Artificial intelligence has moved from research labs into search engines, office tools, phones, classrooms, hospitals, marketing dashboards, and group chats. Along with it came a flood of new vocabulary. Suddenly, everyone is talking about prompts, tokens, multimodal models, hallucinations, and AI agents as if these words have always been part of everyday conversation.
If you have ever nodded along while secretly wondering what a large language model actually is, you are not alone. AI jargon can sound intimidating, but most of the core ideas are easier to understand than they first appear. This glossary breaks down the most common AI terms in plain English, so the next time they come up, you can join the conversation with confidence.
Artificial Intelligence
Artificial intelligence, or AI, is the broad field of creating computer systems that can perform tasks we usually associate with human intelligence. That might include understanding language, recognizing images, making predictions, solving problems, generating text, or learning from patterns in data.
AI does not necessarily mean a machine is conscious or thinking like a person. In most practical uses, AI is software that has been trained to spot patterns and produce useful outputs based on those patterns.
Machine Learning
Machine learning is a branch of AI where systems improve by learning from data instead of being programmed with every single rule by hand. For example, rather than telling a program exactly what every spam email looks like, engineers can train it on many examples of spam and non-spam messages so it learns the difference.
Machine learning powers recommendation engines, fraud detection tools, voice assistants, image recognition, predictive analytics, and many of the AI features people use every day.
Deep Learning
Deep learning is a type of machine learning inspired by the structure of the human brain, using layered systems called neural networks. These layers help software identify complex patterns, such as faces in photos, spoken words in audio, or meaning in a paragraph of text.
The word deep refers to the many layers in the network, not to the system having deep thoughts. Deep learning is one of the reasons modern AI has become so powerful.
Neural Network
A neural network is a computing system made of connected nodes that process information. It learns by adjusting the strength of those connections as it looks at training examples.
Think of it like a flexible pattern-recognition machine. Give it enough relevant examples, and it can learn to classify images, translate languages, predict trends, or generate realistic-sounding text.
Generative AI
Generative AI refers to AI systems that can create new content. This content might be text, images, music, code, video, product ideas, lesson plans, or summaries.
Tools that write emails, design graphics, generate social media captions, or create images from text descriptions are all examples of generative AI. The key idea is creation: the AI is producing something new based on what it has learned from data.
Large Language Model
A large language model, often shortened to LLM, is an AI model trained on massive amounts of text so it can understand and generate language. Chatbots, writing assistants, coding copilots, and many search-based AI tools rely on LLMs.
An LLM predicts what text should come next based on context. That simple description hides a lot of complexity, but it explains why these systems are good at drafting, summarizing, translating, answering questions, and imitating different writing styles.
Model
In AI, a model is the trained system that produces results. You can think of it as the finished engine after the learning process has taken place.
A model might identify whether a photo contains a dog, predict next month’s sales, recommend a song, or answer a question. Different models are designed and trained for different kinds of tasks.
Training Data
Training data is the information used to teach an AI system. It can include documents, websites, books, images, audio clips, code, spreadsheets, videos, or labeled examples.
The quality of training data matters enormously. If the data is incomplete, biased, outdated, or incorrect, the model can learn bad patterns and produce unreliable results.
Parameters
Parameters are internal settings a model adjusts during training. They help determine how the model responds to input. In large AI systems, there can be billions of parameters working together.
More parameters can sometimes mean a model has more capacity to learn complex patterns, but bigger is not always better. Data quality, architecture, training methods, and safety controls also matter.
Prompt
A prompt is the instruction or question you give to an AI system. If you ask a chatbot to write a product description, summarize a report, or explain photosynthesis to a ten-year-old, that request is the prompt.
Good prompts are clear, specific, and include helpful context. The better your instructions, the more likely you are to get a useful response.
Prompt Engineering
Prompt engineering is the practice of writing better prompts to get better AI results. It can involve giving the AI a role, specifying the format you want, adding examples, defining the audience, or setting constraints.
For example, instead of asking, Write a blog intro, you might ask, Write a friendly 120-word introduction for small business owners about using AI to save time, avoiding technical jargon. That second prompt gives the AI much more to work with.
