AI Glossary for Beginners: 40+ Common AI Terms in Plain English

An AI glossary is a short list of the words people use when they talk about artificial intelligence tools, explained in everyday language. This one covers more than 40 terms that beginners run into most often, such as large language model, prompt, token and hallucination, grouped by topic so you can learn related ideas together. You do not need any technical background to use it.

How should you use this glossary?

You do not need to memorize anything. Skim the section that matches what you are doing right now, and come back when a new word trips you up. Where a term relates to a product feature that changes over time, we describe the general idea and suggest checking the provider's own help pages for details.

If you are brand new, read What is an AI chatbot? first. It gives the big picture that makes these terms easier to place.

What are the core AI terms everyone uses?

• Artificial intelligence (AI). A broad label for computer systems that do tasks we usually associate with human thinking, such as understanding language, recognizing images or making predictions. It covers everything from spam filters to chatbots.
• Machine learning. A way of building AI where the system learns patterns from examples instead of being programmed rule by rule. Show it many labeled photos of cats and dogs, and it learns to tell them apart.
• Neural network. A type of machine learning model made of many connected layers of simple math units. The design was loosely inspired by the brain, but it does not work like a brain.
• Deep learning. Machine learning that uses neural networks with many layers. Most of today's headline AI tools, including chatbots and image generators, are built with deep learning.
• Generative AI. AI that creates new content, such as text, images, audio, video or code, rather than only sorting or labeling existing content.
• Model. The trained system that does the actual work. When people say "a newer model," they mean a different trained version, which may be more capable, faster or cheaper to run.
• Large language model (LLM). A model trained on a very large amount of text to predict and generate language. LLMs power general chatbots such as ChatGPT, Claude, Gemini and Microsoft Copilot.

How do AI models learn? Training terms explained

• Training data. The examples a model learns from. For a language model, that is mostly text. The quality and mix of training data strongly shape what a model is good or bad at.
• Training. The process of adjusting a model, often over a long period on powerful computers, until its outputs match the patterns in its training data well.
• Parameters. The internal numbers a model adjusts during training. You may see models described by their number of parameters; more is not automatically better for your task.
• Knowledge cutoff (training cutoff). The point in time after which a model has no information from training. Without a web search feature, a chatbot may not know about recent events.
• Fine-tuning. Extra training on a smaller, focused set of examples to make a model better at a particular style or task, such as answering customer-support questions.
• Inference. The moment a trained model is actually used to produce an answer. When you send a message and get a reply, that is inference.

What words will you see when using a chatbot?

• Chatbot. A program you talk to in everyday language by typing or speaking. Modern AI chatbots generate fresh replies rather than following a fixed script.
• Prompt. The message, question or instruction you give an AI tool. Better prompts usually mean better answers. See What is a prompt? for examples.
• Response (output). What the AI produces in reply to your prompt.
• Token. A small chunk of text that a model reads and writes, often a word or part of a word. Limits on how much you can paste or how long a reply can be are often measured in tokens.
• Context window. How much text a model can take into account at once, including your messages, any pasted documents and its own replies. If a conversation gets very long, earlier parts may be dropped or handled less reliably.
• System prompt and custom instructions. Background instructions that shape how the chatbot behaves across a conversation. Providers set their own system prompts, and many chatbots let you add personal instructions, such as "keep answers short."
• Memory. A feature in some chatbots that saves details between separate conversations. Whether it exists, and how to view or delete what is saved, depends on the product and your settings.
• Temperature. A setting, mostly seen in developer tools, that controls how predictable or varied replies are. Lower values give more consistent answers; higher values give more variety.

Which terms explain how AI answers go wrong?

• Hallucination. When an AI states something false or made up, such as a fake quote, statistic or source, often in a confident tone. It is the single most important term for beginners. Our guide on how to check AI answers for mistakes shows how to catch it.
• Bias. Unfair or one-sided patterns in AI output, often inherited from the training data. It can show up in word choices, assumptions about people, or which viewpoints get presented.
• Grounding and citations. Connecting an answer to specific sources, such as web pages or a document you uploaded, and showing where claims came from. Citations help you verify, but you still need to open them, since a cited source may not say what the summary claims.
• Sycophancy. A tendency for AI to agree with you or tell you what you seem to want to hear. Asking "What are the strongest arguments against this?" helps counter it.

