Skip to Main Menu
Skip to Main Content
Skip to Footer

AI21 Glossary

Learn what foundational large language models are, how they function, and their role in powering advanced AI systems.

Agent swarm refers to a multi-agent system in which multiple AI agents collaborate toward a shared objective. Each agent operates…

Agentic RAG is a method that uses AI agents to control how a language model retrieves information and generates responses.…

AI agility refers to the speed at which artificial intelligence systems can adapt to changing conditions. In enterprise settings, this…

AI Infrastructure

AI infrastructure refers to the combination of hardware and software systems designed to support workloads powered by artificial intelligence (AI).…

AI Integration Roadmap for Enterprises in 2026

By 2026, AI integration has moved from “interesting pilot” to a core part of enterprise architecture. The organizations that pull…

AI models are software systems designed and trained to perform specific tasks. This includes analyzing data to identify trends and…

Artificial General Intelligence (AGI)

Artificial general intelligence (AGI) is a theoretical form of artificial intelligence (AI) that can learn, understand, and perform any intellectual…

Automatic Prompt Engineer

Automatic prompt engineer (APE) is a machine learning–based system that automatically creates and refines text prompts — written instructions used…

Black-Box Models

A black-box model is a type of artificial intelligence (AI) system that generates outputs without exposing the specific steps or…

Chain-of-thought prompting

Chain-of-thought prompting (Cot) is a method used with large language models that encourages them to generate intermediate reasoning steps before…

Closed-Source LLM

A closed-source LLM is a large language model whose source code is not publicly available. It is owned, developed, and…

Continual Learning

Continual learning is the ability of an AI model to retain knowledge over time by continuously incorporating new data while…

Continuous Batching

Continuous Batching What is Continuous Batching? Continuous batching is a scheduling technique for large language model inference that allows new…

Convolutional Neural Network

A convolutional neural network (CNN) is a specialized type of neural network used in deep learning to process visual data.…

Data Augmentation

Data augmentation is the process of creating additional training data by making controlled changes to existing data, while preserving its…

Data poisoning is an attack in artificial intelligence where malicious, false, or manipulated data is deliberately inserted into training data,…

Deep learning is a type of machine learning (ML) used in many artificial intelligence (AI) applications. It is particularly effective…

Directional Stimulus Prompting

Directional stimulus prompting is a technique used to influence how a large language model (LLM) interprets and responds to a…

Document Parsing (PDF Parsing)

Document Parsing (PDF Parsing) What it is Document parsing is the process of extracting structured information – including text, tables,…

Dynamic Prompting

Dynamic prompting is an approach in artificial intelligence (AI) where the input text given to an AI model — known…

Enterprise AI Solutions: From Experiments to Real Business Impact

Enterprise AI solutions are no longer a side project for innovation teams, they are becoming the backbone of how large…

Enterprise Search

Enterprise search is software that lets employees find information across a company’s internal systems through a single search bar. It…

Feature Engineering

Feature engineering is a key step in machine learning (ML) that involves transforming raw data into meaningful input features for…

Federated Learning

Federated learning is a method of training AI models across multiple separate devices or servers, each of which retains its…

Federated Search

Federated search is a method of retrieving information from many different sources using a single search query. It allows users…

Few-Shot Learning

Few-shot learning is a method in artificial intelligence (AI) where a system performs new tasks or recognizes new concepts after…

Few-Shot Prompting

Few-shot prompting is a technique in artificial intelligence (AI) in which a language model is provided with a small number…

Fine-tuning is the process of adapting a pre-trained language model to perform a specific task with greater accuracy on domain-specific…

Function Calling

Function calling is a process where software directs a specific function or external service to execute a task during runtime.…

Generative Adversarial Networks (GANs)

A generative adversarial network is a machine learning model that creates synthetic data by training two neural networks in competition.…

Generative AI (GenAI)

Generative AI (GenAI) is a type of artificial intelligence designed to create new content, such as text, images, videos, music,…

GRPO (Group Relative Policy Optimization)

GRPO (Group Relative Policy Optimization) GRPO, or Group Relative Policy Optimization, is a reinforcement learning algorithm used to fine-tune large…

Homomorphic Encryption

Homomorphic encryption is a cryptographic technique that allows data to remain encrypted while it is being processed. It makes it…

Human-in-the-Loop (HITL)

Human-in-the-loop (HITL) is a process in artificial intelligence (AI) that involves human participation in producing, evaluating, or refining the results…

A hybrid cloud is a computing model that connects on-premises infrastructure with private and public cloud services, allowing data and…

Hybrid RAG is a retrieval method that combines structured and unstructured data sources to support more comprehensive information access in…

Hyperparameter Tuning

Hyperparameter tuning is used in machine learning (ML) to select optimal hyperparameters – parameters set before training that govern model…

In-context Learning

In-context learning is the ability of certain artificial intelligence (AI) models to adjust their outputs based on information included directly…

