DEV Community

Cover image for A-Z AI Glossary
preeti deshmukh
preeti deshmukh

Posted on

A-Z AI Glossary

AI Glossary: A to Z
An A-to-Z glossary of AI terms, created with help from AI itself. Because in 2026, the best way to study AI is apparently to ask AI itself. 🤣

Written for beginners and practitioners alike. Each term includes a plain English definition and a real-world example.


Quick Navigation

A · B · C · D · E · F · G · H · I · J · K · L · M · N · O · P · Q · R · S · T · U · V · W · X · Y · Z

↑ Back to top


A

Term Definition Example
Agent (AI Agent) An AI system that perceives its environment, makes decisions, and takes autonomous actions to achieve a goal A coding agent that writes, runs, and debugs its own code without human intervention
AGI (Artificial General Intelligence) A hypothetical AI that can match or exceed human-level intelligence across any task — does not yet exist Often cited as a long-term goal by companies like OpenAI and DeepMind
AI (Artificial Intelligence) The field of computer science focused on building machines that can perform tasks normally requiring human intelligence ChatGPT writing an essay, an algorithm detecting cancer in X-rays
AI Ethics The principles and practices for developing and deploying AI in ways that are fair, transparent, and safe Auditing a hiring algorithm to ensure it doesn't discriminate by gender or race
AI Safety The field dedicated to ensuring AI systems remain reliable, controllable, and beneficial as they grow more capable Research into preventing AI from pursuing goals that harm people
Alignment The challenge of ensuring an AI system's goals and behaviour match what its designers and users actually intend Preventing a powerful AI from optimising for a metric in a way that causes unintended harm
Annotation The process of labelling raw data so it can be used to train supervised learning models Humans drawing bounding boxes around cars in images to train a self-driving model
API (Application Programming Interface) A defined interface that lets software systems communicate with each other Calling the OpenAI API to add GPT-powered responses to your own application
API Key A private authentication token that identifies you when making API requests Pasting your secret key into code so it has permission to use Claude or OpenAI's API
Attention Mechanism The component of a transformer that lets a model focus on the most relevant parts of the input when producing each output A model knowing that "it" in "The cat sat because it was tired" refers to the cat
Augmented Intelligence Using AI to enhance human decision-making rather than replace it entirely A radiologist using AI to flag suspicious areas in a scan, then making the final call
AutoML Automated Machine Learning — tools that automatically select models, tune hyperparameters, and build pipelines Google AutoML letting non-experts build a custom image classifier without coding

↑ Back to top


B

Term Definition Example
Backpropagation The algorithm used to train neural networks by calculating how much each parameter contributed to the error and adjusting accordingly How a neural network "learns" by working backwards from its mistakes to fix its weights
Batch Size The number of training examples processed together before the model's weights are updated A batch size of 64 means the model updates after every 64 training samples
Benchmark A standardised test used to measure and compare AI model performance MMLU (Massive Multitask Language Understanding) and HumanEval for coding ability
Bias (Data Bias) Systematic unfairness in AI outputs caused by skewed or unrepresentative training data A facial recognition system that performs poorly on darker skin tones because training data was mostly light-skinned faces
BLEU Score A metric used to evaluate the quality of AI-generated text by comparing it to human reference text Measuring how close a machine translation is to a professional human translation
Bot A software program that performs automated tasks, often simulating human interaction A customer service chatbot that answers FAQs on a website 24/7

↑ Back to top


C

Term Definition Example
Chain-of-Thought Prompting A technique that encourages an AI to reason step by step before giving a final answer, improving accuracy on complex tasks Adding "Think step by step" to a maths problem prompt dramatically improves the model's answer
Chatbot A software application that simulates conversation with users, typically powered by an LLM or rule-based system ChatGPT, customer support bots, virtual assistants on bank websites
Classification A machine learning task where a model predicts which category an input belongs to Labelling emails as spam or not spam; detecting whether a tumour is malignant or benign
Clustering Grouping similar data points together without predefined labels, used in unsupervised learning Segmenting customers into groups based on purchasing behaviour
CNN (Convolutional Neural Network) A type of neural network designed specifically for processing grid-like data such as images Used in face recognition, medical imaging, and object detection
Computer Vision The field of AI focused on enabling machines to interpret and understand visual information A self-driving car detecting pedestrians; a quality control camera spotting defects
Context Window The maximum amount of text an AI model can process and retain in a single interaction A model with a 200,000-token context window can read roughly 150,000 words at once
Copilot An AI assistant integrated into a tool or workflow to help users complete tasks more efficiently GitHub Copilot suggesting code completions as a developer types
Cross-Validation A technique for evaluating how well a model generalises by training and testing it on different subsets of the data Splitting data into 5 "folds" and rotating which one is the test set each time
CUDA A parallel computing platform by NVIDIA that enables GPUs to be used for AI training and inference Virtually every large AI model is trained using CUDA on NVIDIA hardware