Token
A token is a small unit of text that an AI model processes. A token can be a whole word, part of a word, punctuation, or even a space, depending on how the model breaks language down.
Tokens matter because AI systems often have limits on how many they can handle at once. Longer documents, detailed prompts, and long responses use more tokens.
Context Window
The context window is the amount of information an AI model can consider at one time. It includes your prompt, previous conversation, uploaded text, and the model’s response.
If a conversation or document is too long for the context window, the AI may lose track of earlier details. Larger context windows allow models to work with longer documents and more complex instructions.
Inference
Inference is what happens when a trained AI model is used to generate an answer or make a prediction. Training is the learning stage; inference is the using stage.
When you ask an AI tool to summarize an article or create a meal plan, the model is performing inference. It is applying what it learned during training to your specific request.
Hallucination
A hallucination is when an AI system produces information that sounds confident but is false, misleading, or made up. This can include fake citations, incorrect facts, invented statistics, or nonexistent legal cases.
Hallucinations happen because language models are designed to generate plausible responses, not to guarantee truth. For important decisions, always verify AI-generated information with reliable sources.
AI Bias
AI bias occurs when an AI system produces unfair or skewed results because of problems in the data, design, or deployment process. If historical data reflects discrimination or imbalance, the model may repeat or amplify those patterns.
Bias is a serious issue in areas like hiring, lending, policing, healthcare, and education. Responsible AI development includes testing for bias and reducing harm wherever possible.
Fine-Tuning
Fine-tuning is the process of taking an existing AI model and training it further on a more specific dataset. This can help the model perform better for a particular industry, brand voice, task, or type of content.
For example, a company might fine-tune a model on its support documents so it can answer customer questions in a more accurate and brand-consistent way.
RAG
RAG stands for retrieval-augmented generation. It is a technique that allows an AI system to retrieve relevant information from a trusted source before generating an answer.
Instead of relying only on what the model learned during training, a RAG system can pull from documents, databases, knowledge bases, or company files. This can make answers more accurate, current, and specific.
Embeddings
Embeddings are numerical representations of words, sentences, images, or other data. They help AI systems understand similarity and meaning.
For example, an AI search tool can use embeddings to understand that affordable running shoes and budget sneakers are related, even though the exact words are different. Embeddings are especially useful for semantic search, recommendations, clustering, and RAG systems.
Multimodal AI
Multimodal AI can work with more than one type of input or output, such as text, images, audio, video, or code. A multimodal model might analyze a chart, answer questions about a photo, transcribe speech, or generate an image from a written description.
This is one of the biggest shifts in modern AI. Instead of being limited to text-only conversations, AI tools are becoming better at understanding the many ways humans communicate.
AI Agent
An AI agent is a system designed to take steps toward a goal, often by using tools, making decisions, and completing tasks with some degree of autonomy. A simple chatbot answers questions; an agent might research options, compare prices, draft an email, and schedule a meeting.
Agents are still developing, and their reliability varies. They can be powerful when supervised carefully, but they should not be trusted blindly with sensitive or high-stakes tasks.
Copilot
An AI copilot is an assistant built into a workflow to help you complete tasks faster. Coding copilots suggest code, writing copilots help draft documents, and business copilots can summarize meetings or analyze data.
The term copilot is useful because it suggests partnership rather than replacement. The human is still responsible for judgment, direction, and final decisions.
Guardrails
Guardrails are rules, filters, and safety measures designed to keep AI behavior within acceptable boundaries. They may prevent the model from sharing harmful instructions, exposing private data, or generating certain types of content.
Guardrails are not perfect, but they are an important part of making AI tools safer and more reliable, especially in business, healthcare, education, and public-facing products.
What This All Means
You do not need to become a machine learning engineer to understand the AI conversation. In most cases, you just need a working vocabulary. Once you know what terms like prompt, model, token, hallucination, and RAG mean, the technology feels much less mysterious.
AI is moving quickly, and the language around it will keep changing. New buzzwords will appear, old terms will evolve, and some overhyped phrases will fade away. But the fundamentals will help you ask better questions, use tools more effectively, and separate genuine innovation from shiny jargon.
The next time someone casually mentions embeddings, context windows, or AI agents, you will not have to simply nod along. You will know what they mean, why it matters, and where to stay curious.