AI news is full of jargon. One article talks about agents and tokens, another about hallucinations and fine-tuning, and a third about multimodal models. If you have ever nodded along without being sure what a term meant, this glossary is for you. Yahoo Tech recently ran an explainer on the AI jargon that has become common, and the idea is worth repeating in a form you can bookmark. Below are 25 terms, grouped by theme, each explained in plain English with a simple example. You do not need a technical background to follow any of them.
In this article
- The Basics
- Language Models and Chatbots
- How Models Are Built and Improved
- Problems and Safety
- Newer Ideas You Will See in the News
- Quick Reference Table
- How to Use This Glossary
- Common Misunderstandings
- How These Terms Show Up in Real Products
- Five Questions to Ask About Any AI Feature
- A Short Example in Plain English
- Frequently Asked Questions
- The Bottom Line
The Basics
1. Artificial intelligence (AI)
A broad term for computer systems that perform tasks that normally need human intelligence, such as understanding language, recognising images or making decisions. The US National Institute of Standards and Technology describes its work on trustworthy AI at nist.gov/artificial-intelligence.
2. Machine learning
A way of building AI by letting computers learn patterns from data instead of following hand-written rules. A spam filter that improves as it sees more emails is a classic example.
Also readMeta’s New Personal Agent Protocol: 7 Things to Know Before You Let AI Shop for You
3. Deep learning
A type of machine learning that uses many-layered neural networks. It powers most modern image recognition, speech recognition and language tools.
4. Neural network
A mathematical system loosely inspired by the brain, made of layers of connected nodes that process information. Training adjusts the connections so that the network gets better at its task.
5. Algorithm
A set of step-by-step instructions for solving a problem. In AI, algorithms define how a system learns from data or makes a prediction.
Language Models and Chatbots
6. Large language model (LLM)
An AI model trained on huge amounts of text to predict and generate language. ChatGPT, Claude and Gemini are built on LLMs. Our explainer on what a large language model is goes deeper.
7. Chatbot
A program that talks with you in natural language. Some follow fixed scripts, and modern ones use LLMs to produce flexible responses.
8. Prompt
The instruction or question you give an AI. Clear prompts with context, goals and examples usually produce better answers.
9. Token
A small chunk of text, such as a word or part of a word, that a model processes. Limits and pricing for AI tools are often measured in tokens.
10. Context window
The amount of text a model can consider at once, including your prompt and its answer. A larger window lets you work with longer documents.
How Models Are Built and Improved
11. Training data
The information used to teach a model. Its quality, variety and fairness shape how the model behaves.
12. Parameters
The internal numbers a model adjusts during training. Bigger models have more parameters, but size alone does not guarantee quality.
13. Fine-tuning
Further training of an existing model on specialised data so that it performs better in a particular area, such as legal documents or customer support.
14. Inference
The stage where a trained model is used to produce answers. Every time you ask a chatbot something, it performs inference.
15. Benchmark
A standard test used to compare AI models. Benchmarks are useful but do not always reflect real-world performance.
Problems and Safety
16. Hallucination
When an AI states something false but plausible, such as an invented quote or a fake source. Always verify important facts.
17. Bias
Systematic unfairness in a model’s output, often reflecting imbalances in the data it learned from.
18. Alignment
The effort to make AI systems behave in line with human values and intentions.
19. Guardrails
Rules and filters that limit what an AI will say or do, for example refusing harmful requests.
20. Deepfake
Synthetic video, audio or images that convincingly imitate real people. They can be used creatively, but also for fraud and misinformation.
Newer Ideas You Will See in the News
21. AI agent
An AI system that can take actions toward a goal, such as booking, searching or filling in forms, instead of just answering questions. Agents raise new questions about safety and oversight.
22. Multimodal
A model that can work with more than one type of input or output, such as text, images, audio and video.
23. Retrieval-augmented generation (RAG)
A method where the AI looks up relevant documents before answering, so its response is grounded in specific sources. It can reduce, but not remove, hallucinations.
24. On-device AI
AI that runs directly on your phone or laptop instead of in the cloud, which can improve speed and privacy.
25. Generative AI
AI that creates new content, such as text, images, music or code, rather than only classifying or predicting. Practical use is covered in our guide to how to use ChatGPT for work.
Quick Reference Table
| Term | One-line meaning |
| LLM | A model trained on text to generate language |
| Prompt | What you ask the AI |
| Token | A chunk of text the model reads |
| Hallucination | A confident but false answer |
| Fine-tuning | Extra training for a specific job |
| RAG | Looking things up before answering |
| Agent | AI that takes actions |
| Multimodal | Handles text, images, audio and more |
| Inference | Using a trained model |
| Guardrails | Rules that limit AI behaviour |
How to Use This Glossary
- Skim the groups first to get the big picture
- Come back to terms as you meet them in articles and product pages
- Test your understanding by explaining a term to a friend in one sentence
- Notice which terms marketers use loosely, such as agent or AI-powered, and ask what the product actually does
- Keep learning, since AI vocabulary changes quickly
Common Misunderstandings
- AI is not conscious or alive; it identifies patterns in data
- Larger models are not always better for every task
- Confident language does not mean a correct answer
- AI does not know today’s news unless it is connected to current sources
- Different AI products can use the same underlying model with different settings
How These Terms Show Up in Real Products
Knowing a word is one thing, and spotting it in the wild is another. When a company says its app is “AI-powered”, it usually means that it uses a machine learning model for one feature, such as search, recommendations or summaries. When it says it offers an “assistant”, it often means an LLM wrapped in a chat window. When you see “agent”, ask what actions it can take and who approves them. When you see “private” or “secure AI”, ask where the model runs and what happens to your data. Those simple questions cut through marketing quickly.
Five Questions to Ask About Any AI Feature
- What is the model actually doing: generating text, classifying something, recommending items or taking actions?
- Where does the processing happen, on your device or in the cloud?
- What data does it use, and is it kept or used to train future models?
- How does it handle mistakes, and can you correct or override it?
- Is there a human in the loop for important decisions?
You do not need to be an engineer to ask these questions, and a clear answer is a good sign that a company has thought carefully about its product.
A Short Example in Plain English
Imagine you ask an assistant to plan a weekend trip. The assistant is built on a large language model that was trained on huge amounts of text. Your request is the prompt, which gets split into tokens, and the model uses its context window to hold everything in view. It may look up travel information (retrieval) and then generate a plan. If it invents a restaurant that does not exist, that is a hallucination. If it can also book the hotel for you, it is acting as an agent, and that is the moment to check permissions and payment settings carefully.
Frequently Asked Questions
What is the difference between AI and machine learning?
AI is the broad field, and machine learning is a method within it that learns from data.
What is a hallucination in AI?
It is an answer that sounds confident but is false or made up.
What is an AI agent?
It is a system that can take actions toward a goal, such as using tools or completing tasks, not just chat.
Do I need to understand all of this to use AI?
No. A few core terms, such as prompt, LLM and hallucination, are enough to use AI tools sensibly.
The Bottom Line
AI vocabulary can feel intimidating, but most of it describes simple ideas. A model learns from data, you give it a prompt, it generates an answer, and sometimes that answer is wrong. With the 25 terms above, you can follow most AI news, ask better questions of vendors and use tools with more confidence. Bookmark this page and return whenever a new buzzword appears.
1 Comment