↑ Back to top


D

Term Definition Example
Data Augmentation Artificially expanding a training dataset by creating modified versions of existing data Flipping, rotating, and cropping images to give a computer vision model more variety
Data Pipeline An automated workflow that collects, processes, and delivers data for AI training or inference A system that ingests raw sensor data, cleans it, and feeds it to a fraud detection model
Dataset A structured collection of data used to train or evaluate an AI model ImageNet — a dataset of 14 million labelled images used to train and benchmark vision models
Deep Learning An advanced form of machine learning that uses multi-layered neural networks to learn complex patterns Powering speech recognition, image generation, and language understanding
Deepfake AI-generated media (video, audio, or images) that realistically depicts someone saying or doing something they never did Synthetic video of a public figure making a false statement
Deployment The process of making a trained AI model available for use in a real-world product or system Releasing a trained customer churn model into a company's CRM platform
Diffusion Model A type of generative AI that learns to create data by learning to reverse a process of adding noise Stable Diffusion and DALL·E use diffusion models to generate images from text prompts
Distillation A technique where a smaller "student" model is trained to mimic the behaviour of a larger "teacher" model, reducing size and cost Creating a lightweight model for mobile devices by distilling a large cloud-based model

↑ Back to top


E

Term Definition Example
Edge AI Running AI models directly on a local device rather than sending data to the cloud A smart security camera that detects intruders locally without needing an internet connection
Embeddings Numerical vector representations of text (or other data) that capture semantic meaning and relationships Words with similar meanings have embeddings that are close together in vector space
Epoch One complete pass through the entire training dataset during model training Training for 10 epochs means the model has seen every training example 10 times
Ensemble Learning Combining multiple models and averaging their outputs to get better predictions than any single model Random Forests, which combine hundreds of decision trees to make more accurate predictions
Evaluation Metrics Measurements used to assess how well an AI model is performing Accuracy, precision, recall, F1 score, and BLEU score
Explainable AI (XAI) AI systems designed so their reasoning and decisions can be understood and audited by humans A loan-rejection system that shows which factors (income, debt ratio) drove the decision

↑ Back to top


F

Term Definition Example
Feature An individual measurable property used as input to a machine learning model In a house-price model: square footage, number of bedrooms, and location are features
Feature Engineering The process of selecting, transforming, or creating input variables to improve model performance Combining "day of week" and "time of day" into a single "rush hour" feature for a traffic model
Few-Shot Prompting Giving an AI a small number of examples in the prompt before asking it to complete a task Showing 3 example customer reviews before asking the model to classify a new one
Fine-Tuning Further training a pre-trained model on a specific, smaller dataset to specialise its behaviour Training a general LLM on legal documents to create a legal research assistant
Foundation Model A large AI model trained on broad, general data that can be adapted to many downstream tasks GPT-4, Claude, and Gemini are all foundation models
Function Calling A feature that allows an LLM to trigger external tools or APIs as part of generating a response An AI assistant calling a weather API to answer "Should I bring an umbrella tomorrow?"

