AI Glossary
AI moves fast. Here's a glossary to help you keep up. Plain-English definitions for every major AI term.
161 terms · 24 letters covered · Always free
Total Terms
161
and growing weekly
Letters Covered
24
of the alphabet
Categories
8
topic areas
Accuracy
Accuracy is the proportion of correct predictions out of total predictions made by a classification model, cal…
Activation Function
Activation Function is a mathematical function applied to the output of a neural network node that introduces…
Adversarial Attack
Adversarial Attack is a technique where carefully crafted inputs are designed to deceive AI models into making…
Agentic AI
Agentic AI is aI systems that can autonomously plan, reason, and take actions to achieve goals without constan…
AI Agent
AI Agent is an autonomous software system that perceives its environment, makes decisions, and takes actions t…
AI Alignment
AI Alignment is the challenge of ensuring AI systems pursue goals that are consistent with human values and in…
AI Ethics
AI Ethics is an interdisciplinary field examining the moral implications of artificial intelligence, addressin…
AI Governance
AI Governance is the frameworks, regulations, and organizational structures that guide the development and dep…
AI Safety
AI Safety is the interdisciplinary field focused on ensuring AI systems operate reliably, ethically, and witho…
AI-as-a-Service
AI-as-a-Service (AIaaS) delivers artificial intelligence capabilities through cloud-based APIs and platforms,…
Alignment Tax
Alignment Tax is the performance cost incurred when making AI models safer and more aligned with human values.…
Annotation
Annotation is the process of labeling data with metadata that AI models can learn from during supervised train…
API
API application Programming Interface — in the AI context, a standardized way for developers to send requests…
Artificial Intelligence
Artificial Intelligence (AI) is a field of computer science focused on building systems capable of performing…
Attention Mechanism
Attention Mechanism is a technique that allows neural networks to focus on the most relevant parts of the inpu…
AUC-ROC
AUC-ROC (Area Under the Receiver Operating Characteristic Curve) is a classification performance metric that m…
Autoregressive Model
Autoregressive Model is a type of generative model that produces output one element at a time, with each new e…
Backpropagation
Backpropagation is the algorithm that computes gradients of the loss function with respect to each weight in a…
Batch Size
Batch Size is the number of training examples processed simultaneously in one forward and backward pass of a n…
Benchmark
Benchmark is a standardized test or dataset used to evaluate and compare the performance of AI models. Common…
Bias
Bias in AI refers to systematic errors in model predictions that arise from skewed training data, flawed assum…
BLEU Score
BLEU Score is bilingual Evaluation Understudy score, a metric for evaluating the quality of machine-translated…
Catastrophic Forgetting
Catastrophic Forgetting is a phenomenon where neural networks lose previously learned knowledge when trained o…
Chain of Thought
Chain of Thought is a prompting technique where AI models show their reasoning step-by-step before arriving at…
Chatbot
A Chatbot is an AI-powered software application that simulates human conversation through text or voice interf…
Chinchilla Scaling
Chinchilla Scaling is a training methodology derived from DeepMind's Chinchilla paper showing that many large…
CLIP
CLIP is contrastive Language-Image Pre-training, a model developed by OpenAI that learns visual concepts from…
Cloud AI
Cloud AI refers to artificial intelligence services delivered through cloud computing platforms — including pr…
CNN (Convolutional Neural Network)
CNN (Convolutional Neural Network) is a deep learning architecture designed for processing grid-structured dat…
Code Generation
Code Generation is an AI capability where language models produce functional source code from natural language…
Compute
Compute is the computational resources (GPUs, TPUs) required to train and run AI models. Compute costs are the…
Computer Vision
Computer Vision is an AI discipline that trains machines to interpret and understand visual information from i…
Constitutional AI
Constitutional AI is an AI alignment technique developed by Anthropic where AI systems are trained to follow a…
Context Window
Context Window is the maximum amount of text an AI model can process in a single interaction, measured in toke…
Contrastive Learning
Contrastive Learning is a self-supervised learning technique where models learn by comparing similar and dissi…
CUDA
CUDA (Compute Unified Device Architecture) is NVIDIA proprietary parallel computing platform and programming m…
Data Augmentation
Data Augmentation is a technique that artificially expands training datasets by applying transformations to ex…
Data Flywheel
Data Flywheel is a self-reinforcing cycle where an AI product generates more user data, which improves the mod…
