"Half of feeling confident with AI is simply knowing what the words mean. Here are the essentials, in plain English."
Learning Objectives
By the end of this module, you will be able to:
- Recognize the most common Gen AI terms
- Explain each one in simple language
- Read AI articles and tool docs with more confidence
The Essential Terms
Keep this page handy — bookmark it and come back whenever a term trips you up. You don't need to learn it all at once; treat it like a dictionary you glance at when a word snags you mid-article. Over a few weeks of using AI tools, most of these will quietly become second nature.
| Term | Plain-language meaning |
|---|---|
| Generative AI | AI that creates new content (text, images, audio, video) from a prompt. |
| Model | The trained "brain" that generates content. Examples: GPT, Gemini, Claude. |
| LLM (Large Language Model) | A model trained on huge amounts of text to understand and generate language. |
| Prompt | The instruction or question you give the AI. |
| Prompt engineering | The skill of writing prompts that get the results you want. |
| Token | A small piece of text (about ¾ of a word) that models read and write. |
| Context window | How much text the model can consider at once. Go over it and early text is forgotten. |
| Training | The one-time process where a model learns patterns from data. |
| Inference | Using the trained model to answer a prompt — what happens when you chat. |
| Hallucination | When the AI states something false but sounds confident. |
| Bias | Unfair patterns a model picks up from its training data. |
| Fine-tuning | Extra training to specialize a model for a specific task or style. |
| Embedding | Turning text into numbers that capture meaning, so similar ideas sit close together. |
| RAG | Retrieval-Augmented Generation — fetching trusted facts and giving them to the model before it answers. |
| Agent | An AI that can use tools and take actions, not just chat. |
| Multimodal | A model that handles more than one type of content (e.g., text and images). |
| Diffusion model | The technique behind most image generators — turning random noise into a picture. |
| System prompt | Hidden instructions that set an AI's role, rules, and tone. |
| Temperature | A setting for how "creative" vs. predictable the output is. |
| Open-source model | A model whose code/weights are publicly available to run and adapt. |
Grouping the Terms
Twenty definitions can feel like a lot, so it helps to see how they cluster. Most fall into just a few families:
- The basics — generative AI, model, LLM, prompt, prompt engineering. What the tool is and how you talk to it.
- How it works inside — token, context window, training, inference, embedding. The moving parts under the hood.
- The risks — hallucination, bias. The pitfalls to stay alert to.
- The building blocks — RAG, agent, system prompt, fine-tuning, diffusion model. How people build useful things on top of a model.
- The settings and flavors — multimodal, temperature, open-source model. Options and varieties you'll run into.
Seeing the map this way means you don't have to hold twenty separate facts in your head — just five small groups.
The Terms in a Single Story
Imagine you open a chat assistant to plan a trip. The model is the trained brain answering you, and because it's a large language model, it works with language. You type a prompt, and shaping it well is prompt engineering. Behind the scenes your words become tokens, and everything considered at once must fit in the context window. It answers using what it learned during training — and answering you right now is inference. Invent a hotel that doesn't exist? That's a hallucination; always suggesting the same kind of place could be bias. Connect it to live travel data and you'd use RAG; let it actually book the flight and it becomes an agent. In one short paragraph, more than half the glossary showed up naturally — exactly how you'll meet these words in real life.
Explain like I'm new: Learning these words is like learning the names of tools in a workshop. You don't memorize a dictionary before you start building — you pick up each name the first time you reach for that tool, and soon the whole workshop feels familiar.
Key idea: You don't need to memorize these — just recognize them. Once these twenty words feel familiar, most AI articles, tutorials, and tool settings suddenly make sense.
Key Takeaway: These twenty terms cover almost everything a beginner meets: what a model and prompt are, how tokens and the context window work, the difference between training and inference, the risks (hallucination, bias), and the building blocks (RAG, agents, embeddings, diffusion). Bookmark this page as your quick reference.