AI Glossary: 69 AI Terms Explained in Simple Words
The 69 AI terms you will meet most in AI news, courses and interviews, from LLM and token to RAG and hallucination, each explained in two or three plain sentences.
Start here:Artificial intelligence (AI)Machine learning (ML)Generative AILarge language model (LLM)PromptTokenAI hallucinationRetrieval-augmented generation (RAG)AI agentTemperature
69 AI terms
AI basics terms
The words you meet first when you start learning AI.
Algorithm
- A step-by-step set of instructions for solving a problem. In machine learning, the algorithm is the method used to learn from data, such as linear regression or a decision tree.
Artificial general intelligence (AGI)(AGI)
- A hypothetical AI that can learn and perform almost any intellectual task a human can, across domains. Today's AI systems are strong at many tasks but are not generally accepted as AGI.
Artificial intelligence (AI)(AI)
- The field of building computer systems that do tasks which normally need human intelligence, such as understanding language, recognising images, making predictions and taking decisions.Full explanation
Computer Vision
- Computer vision is a field of artificial intelligence that teaches computers to see and understand images and videos. It lets machines recognise objects, read text, detect faces and make decisions based on what a camera captures.Full explanation
Deep learning
- A type of machine learning that uses neural networks with many layers to learn complex patterns directly from raw data such as images, audio and text.Full explanation
Generative AI(GenAI)
- AI that creates new content, such as text, images, code, audio or video, by learning patterns from large amounts of existing data. ChatGPT and image generators are common examples.Full explanation
Machine learning (ML)(ML)
- A branch of AI where systems learn patterns from data instead of following hand-written rules, so they improve at a task like spam detection or price prediction as they see more examples.Full explanation
Model
- The trained result of a machine learning algorithm: a file of learned numbers (parameters) that takes an input, such as a sentence or an image, and produces an output, such as a label or a reply.
Natural language processing (NLP)(NLP)
- The area of AI focused on understanding and generating human language, covering tasks like translation, sentiment analysis, chatbots and summarisation.
Neural network
- A model made of layers of connected units (neurons) that pass numbers forward and adjust their connection strengths (weights) during training to learn patterns from data.Full explanation
Reinforcement learning
- Learning by trial and error, where an agent takes actions, receives rewards or penalties, and gradually learns a strategy that earns the most reward over time.
Supervised learning
- Machine learning from labelled examples, where each input comes with the correct answer, for example emails marked spam or not spam, so the model learns to predict labels for new data.
Unsupervised learning
- Machine learning on data without labels, where the model finds structure on its own, such as grouping customers with similar buying habits (clustering).
Prompting terms
Terms for talking to AI models and getting better answers.
Chain-of-Thought Prompting
- Chain-of-thought (CoT) prompting is a technique where you ask an AI model to write out its intermediate reasoning steps before giving a final answer. Breaking a problem into steps often improves accuracy on maths, logic and multi-step questions.Full explanation
Context engineering
- Designing everything an AI model sees in its context window, including instructions, retrieved documents, tool results and conversation history, so it has the right information to do a task well.
Few-Shot Prompting
- Few-shot prompting is a technique where you include a small number of worked examples, usually two to five input and output pairs, inside the prompt so the AI model copies the pattern, format and style for your real question.Full explanation
Prompt
- The instruction or input you give an AI model, such as a question, a task description or examples. The clarity of the prompt strongly affects the quality of the output.
Prompt engineering
- The practice of writing and refining prompts, with clear roles, context, examples and output formats, to get accurate and consistent results from AI models.Full explanation
Prompt injection
- An attack where hidden or malicious instructions inside user input, documents or web pages trick an AI model into ignoring its rules. OWASP lists it as a top risk for LLM applications.
System prompt
- Hidden standing instructions given to a chatbot or AI app that set its role, tone and rules before the user types anything, for example "You are a polite support agent for a bank".
Zero-Shot Prompting
- Zero-shot prompting means asking an AI model to do a task using only an instruction, with no examples of the expected answer. The model relies entirely on what it learned during training to understand the task and respond.Full explanation
Training and models terms
How models are built, trained, tested and shipped.
Backpropagation
- The method neural networks use to learn: the error at the output is sent backwards through the layers to work out how much each weight contributed, so the weights can be adjusted.
Benchmark
- A standard test set used to compare AI models on a skill, such as maths, coding or general knowledge. Benchmark scores are useful but do not always reflect real-world performance.
Dataset
- An organised collection of examples, such as images, text or table rows, used to train, validate or test a machine learning model.