Data Labeling
Data Labeling is the process of annotating raw data with meaningful tags or categories that enable supervised…
Data Poisoning
Data Poisoning is an attack where malicious data is injected into a training dataset to compromise an AI model…
Dataset
Dataset is a structured collection of data used for training, validating, and testing machine learning models,…
Decoder
Decoder is a neural network component that generates output sequences from encoded representations. In transfo…
Deep Learning
Deep Learning is a machine learning technique that uses multi-layered neural networks (deep neural networks) t…
Depthwise Separable Convolution
Depthwise Separable Convolution is an efficient neural network operation that factorizes a standard convolutio…
Diffusion Model
Diffusion Model is a generative AI architecture that learns to create data by reversing a gradual noise-additi…
Diffusion Models
Diffusion Models is a class of generative AI models that create images by gradually denoising random noise. St…
Direct Preference Optimization (DPO)
Direct Preference Optimization (DPO) is a simplified alternative to RLHF for aligning language models with hum…
Distillation
Distillation is a model compression technique that transfers knowledge from a large teacher model to a smaller…
Distributed Training
Distributed Training is the practice of training AI models across multiple GPUs or machines simultaneously to…
Edge AI
Edge AI is aI processing performed locally on devices (phones, IoT sensors, cars) rather than in the cloud. Ed…
Embedding
Embedding is a dense numerical representation of data (text, images, audio) in a continuous vector space where…
Emergent Abilities
Emergent Abilities is capabilities that appear in large language models only at sufficient scale, such as arit…
Encoder
Encoder is a neural network component that processes input data and produces a compressed representation captu…
Encoder-Decoder
Encoder-Decoder is a neural network architecture where an encoder compresses input into a dense representation…
Epoch
Epoch is one complete pass through the entire training dataset during model training. Training typically invol…
Explainability
Explainability is the degree to which an AI model's decision-making process can be understood by humans. Expla…
F1 Score
F1 Score is the harmonic mean of precision and recall, providing a single metric that balances both false posi…
Feature Store
Feature Store is a centralized repository for storing, managing, and serving machine learning features across…
Federated Learning
Federated Learning is a machine learning approach where models are trained across multiple decentralized devic…
Few-Shot Learning
Few-Shot Learning is the ability of AI models to learn new tasks from just a handful of examples, rather than…
Fine-Tuning
Fine-Tuning is the process of further training a pre-trained model on a smaller, task-specific dataset to adap…
Flash Attention
Flash Attention is an optimized attention algorithm that dramatically reduces the memory requirements and spee…
Foundation Model
Foundation Model is a large AI model trained on broad data that can be adapted to many downstream tasks throug…
Frontier Models
Frontier Models is the most capable and advanced AI models available at any given time. As of 2026, frontier m…
GAN (Generative Adversarial Network)
GAN (Generative Adversarial Network) is a neural network architecture consisting of two networks — a generator…
Generative AI
Generative AI refers to artificial intelligence systems that create new content — text, images, video, audio,…
GPT (Generative Pre-trained Transformer)
GPT (Generative Pre-trained Transformer) is OpenAI family of autoregressive language models that predict the n…
GPU
GPU is graphics Processing Unit — a specialized processor originally designed for rendering graphics but now e…
GPU Cloud
GPU Cloud is cloud computing services that provide on-demand access to GPU hardware for AI training and infere…
Gradient Descent
Gradient Descent is the fundamental optimization algorithm used to train neural networks, iteratively adjustin…
Grounding
Grounding is the process of connecting AI model outputs to verified external data sources to reduce hallucinat…
Guardrails
Guardrails is safety mechanisms built into AI systems to prevent harmful, biased, or inappropriate outputs. Gu…
Hallucination
Hallucination is when AI models generate information that is factually incorrect or fabricated but presented w…
HumanEval
HumanEval is a code generation benchmark created by OpenAI containing 164 hand-written programming problems wi…
Hyperparameter
Hyperparameter is a configuration setting for model training that is set before the learning process begins, a…
In-Context Learning
In-Context Learning is the ability of large language models to adapt their behavior based on examples provided…
Inference
Inference is the process of running a trained AI model to generate predictions or outputs. Inference costs oft…
Inference Cost
Inference Cost is the computational expense of running a trained AI model to generate predictions, measured in…