↑ Back to top


G

Term Definition Example
GAN (Generative Adversarial Network) A model architecture where two networks — a generator and a discriminator — compete to produce increasingly realistic outputs Used to generate photorealistic synthetic faces or artistic images
Generative AI AI that can create new content — text, images, audio, video, or code — rather than just analysing existing data ChatGPT writing an article; Midjourney generating artwork
GPU (Graphics Processing Unit) Specialised hardware with thousands of cores that dramatically accelerates AI training and inference NVIDIA A100 and H100 GPUs are the standard for training large AI models
Gradient Descent The core optimisation algorithm that iteratively adjusts a model's weights to minimise prediction error during training The mathematical engine behind how every neural network learns
Guardrails Constraints or filters applied to an AI system to prevent it from producing harmful, offensive, or off-topic outputs A customer service bot that refuses to discuss competitors or give legal advice

↑ Back to top


H

Term Definition Example
Hallucination When an AI model confidently generates information that is factually incorrect or entirely fabricated An AI citing a scientific paper that doesn't exist, with a realistic-looking author and journal
Hugging Face A popular open-source platform for sharing, discovering, and running AI models and datasets Often called "the GitHub of AI" — thousands of models are freely available there
Human-in-the-Loop (HITL) A system design where a human reviews or approves AI decisions before they take effect A doctor reviewing an AI-flagged medical scan before acting on the recommendation
Hyperparameter A configuration value set before training begins that controls how the model learns, not what it learns Learning rate, batch size, and number of layers are all hyperparameters

↑ Back to top


I

Term Definition Example
Image Recognition AI's ability to identify and classify objects, people, or scenes within images Google Photos automatically tagging people and places in your photo library
Inference The process of using a trained AI model to generate predictions or outputs on new, unseen inputs Every time you send a message to ChatGPT, it runs inference
Interpretability The degree to which humans can understand why an AI model made a specific decision Being able to explain why a credit scoring model rejected an application

↑ Back to top


J

Term Definition Example
Jailbreak A technique used to trick an AI model into bypassing its safety rules or guidelines A roleplaying prompt designed to make an AI ignore its ethical restrictions
JSON Mode A setting in some LLM APIs that forces the model to return responses in valid JSON format Useful when building apps that need to parse AI output programmatically

↑ Back to top


K

Term Definition Example
Knowledge Base A structured repository of information that an AI can query to answer questions or complete tasks A company's internal FAQ documents connected to a RAG-powered support chatbot
Knowledge Graph A network of entities and the relationships between them, used to represent and query structured knowledge Google's Knowledge Graph connecting "Albert Einstein" to "physicist", "Germany", and "Theory of Relativity"
Knowledge Distillation Training a smaller model to replicate the performance of a larger one by learning from its outputs Creating a fast, lightweight model for edge deployment by mimicking a large cloud model

↑ Back to top


L

Term Definition Example
Label The correct answer or category assigned to a training example in supervised learning In a spam dataset, each email is labelled "spam" or "not spam"
Latency The delay between sending a request to an AI model and receiving its response A model with low latency feels instant; high latency feels slow and frustrating
Large Language Model (LLM) An AI model trained on vast amounts of text data, capable of generating, summarising, and reasoning about language GPT-4, Claude, Gemini, and Llama are all LLMs
Latent Space The compressed internal representation a model learns, where similar concepts are encoded close together In image generation models, nearby points in latent space produce visually similar images
Learning Rate A hyperparameter that controls how large a step the model takes when updating its weights during training Too high and the model overshoots; too low and it trains too slowly
LLMOps The set of practices and tools for deploying, monitoring, and maintaining LLMs in production Managing prompt versions, monitoring for drift, and evaluating model outputs at scale
LoRA (Low-Rank Adaptation) A parameter-efficient fine-tuning technique that adds small trainable layers to a model without modifying the original weights Fine-tuning a large model on a custom dataset using a fraction of the compute cost

↑ Back to top


M

Term Definition Example
Machine Learning (ML) A branch of AI where systems learn patterns from data rather than being explicitly programmed with rules A spam filter that improves over time by learning from emails users mark as spam
MCP (Model Context Protocol) An open standard created by Anthropic that allows AI models to connect to external tools, databases, and services in a consistent and secure way — think of it as a universal plug for AI integrations Connecting Claude to your GitHub repo, Google Drive, or a SQL database so it can read, write, and act on real data
Model A trained AI system that maps inputs to outputs based on what it learned from data A trained neural network that predicts tomorrow's stock price from historical data
Model Card A document published alongside an AI model describing its purpose, training data, capabilities, and limitations Hugging Face model cards provide transparency about what a model can and can't do
Model Collapse A phenomenon where AI models trained on AI-generated data degrade in quality over time A concern as the internet fills with AI-generated content used to train future models
Multimodal AI AI that can process and generate multiple types of content — text, images, audio, and video — together GPT-4o accepting an image and a question, then answering about the image in text