What do common AI tool and feature terms mean?

• Multimodal. Able to work with more than one type of input or output, such as text plus images, audio or files.
• Image generator. A tool that creates pictures from a text description. Many use a technique called a diffusion model, which starts from random visual noise and gradually refines it into an image. See AI image generators for beginners.
• Retrieval-augmented generation (RAG). A method where the system first looks up relevant information, such as search results or company documents, and then uses it to write the answer. It can reduce, but not eliminate, made-up answers.
• Reasoning model. A model designed to work through a problem in more steps before answering. These can do better on multi-step questions but may take longer to reply.
• AI agent. An AI system that can take actions toward a goal, such as searching, filling in forms or using other apps, rather than only chatting. Agents are useful but deserve extra care, because their actions can have real effects.
• API (application programming interface). A way for software to talk to other software. Developers use AI providers' APIs to build AI features into their own apps.
• Open-source and open-weight models. Models whose inner workings (the trained "weights") are published so others can download and run them. Licenses differ, so "open" does not always mean free to use for any purpose.
• Copilot or assistant. Common names for AI helpers built into other software, like a writing assistant inside an email or document app. "Microsoft Copilot" is also a specific product name.

Which privacy and safety terms should you know?

• Data retention. How long a service keeps your conversations and files. Policies vary by provider, plan and settings, so check each service's privacy page.
• Training on your data (opt-out). Some services may use conversations to improve their models unless you change a setting, and some business plans handle this differently. Read our privacy basics guide for safe habits.
• Guardrails. Rules and filters that stop an AI from producing certain harmful content. They are why a chatbot sometimes declines a request.
• Prompt injection. Hidden or sneaky instructions placed in content an AI reads, such as a web page or email, to trick it into doing something you did not ask for. It matters most when AI tools can browse or take actions for you.
• Deepfake. Realistic fake images, video or audio of real people made with AI. Be cautious with surprising clips, especially ones asking for money or urgent action.
• AI detector. A tool that tries to guess whether text or images were made by AI. Results are estimates and can be wrong in both directions, so they should not be treated as proof.

What do plan and pricing terms mean?

• Free tier. A no-cost version of a tool, usually with fewer features or lower limits than paid plans.
• Usage limits (rate limits). Caps on how much you can use a tool in a period of time, such as a number of messages or images. Limits change often, so check the official plan page rather than a blog post.

Deciding whether to pay? Our guide to free vs paid AI tools gives a simple way to test before you upgrade.

Frequently asked questions

What is the difference between AI, machine learning and deep learning?

Think of them as nested circles. AI is the broadest idea: computers doing tasks that seem intelligent. Machine learning is one way to build AI, by learning from examples. Deep learning is a kind of machine learning that uses many-layered neural networks, and it powers most modern chatbots and image tools.

Is a chatbot the same thing as a large language model?

Not quite. The large language model is the engine; the chatbot is the app you talk to. A chatbot usually adds a chat screen, safety rules, and sometimes extras such as web search, file uploads or memory.

Why do AI chatbots hallucinate?

Language models generate text that is likely to fit, based on patterns learned in training. They do not check each sentence against a database of verified facts, so a plausible-sounding but wrong detail can slip in. Verifying names, numbers, quotes and sources is the best defense.

What is a token in simple terms?

A token is a small piece of text, often a short word or part of a longer one. Models process text as tokens, which is why some limits on input length and reply length are measured in tokens rather than words.

Do I need to learn these terms to use AI tools?

No. You can start using a chatbot today with plain sentences. Knowing a handful of terms, especially prompt, context window, hallucination and data retention, simply makes it easier to use the tools well and safely.

Key takeaways

• Most AI chatbots run on large language models, which generate text by predicting likely words.
• A prompt is what you type; tokens and the context window describe how much the model can handle.
• Hallucination, bias and sycophancy are the main ways answers go wrong, so verify what matters.
• Privacy terms such as data retention and training opt-out are worth checking in every tool you use.
• Product features and limits change, so rely on official help pages for current details.