The AI Glossary: 30 Terms Every Beginner Should Know

AI conversations are full of unfamiliar words, and nobody hands you a dictionary. This guide explains 30 of the most common terms in plain language, grouped by theme, so you can follow any AI discussion without guessing.

The basics: what AI actually is

Start with the words that describe the field itself.

• Artificial intelligence (AI): Software that performs tasks we usually associate with human thinking, like writing, recognizing images, or making predictions.
• Machine learning: A branch of AI where software learns patterns from examples instead of being programmed with step-by-step rules.
• Model: The finished, trained piece of AI software that actually does the work, like answering questions or editing photos.
• Training: The process of showing a model huge amounts of example data so it learns patterns. Training happens once, before you ever use the model.
• Training data: The books, websites, images, or other examples a model learned from during training.
• Inference: The moment you use a model: it takes your input and produces an answer. If training is studying, inference is taking the test.
• Neural network: The structure most modern AI is built on, loosely inspired by how brain cells connect, though it works very differently from an actual brain.
• Parameters: The adjustable settings inside a model that training fine-tunes. More parameters usually mean a bigger, more capable model.

How models learn and remember

These terms explain what happens during training and what a model knows.

• Large language model (LLM): A model trained mostly on text, able to understand and produce human language. The chatbots you have heard of are built on LLMs.
• Token: The small chunk of text, often part of a word, that a model reads and writes. When you hear about limits on input size, tokens are the unit being counted.
• Context window: How much text a model can consider at once, including your question and its own reply. A bigger window means longer conversations stay coherent.
• Hallucination: When a model confidently makes up facts, quotes, or details that sound right but are wrong. It happens because models predict likely text, not verified truth.
• Fine-tuning: Extra training on a smaller, specialized set of examples to make a general model better at one particular job.
• Bias: Skewed or unfair patterns in a model's answers, usually inherited from skewed patterns in its training data.
• Knowledge cutoff: The date up to which a model's training data goes. Anything that happened after that date is unknown to it unless it can search the web.
• Multimodal: A model that can work with more than one type of input, such as text, images, and voice, all in the same conversation.

Using AI day to day

The words you will run into while actually typing into AI tools.

• Prompt: The instructions or question you type into an AI tool. Better prompts give better answers.
• Prompt engineering: The practice of writing clearer, more structured prompts to get more useful results.
• System prompt: Hidden instructions set by the tool's maker that shape how the AI behaves, such as its tone or what it refuses to do.
• Zero-shot: Asking the model to do a task it was never shown an example of, relying only on its general training.
• Few-shot: Showing the model two or three examples of what you want before asking, which usually improves the result.
• Temperature: A setting that controls randomness. Low temperature gives predictable, focused answers; high temperature gives more creative, varied ones.
• Chatbot: A conversational interface where you type messages and the AI replies, the most common way beginners use AI.
• Copilot or assistant: AI built into another app, like a writing or office program, that helps you inside the tool you are already using.

Capabilities and tools around AI

Terms for what AI can connect to and produce.

• Generative AI: AI that creates new content, like text, images, music, or video, rather than just sorting or analyzing existing data.
• Retrieval-augmented generation (RAG): A setup where the AI looks up relevant documents before answering, which makes answers more accurate and grounded in real sources.
• API: A way for one piece of software to use an AI service directly, so apps can build AI features without running the model themselves.
• Open-weight / open-source model: A model whose internal files are publicly shared, so anyone can download, study, and run it themselves.
• Agent: An AI setup that can take multiple steps on its own, like searching, reading, and then writing, toward a goal you set.
• Embedding: A way of converting text into numbers so the AI can compare meanings, used behind the scenes for search and recommendations.

Copy this: a one-week learning plan. Day 1: try one chatbot with five simple questions. Day 2: compare two different AI tools on the same task. Day 3: read back your own prompts and rewrite the weakest one. Day 4: ask the AI to explain a topic you know well, and check for hallucinations. Day 5: try an image or voice feature.

Limitations to keep in mind

Knowing the vocabulary does not mean trusting everything. Models hallucinate, repeat biases from their training data, and know nothing past their knowledge cutoff. A glossary teaches you what the words mean, not which claims are true. Keep a habit of checking important answers, especially anything involving money, health, or legal decisions, against reliable sources.