↑ Back to top


N

Term Definition Example
Natural Language Processing (NLP) The field of AI focused on enabling machines to understand, interpret, and generate human language Machine translation, sentiment analysis, chatbots, and text summarisation
Neural Network A computational model loosely inspired by the structure of the human brain, made up of layers of interconnected nodes The underlying architecture used by most modern AI systems
NLP Pipeline A sequence of processing steps applied to text data, from raw input to final output Tokenisation → embedding → classification → output
Node An individual computational unit in a neural network that receives inputs, applies a function, and passes an output Billions of nodes work together in a large neural network

↑ Back to top


O

Term Definition Example
Object Detection A computer vision task that identifies what objects are in an image and where they are located A self-driving car identifying pedestrians, traffic lights, and other vehicles in real time
Ontology A formal representation of concepts and the relationships between them within a specific domain A medical ontology defining how "disease", "symptom", and "treatment" relate to each other
Open Source Model An AI model whose weights and/or code are publicly available for anyone to use, modify, and distribute Meta's Llama models, Mistral, and Stable Diffusion
Overfitting When a model learns the training data too precisely — including its noise — and fails to generalise to new data A model that scores 99% on training data but only 60% on real-world data

↑ Back to top


P

Term Definition Example
Parameter An internal numerical value a model learns during training that shapes how it processes and generates outputs GPT-4 is estimated to have over a trillion parameters
Pre-training The initial large-scale training phase where a model learns from a massive general dataset before specialisation Training an LLM on hundreds of billions of words from the internet and books
Precision The percentage of positive predictions that were actually correct Of all emails the model flagged as spam, what percentage were truly spam?
Prompt The instruction, question, or input you give to an AI model to guide its response "Summarise this article in three bullet points for a non-technical audience"
Prompt Engineering The practice of designing and refining prompts to get better, more reliable outputs from AI models Using structured formatting, role assignment, and examples to improve response quality
Prompt Injection An attack where malicious instructions hidden in content the AI reads attempt to hijack its behaviour A webpage containing invisible text instructing a browsing AI to leak your personal data

↑ Back to top


Q

Term Definition Example
Quantisation A technique that reduces a model's memory usage by representing its weights with lower numerical precision, making it faster and cheaper to run Running a compressed Llama model on a laptop instead of a high-end server
Query The input or question sent to an AI model or database to retrieve information "What are the side effects of ibuprofen?" sent to a medical AI system

↑ Back to top


R

Term Definition Example
RAG (Retrieval-Augmented Generation) A technique that combines real-time document retrieval with AI generation, reducing hallucination and keeping responses current A chatbot that searches your company's knowledge base before answering a support question
Recall The percentage of actual positives that the model successfully identified Of all actual fraud cases, what percentage did the model correctly flag?
Recommendation System An AI system that predicts and surfaces content or products a user is likely to want, based on past behaviour Netflix's "Because you watched" suggestions; Spotify's Discover Weekly playlist
Red Teaming Deliberately attempting to break or manipulate an AI system to discover safety vulnerabilities before release Researchers probing a model with adversarial prompts to expose harmful outputs
Regression A machine learning task where the model predicts a continuous numerical value Predicting a house's sale price based on size, location, and age
Reinforcement Learning (RL) Training a model through a system of rewards and penalties, so it learns to maximise cumulative reward AlphaGo learning to play Go by playing millions of games and receiving rewards for winning
RLHF (Reinforcement Learning from Human Feedback) A training technique where humans rate AI outputs, and the model learns to produce outputs humans prefer The technique used to align ChatGPT and Claude to be helpful, harmless, and honest
RNN (Recurrent Neural Network) A neural network designed for sequential data, where outputs feed back as inputs — largely replaced by transformers Used in early speech recognition and text generation before transformers dominated