Distillation(Knowledge distillation)
- Training a smaller "student" model to imitate the outputs of a larger "teacher" model, so the student keeps much of the quality while being cheaper and faster to run.
Epoch
- An epoch is one complete pass of the entire training dataset through a machine learning model. Training usually runs for many epochs, and after each one the model has seen every training example once.Full explanation
Fine-tuning
- Further training a pretrained model on a smaller, task-specific dataset so it learns a particular style, format or skill, such as answering in a company's tone.Full explanation
Hyperparameter
- A setting chosen before training that controls how a model learns, such as the learning rate, batch size or number of epochs, unlike parameters, which the model learns itself.
Large language model (LLM)(LLM)
- A large neural network, usually a transformer, trained on huge amounts of text to predict the next token. That skill lets it write, summarise, translate, answer questions and generate code.Full explanation
Loss function
- A formula that measures how wrong a model's predictions are during training. Training adjusts the model's parameters to make this loss as small as possible.
MLOps
- The practices and tools for deploying, monitoring and updating machine learning models in production reliably, similar to DevOps for software. LLMOps applies the same ideas to LLM apps.
Open-Source AI Model
- An open-source AI model is a model whose weights, and ideally its code and training details, are publicly released under a licence that lets anyone use, study, modify and share it. Many popular "open" models are more precisely open-weight.Full explanation
Overfitting
- Overfitting happens when a machine learning model learns its training data too closely, including noise and random quirks, so it scores very well on that data but performs poorly on new, unseen data.Full explanation
Parameters
- Parameters are the internal numbers (weights and biases) that an AI model learns during training. They store everything the model has picked up from data, and the parameter count, such as 7 billion, is a rough measure of model size.Full explanation
RLHF
- RLHF (Reinforcement Learning from Human Feedback) is a training method where people rank or rate a model's answers, a reward model learns those preferences, and the language model is then trained to produce answers people prefer.Full explanation
Small language model (SLM)(SLM)
- A language model with far fewer parameters than frontier LLMs, designed to be cheaper and faster, and often small enough to run on a laptop or phone for focused tasks.
Synthetic data
- Artificially generated data, often created by AI models or simulations, used to train or test models when real data is scarce, expensive or too sensitive to use.
Training Data
- Training data is the set of examples an AI or machine learning model learns from. The model studies patterns in this data, such as labelled photos or text, and adjusts itself so it can make predictions on new data.Full explanation
How LLMs work terms
What happens inside a large language model when it answers.
AI Inference
- AI inference is the stage where a trained model uses what it learned to produce an output for new input, such as answering a question, labelling a photo or predicting a price. Training builds the model; inference is using it.Full explanation
Attention mechanism
- The part of a transformer that lets each token weigh how relevant every other token is, so the model can focus on the right words when building meaning.Full explanation
Context window
- The maximum number of tokens a model can consider at once, counting both your input and its reply. Text beyond the limit must be shortened, summarised or retrieved selectively.Full explanation
Embedding
- A list of numbers that represents the meaning of text, an image or other data, so that similar meanings sit close together. Embeddings power semantic search and RAG.Full explanation
Knowledge cutoff
- The date after which a model has no training data. Without search or retrieval, it does not know about events after this date and may give outdated answers.
Latency
- How long an AI system takes to respond. For LLM apps it is often measured as time to the first token and total time, and it grows with model size and output length.
Mixture of Experts (MoE)
- Mixture of experts (MoE) is a neural network design where a model contains many sub-networks called experts, and a small router picks only a few of them for each input token. The model can be huge while each token uses a fraction of it.Full explanation
Quantization
- Quantization is a technique that stores an AI model's numbers (its weights, and sometimes activations) with fewer bits, such as 8 or 4 instead of 16. It shrinks memory use and often speeds up inference, with some loss in accuracy.Full explanation
Reasoning model
- An LLM trained to work through problems step by step internally before answering, which usually improves results on maths, coding and logic tasks at the cost of extra time and tokens.
Temperature
- Temperature is a setting that controls how random a language model's output is. Low values (near 0) make answers focused and repeatable; higher values make them more varied and creative, but also more likely to drift or make mistakes.Full explanation
Token
- A small chunk of text, often part of a word, that language models read and write. In English one token is roughly three quarters of a word, and API pricing is usually per token.Full explanation
Tokenizer
- The tool that splits text into tokens and converts them to numbers a model can process. Different models use different tokenizers, so the same sentence can have different token counts.