Inference Endpoint
Inference Endpoint is a deployed API server that hosts a trained AI model and accepts requests to generate pre…
Large Language Model
Large Language Model (LLM) is a neural network with billions or trillions of parameters trained on massive tex…
Latency
Latency in AI systems measures the time delay between sending a request and receiving a response, typically re…
Latent Space
Latent Space is the abstract, lower-dimensional representation space learned by neural networks to encode the…
Learning Rate
Learning Rate is a hyperparameter that controls how much a model adjusts its weights in response to each batch…
LLM (Large Language Model)
LLM (Large Language Model) is an AI model trained on vast amounts of text data, capable of understanding and g…
LoRA (Low-Rank Adaptation)
LoRA (Low-Rank Adaptation) is a parameter-efficient fine-tuning technique that adds small trainable matrices t…
Loss Function
Loss Function is a mathematical function that quantifies the difference between a model predictions and actual…
LSTM
LSTM (Long Short-Term Memory) is a specialized recurrent neural network architecture that uses gating mechanis…
Machine Learning
Machine Learning is a subset of artificial intelligence where algorithms learn patterns from data to make pred…
Machine Translation
Machine Translation is an NLP application that automatically translates text or speech from one natural langua…
Mixture of Agents
Mixture of Agents is an AI system architecture where multiple specialized AI agents collaborate to solve compl…
Mixture of Experts
Mixture of Experts is an architecture where multiple specialized sub-networks (experts) are combined, with a r…
MLOps
MLOps is the set of practices for deploying, monitoring, and maintaining machine learning models in production…
MMLU
MMLU (Massive Multitask Language Understanding) is a benchmark comprising 15,908 multiple-choice questions acr…
Model Card
Model Card is a standardized document that accompanies an AI model describing its intended use cases, training…
Model Collapse
Model Collapse is a phenomenon where AI models trained on data generated by other AI models progressively degr…
Model Serving
Model Serving is the infrastructure and process of deploying trained AI models to production environments wher…
Multi-Head Attention
Multi-Head Attention is a transformer architecture mechanism that runs multiple attention computations in para…
Multimodal AI
Multimodal AI is aI models that can process and generate multiple types of data — text, images, audio, video —…
Named Entity Recognition
Named Entity Recognition (NER) is an NLP task that identifies and classifies named entities in text — such as…
Natural Language Processing
Natural Language Processing (NLP) is a branch of artificial intelligence that enables computers to understand,…
Neural Architecture Search
Neural Architecture Search is an automated process where AI is used to design optimal neural network architect…
Neural Network
Neural Network is a computing system inspired by biological brain structure, composed of interconnected nodes…
Perplexity
Perplexity is a metric that evaluates language model quality by measuring how well the model predicts a sample…
Positional Encoding
Positional Encoding adds information about token position in a sequence to transformer models, which lack inhe…
Pre-training
Pre-training is the initial phase of training a foundation model on massive amounts of unlabeled data to learn…
Precision
Precision is a classification metric that measures the proportion of positive predictions that are actually co…
Prompt Engineering
Prompt Engineering is the art and science of crafting effective inputs to AI models to elicit desired outputs.…
Prompt Injection
Prompt Injection is a security vulnerability where malicious instructions embedded in user input or external d…
RAG (Retrieval-Augmented Generation)
RAG (Retrieval-Augmented Generation) is a technique that enhances LLM responses by retrieving relevant documen…
Reasoning Models
Reasoning Models is aI models specifically designed for complex logical and mathematical reasoning, such as Op…
Recall
Recall is a classification metric measuring the proportion of actual positive cases that the model correctly i…
Red Teaming
Red Teaming is the practice of deliberately probing AI systems for vulnerabilities, biases, and failure modes…
Regularization
Regularization encompasses techniques that prevent neural networks from overfitting training data by adding co…
Reinforcement Learning
Reinforcement Learning is a machine learning paradigm where an AI agent learns optimal behavior through trial…
Reinforcement Learning from Human Feedback (RLHF)
Reinforcement Learning from Human Feedback (RLHF) is a training technique where AI models are fine-tuned using…
Responsible AI
Responsible AI is the practice of developing and deploying AI systems that are fair, transparent, accountable,…
Retrieval
Retrieval is the process of searching and fetching relevant information from external knowledge sources to aug…
