AI has its own language, and most of it sounds harder than it is. This is the AI terms explained glossary, 30 of the words you actually run into, each in one or two plain sentences with no jargon and no hype. Skim it, search it, or bookmark it for the next time someone drops a term like token or RAG and expects you to nod along. We grouped the terms so related ideas sit together, starting with the basics and ending with the newer stuff everyone is arguing about.
You don’t need to memorize any of this. The goal is that when you see one of these AI terms explained below, you know roughly what it means and why it matters. Let’s start with the foundation.
The basics
Start here. These are the words the rest of the glossary quietly builds on, so a few minutes with this section makes everything after it easier.
Artificial intelligence. Software that does things we used to think needed a human brain, like understanding language, recognizing images, or making a decision. It’s a broad umbrella, not one single technology.
Machine learning. The main way modern AI is built. Instead of a person writing every rule, the system learns patterns from lots of examples and gets better with more data.
Neural network. The structure most AI learning happens in, loosely inspired by how brain cells connect. It’s layers of simple math units that pass signals along and adjust as they learn.
Large language model. Often shortened to LLM, this is the kind of AI behind chatbots like ChatGPT and Claude. It’s trained on huge amounts of text to predict the next word, which turns out to be enough to hold a conversation, write, and summarize.
Generative AI. AI that creates new content on request, whether that’s text, images, audio, or code. If it makes something rather than just sorting or scoring, it’s generative.
Model. The trained system itself, the thing that actually does the work. When people say “which model are you using,” they mean which specific trained AI, like GPT or Gemini.
Parameters. The internal settings a model tunes during training, sometimes numbering in the billions. More parameters can mean more capability, but bigger is not always better.
How models learn and run
This group is about where a model’s smarts come from and what actually happens each time you press send.
Training. The process of feeding a model examples so it learns patterns. It happens once, up front, and is expensive and slow, which is why models have a knowledge cutoff date.
Fine tuning. Taking a model that already knows a lot and training it a little more on a narrower set of examples, so it gets better at a specific job or style.
Token. The small chunk of text a model actually reads and writes, usually a word or piece of a word. Pricing and length limits are counted in tokens, not words, and roughly four characters make a token, so a short email is a few hundred tokens.
Prompt. Whatever you type to the AI, your question or instruction. The prompt is the single biggest thing you control, and small changes to it can change the answer a lot.
Context window. How much text a model can hold in mind at once, measured in tokens. Go past it and the model starts forgetting the earliest parts of a long chat or document, which is why a very long conversation can feel like it lost the plot.
Inference. The moment the model actually answers you, using what it learned in training. Every message you send is an inference, and it’s what you pay for per use.
Embedding. A way of turning words or images into a list of numbers that captures their meaning, so a computer can measure how similar two things are. It’s the quiet engine behind search and recommendations.
Using AI well
These terms show up the moment you try to get sharper results out of a tool instead of settling for the first answer.
Prompt engineering. The skill of writing clear instructions to get better results, like giving examples or asking for a specific format. It sounds fancy but mostly means being specific about what you want.
Retrieval augmented generation. Usually shortened to RAG, this feeds a model your own documents at the moment you ask, so it answers from your facts instead of guessing from memory. It’s how a company help desk bot can quote your actual return policy instead of making one up.
Multimodal. A model that handles more than just text, so it can also see images, hear audio, or read a chart. Most of the leading assistants are multimodal now.
System prompt. The hidden instruction that sets a model’s role and rules before you ever type, like telling it to be a friendly support agent. You usually don’t see it, but it shapes everything.
Temperature. A setting that controls how predictable or creative the output is. Low temperature gives safe, repeatable answers, higher gives more variety and surprise.
API. A way for other software to talk to an AI model directly, without a chat window. It’s how apps bake AI features into their own products.
The new wave
The 2026 buzzwords, sorted so you can tell the real ideas from the marketing gloss.
Foundation model. A large, general model trained on broad data that many other tools are built on top of. Think of it as the base engine that gets adapted for specific jobs.
AI agent. An AI that does not just answer but takes actions toward a goal, using tools and working through several steps. We cover this more in our guide to what agentic AI is.
Agentic AI. The broader idea of AI that plans, uses tools, and checks its own work rather than responding one message at a time. It’s the buzzword of 2026, with real uses and real limits.
Reasoning model. A model built to work through a problem in steps before answering, which helps on math, logic, and coding. Our explainer on what reasoning models actually do goes deeper.
Open weight model. A model whose trained settings are released publicly, so anyone can download and run it. It’s different from fully open source, since the training data usually stays private.
Safety and limits
Every term in this group is really a reason to keep a human in the loop.
Hallucination. When a model states something false with total confidence, like inventing a fake quote or a court case that never existed. It’s the single biggest reason to check important AI output before you rely on it.
Bias. When a model’s answers unfairly favor or harm certain groups, usually because that pattern was in its training data. It’s a real problem in sensitive uses like hiring.
Guardrails. The rules and filters that keep a model from doing harmful or off limits things. They are why an assistant will refuse some requests.
Alignment. The ongoing effort to make AI actually do what people want and intend, safely. It’s easy to state and genuinely hard to get right.
AI literacy. Knowing enough about how AI works to use it well and spot when it’s wrong. Reading a glossary like this one counts as a start.
How to use this glossary
Keep this page handy and come back when a term trips you up. If you learn just a handful, make them prompt, token, context window, hallucination, and agent, since those five come up in almost every real conversation about using AI. For the terms with a link, click through when you want the full plain English version rather than the one line summary.
Frequently asked questions
What is the most important AI term to understand?
Prompt, because it’s the one you control. Everything the AI gives back starts with what you asked and how clearly you asked it, so getting good at prompts improves every result.
Is a large language model the same as AI?
No. A large language model is one type of AI, the kind behind chatbots. Artificial intelligence is the wider umbrella that also covers image recognition, recommendations, and much more.
What is the difference between generative AI and agentic AI?
Generative AI makes content when you ask. Agentic AI uses that same intelligence to complete a multi step task on its own, planning and using tools along the way.
Why do AI models hallucinate?
Because they predict likely text rather than looking up verified facts, so a confident guess can come out as a made up detail. That’s why you check anything that matters.
This glossary is general educational information about common AI terms, written to be accurate as of 2026. AI moves fast and definitions shift, so treat this as a plain English starting point, not a technical or legal reference.