements for a robot; one common kind is a VLA model.
RSS
RSS is a simple format websites use to publish a feed of new posts, so a reader app can show updates without visiting each site.
S
Sandbox
A sandbox is a walled-off space on a computer where an AI agent or program can run without reaching your real files or accounts.
Science model
A science model is a specialist AI model for research jobs, such as predicting a protein's 3D shape from its building blocks.
Scope
A scope is one specific permission you approve when you connect an app with OAuth, such as 'read your calendar'.
SDK
An SDK is a ready-made set of code and tools a company gives developers to build on its service more easily.
Speech-to-text
Speech-to-text turns spoken words into written text, like automatic captions or voice notes.
Subagent
A subagent is a helper agent that a main agent hands a side task to, so the main conversation stays uncluttered.
Subscription plan
A subscription plan is a flat monthly fee for an AI app, with usage limits instead of a per-token bill.
System prompt
A system prompt is a set of instructions the app gives the model before your message, setting its role, rules and tone.
T
Temperature
Temperature is a setting that controls how random a model's answers are; low is steadier, high is more varied.
Terminal
A terminal is the app on a computer where you type commands and see text results, used to run CLI tools.
Text-to-speech
Text-to-speech turns written text into a spoken voice.
Token
A token is a small chunk of text, like a word, part of a word or a punctuation mark, that a model reads and writes.
Tool use
Tool use is when an AI model asks the app to do an action, like searching the web or running code, then reads the result.
Training
Training is how a model learns: it studies many examples and slowly adjusts its numbers until its answers get better.
Transformer
A transformer is the neural network design behind most modern language models; it looks at how all the words in a text relate.
Types of AI models
AI models come in kinds for text, images, video, sound, the physical world, science and robots, and some mix several.
V
Vibe coding
Vibe coding is making software by describing what you want in plain words and letting AI write the code, then testing the result.
Video model
A video model creates moving pictures from a text or image prompt, and some also make the sound.
W
Weights
Weights are the main learned numbers in a model, the part you download when a model is shared as open weights.
World model
A world model is an AI model that learns how the world works so it can predict what happens next, especially after an action.