RNN (Recurrent Neural Network)
RNN (Recurrent Neural Network) is a neural network architecture designed for sequential data, where the output…
SaaS AI
SaaS AI is software-as-a-Service products that embed AI capabilities as core features. SaaS AI includes tools…
Scaling Laws
Scaling Laws is empirical observations that AI model performance improves predictably as compute, data, and pa…
Self-Supervised Learning
Self-Supervised Learning is a training paradigm where models learn from unlabeled data by predicting missing p…
Sentiment Analysis
Sentiment Analysis is an NLP technique that identifies and classifies the emotional tone expressed in text — p…
Sovereign AI
Sovereign AI is the concept that nations should develop their own AI capabilities, models, and infrastructure…
Speculative Decoding
Speculative Decoding is an inference optimization technique where a smaller, faster draft model generates cand…
Speech-to-Text
Speech-to-Text (STT), also called automatic speech recognition (ASR), is an AI technology that converts spoken…
State Space Model
State Space Model is an alternative to transformer architecture that processes sequences using principles from…
Summarization
Summarization is an NLP task where AI models condense long documents into shorter versions while preserving ke…
Superalignment
Superalignment is openAI's research initiative focused on ensuring superintelligent AI systems remain aligned…
Supervised Learning
Supervised Learning is a machine learning paradigm where models are trained on labeled datasets — input-output…
Synthetic Data
Synthetic Data is artificially generated data used to train AI models when real-world data is scarce, expensiv…
System Prompt
System Prompt is hidden instructions given to an AI model that define its behavior, personality, and constrain…
Temperature
Temperature is a parameter that controls the randomness of AI model outputs. Low temperature (near 0) produces…
Test-Time Compute
Test-Time Compute is additional computational resources allocated during inference to improve model performanc…
Text-to-Image
Text-to-Image generation is an AI capability that creates visual images from natural language descriptions, po…
Text-to-Speech
Text-to-Speech (TTS) is an AI technology that converts written text into natural-sounding spoken audio, using…
Text-to-Video
Text-to-Video generation is an AI capability that creates video content from natural language descriptions, wi…
Throughput
Throughput in AI systems measures the rate of data processing, typically reported as tokens per second for lan…
Token
Token is the basic unit of text processed by language models. A token is roughly 3/4 of a word in English. Mod…
Tokenizer
Tokenizer is a component that converts raw text into a sequence of tokens that a language model can process. T…
Tool Use
Tool Use is the ability of an AI model to interact with external tools and APIs — such as web search, code int…
Tool Use (Function Calling)
Tool Use (Function Calling) is the ability of AI models to invoke external tools, APIs, or databases during co…
TPU
TPU (Tensor Processing Unit) is a custom-designed AI accelerator chip developed by Google specifically for mac…
Training Data
Training Data is the dataset used to teach machine learning models patterns and relationships, comprising inpu…
Transfer Learning
Transfer Learning is the practice of applying knowledge learned from one task or domain to improve performance…
Transformer
Transformer is a neural network architecture introduced in 2017 that uses self-attention mechanisms to process…
Underfitting
Underfitting occurs when a machine learning model is too simple to capture the underlying patterns in training…
Unicorn
Unicorn is a privately held startup valued at over $1 billion. The AI sector has produced more unicorns than a…
Unsupervised Learning
Unsupervised Learning is a machine learning approach where models discover hidden patterns, groupings, or stru…
VAE (Variational Autoencoder)
VAE (Variational Autoencoder) is a generative model that learns a compressed latent representation of input da…
Vector Database
Vector Database is a specialized database optimized for storing and querying high-dimensional vector embedding…
Vision-Language Model
Vision-Language Model is aI models that can understand and reason about both images and text simultaneously. V…
Watermarking
Watermarking is a technique for embedding invisible statistical patterns in AI-generated content to enable det…
Weight
Weight is a numerical parameter in a neural network that determines the strength of the connection between neu…
World Model
World Model is an AI system's internal representation of how the world works, enabling it to predict future st…
Common Questions
FAQ
What is the Awaira AI Glossary?
Who is this glossary for?
How often is the glossary updated?
Can I suggest a term?
Know the terms. Now explore the companies.
Awaira tracks 400+ AI companies -- valuations, funding, founders, and more. All free.
Browse Companies