↑ Back to top


S

Term Definition Example
Semantic Search Search that understands the meaning and intent behind a query rather than matching exact keywords Searching "how to fix a broken bone" and getting results about fracture treatment, not carpentry
Sentiment Analysis AI that determines the emotional tone — positive, negative, or neutral — of a piece of text Automatically classifying thousands of customer reviews to measure product satisfaction
Speech Recognition AI that converts spoken audio into written text Apple's Siri, Google Voice, and OpenAI's Whisper model
Supervised Learning A training approach where the model learns from labelled input-output pairs Training a model on thousands of (email, spam/not spam) pairs so it can classify new emails
Synthetic Data Artificially generated data used to train or test models when real data is scarce, costly, or sensitive Generating fake patient records to train a healthcare AI without privacy concerns
System Prompt A hidden set of instructions given to an AI before the user conversation begins, used to shape its behaviour and persona A company using a system prompt to make Claude respond only about their products

↑ Back to top


T

Term Definition Example
Temperature A setting that controls how predictable or creative an AI's outputs are — low is focused and deterministic, high is varied and creative Set temperature low for factual Q&A; set it high for creative brainstorming
Text-to-Image AI that generates images from a natural language description DALL·E, Midjourney, and Stable Diffusion generating artwork from a text prompt
Text-to-Speech (TTS) AI that converts written text into natural-sounding spoken audio ElevenLabs generating a realistic voice clone from a few seconds of audio
Token The basic unit of text an LLM processes — roughly a word or part of a word "Artificial" might be split into "Art", "ific", "ial" — three tokens
Top-p Sampling A setting that controls output variety by limiting the pool of next-word candidates to a cumulative probability threshold Often tuned alongside temperature to balance quality and creativity
TPU (Tensor Processing Unit) Hardware designed specifically to accelerate AI workloads, developed by Google Used to train Google's Gemini and other large models
Training The process of exposing a model to data and adjusting its weights to minimise prediction error Training GPT-4 required thousands of GPUs running for months
Transfer Learning Reusing a model trained on one task as the starting point for a new but related task Adapting a model trained on English text to work with French by fine-tuning on French data
Transformer An attention-based neural network architecture that is the backbone of virtually all modern LLMs GPT, Claude, Gemini, and Llama are all transformer-based models
TTS — see Text-to-Speech

↑ Back to top


U

Term Definition Example
Underfitting When a model is too simple to capture the underlying patterns in the data, resulting in poor performance A linear model trying to predict stock prices — too simple for the complexity of the problem
Unsupervised Learning Training a model on unlabelled data so it discovers its own patterns and structure Grouping news articles into topic clusters without being told what the topics are

↑ Back to top


V

Term Definition Example
Validation Set A portion of data held back from training, used to tune the model and catch overfitting before final evaluation Monitoring validation loss during training to decide when to stop
Vector A list of numbers that represents data (like a word or image) in a mathematical space The word "king" might be represented as a vector of 768 numbers in an embedding model
Vector Database A database that stores and indexes embeddings (vectors) so AI can retrieve semantically relevant information quickly Pinecone, Weaviate, and Chroma are popular vector databases used in RAG systems

↑ Back to top


W

Term Definition Example
Weight A numerical parameter inside a neural network that is adjusted during training to reduce error A model with 70 billion parameters has 70 billion weights stored in memory
Weight Decay A regularisation technique that penalises large weights during training to prevent overfitting Commonly used alongside dropout to keep models from memorising training data

↑ Back to top


X

Term Definition Example
XAI (Explainable AI) AI systems and techniques designed to make model decisions interpretable and understandable to humans A credit scoring model that explains: "Rejected due to high debt-to-income ratio and short credit history"

↑ Back to top


Y

Term Definition Example
YAML A human-readable data format commonly used to write configuration files for AI tools and ML pipelines Writing a training configuration file for a machine learning experiment
YOLO (You Only Look Once) A real-time object detection algorithm known for its speed and efficiency Detecting and tracking multiple objects in a live video feed at 60 frames per second

↑ Back to top


Z

Term Definition Example
Zero-Shot Learning A model's ability to perform a task it was never explicitly trained on, by generalising from related knowledge Asking GPT-4 to translate a language it saw rarely during training with no translation-specific training
Zero-Shot Prompting Giving an AI a task with no examples — relying entirely on its pre-trained knowledge "Classify this review as positive or negative: 'The food was amazing!'" — no examples given

This glossary covers 100+ terms across the full AI landscape. Bookmark it, share it, and revisit it as you grow.

Top comments (0)