Transformer
- The neural network architecture behind most modern LLMs, introduced by Google researchers in 2017. It uses attention to process all tokens in parallel and understand how words relate.Full explanation
Safety and ethics terms
Risks, limits and the guardrails around AI systems.
AI Bias
- AI bias is when an artificial intelligence system produces results that are unfair to certain people or groups, usually because it learned from unbalanced or prejudiced data, was labelled poorly, or was designed with flawed assumptions.Full explanation
AI Guardrails
- AI guardrails are rules, filters and checks placed around an AI model to keep its inputs and outputs safe, accurate and on-topic. They can block harmful requests, hide personal data, enforce formats and stop the model from taking risky actions.Full explanation
AI hallucination(Hallucination)
- When an AI model confidently produces false or made-up information, such as fake citations or invented facts, because it predicts likely text rather than checking truth.Full explanation
Deepfake
- A deepfake is a fake video, image or audio clip made with AI that shows a real person saying or doing something they never did. It usually swaps faces, copies a voice or changes lip movements to look real.Full explanation
Grounding
- Connecting a model's answer to trusted sources, such as retrieved documents or search results, so its output is based on real information and can cite where it came from.
Apps and tools terms
The building blocks of real AI products and agents.
Agentic AI
- AI systems designed to act with some autonomy, planning and carrying out multi-step tasks with tools and limited human input. The term describes a degree of independence rather than one product.Full explanation
AI agent
- An AI system that works towards a goal by planning steps, calling tools such as search or APIs, checking the results and deciding what to do next, rather than only replying once.Full explanation
API(Application programming interface)
- A defined way for programs to talk to each other. AI APIs let developers send prompts to a model over the internet and receive its responses inside their own apps.
Chatbot
- A program that holds a conversation with users in text or voice. Modern chatbots such as ChatGPT, Claude and Gemini are built on large language models.
Diffusion model
- A generative model that creates images (and increasingly video and audio) by starting from random noise and removing it step by step, guided by a text prompt.
Function calling(Tool calling)
- A model feature where the AI outputs a structured request to use a tool, such as a weather API, your code runs it, and the result is sent back for the model to use in its answer.
GPU(Graphics processing unit)
- A chip built to do many calculations in parallel. GPUs are the main hardware used to train and run deep learning models because neural networks are mostly large matrix maths.
Model Context Protocol (MCP)(MCP)
- An open standard, released by Anthropic in November 2024, that lets AI apps connect to tools and data sources through one common interface, like a USB-C port for AI.Full explanation
Multimodal AI
- AI that can understand or generate more than one type of data, such as text, images, audio and video, for example describing a photo or answering questions about a chart.
Retrieval-augmented generation (RAG)(RAG)
- A method that retrieves relevant documents from your own data and adds them to the prompt, so the model answers from that information instead of memory alone.Full explanation
Semantic Search
- Semantic search is a way of searching that matches the meaning of your query instead of its exact words. It uses AI embeddings to find results that are about the same idea, even when they share no keywords.Full explanation
Vector database
- A database that stores embeddings and quickly finds the ones most similar to a query, which is how RAG systems fetch relevant text in milliseconds.Full explanation
Vibe coding
- Building software by describing what you want in plain language and letting an AI coding tool write most of the code, then testing and steering it through conversation.
In the Unrot AI app
Turn these AI terms into real understanding with 5-minute lessons.
- Interactive lessons
- Quizzes
- Progress tracking
- Daily AI news
AI glossary: frequently asked questions
What is an AI glossary?
An AI glossary is a list of artificial intelligence terms with short, plain-English definitions. It helps beginners, students and professionals understand words like LLM, token, RAG and hallucination without reading a full course first.
What are the most important AI terms for beginners?
Start with artificial intelligence, machine learning, generative AI, large language model (LLM), prompt, token, hallucination, RAG, AI agent and temperature. Those ten words cover most of what you will hear in AI news, courses and interviews.
Where can I learn AI terms in Hindi?
Our AI terms in Hindi page explains the most common AI words with simple Hindi and Hinglish meanings, such as what AI, LLM and prompt engineering mean, so you can learn in the language you think in.
Explore more free AI guides
AI ConceptsLLMs, RAG, agents and more, explained simply19 conceptsInterview PrepAI interview questions with model answers17 rolesAI ProjectsProject ideas and step-by-step builds17 guidesLearning PathsFree week-by-week AI paths by role16 pathsBest ResourcesHonest lists of courses, apps and books15 lists