Instruction Tuning

Instruction tuning is a method in artificial intelligence (AI) for aligning a language model’s responses with specific written instructions, shaping…

Knowledge Distillation

Knowledge distillation is a training method where a smaller machine learning model learns to replicate the behavior of a larger,…

KV Cache Glossary

KV Cache What is a KV Cache? A KV cache (key-value cache) is a memory structure used in large language…

Labeled data is data that has been annotated with tags or identifiers to add meaning and context. It is essential…

Language Models (LLMs)

Large language models (LLMs) are deep learning models designed to process and generate human-like language. They are capable of text…

Large Reasoning Models (LRMs)

A Large Reasoning Model (LRM) is an artificial intelligence system that combines natural language understanding with logical reasoning to solve…

Least-to-Most Prompting

Least-to-most (LtM) prompting is a strategy in artificial intelligence (AI) in which a model begins with minimal input and receives…

Lemmatization is a technique that replaces different forms of a word with a single base form, known as a lemma.…

LLM agents are AI systems that use large language models (LLMs) to carry out tasks based on language input. They…

LLM Rate Limiting

LLM rate limiting is the practice of controlling how often a large language model (LLM) can be accessed within a…

LLM Red Teaming

LLM red teaming is a structured process used to uncover risks, weaknesses, or unintended behavior in large language models (LLMs)…

LLM Temperature

Temperature is a setting that controls how predictable or varied a language model’s output is — from consistent, fact-based responses…

LLM-as-a-Judge (LaaJ) is an evaluation method in which a large language model (LLM) is used to assess the quality of…

LLMOps is the practice of deploying, monitoring and updating large language models (LLMs) across their full development cycle. These models…

Machine Learning

Machine learning (ML) is a subset of artificial intelligence (AI) that enables computers to perform tasks by learning from data,…

Meta prompting is the practice of designing prompts that define the structure, tone, or logic of other prompts used by…

Multi-Party Computation

Multi-party computation is a cryptographic technique used in private AI that allows several parties to compute a shared result without…

Multimodal Large Language Model

Multimodal large language models (MLLMs) are deep learning algorithms that can process and analyze data across multiple content types, including…

Natural Language Processing (NLP)

Natural Language Processing (NLP) is a type of artificial intelligence that enables computers to interpret human language. It assists machines…

Natural Language Understanding (NLU)

Natural Language Understanding (NLU) is a branch of artificial intelligence that enables machines to interpret the meaning behind human language,…

Neural Networks

Neural networks are computer systems modeled after the human brain that can recognize patterns, make predictions, and improve over time…

Online Reinforcement Learning (Online-RL) Training

Online Reinforcement Learning (Online-RL) Training Online reinforcement learning (online-RL) training is a method for training large language models in which…

Open-Source LLM

An open-source LLM is a large language model whose code and design are freely available for anyone to access, use,…

Open-Weights Model

Open-weights models are large language models (LLMs) whose trained parameters — known as weights — are publicly available. These weights…

Orchestration Layer

An orchestration layer is a software component that manages how different systems coordinate with one another. It can automate workflows…

Overfitting occurs when a machine learning (ML) model performs well on training data but fails to generalize to new, unseen…

Prefill and Decode

Prefill and Decode Prefill and decode are the two distinct computational phases of LLM inference. Prefill is the phase in…

Privacy-Enhancing Technologies (PETs)

Privacy-enhancing technologies (PETs) are tools and methods that protect personal data throughout its lifecycle, and are pivotal for the deployment…

A private cloud is a cloud computing setup where one organization has exclusive use of the infrastructure. Teams configure systems…

Prompt Injection

Prompt injection is a technique used to manipulate how an AI model interprets instructions by inserting unexpected input into a…

Prompt Template

A prompt template is a predefined structure used to organize input provided to an AI model. It ensures that information…

Prompt tuning is a technique in artificial intelligence (AI) that customizes how a language model responds by optimizing the prompts…

A public cloud is a model in which computing resources are delivered over the internet by external providers and shared…

RAG as a Service (RaaS)

RAG as a Service (RaaS) is a managed service where providers handle the entire RAG process, including data ingestion, indexing,…

RAG for Unstructured Data

RAG for unstructured data refers to a retrieval-augmented generation method that combines information retrieval with a large language model (LLM)…

Retrieval Latency

Retrieval latency refers to the time it takes for a system to retrieve relevant information after a request is made.…

Retrieval-Augmented Generation (RAG)

Retrieval-augmented generation (RAG) is a technique that improves the accuracy of AI-generated responses by combining a language model – a…

Self-Consistency (LLM Evaluation)

Self-Consistency (LLM Evaluation) What it is Self-consistency is a quantitative metric used to evaluate how reliably a language model produces…

Self-Supervised Learning

Self-supervised learning is a machine learning technique where models learn from unlabeled data by predicting missing or masked elements, such…

Semantic Search

Semantic search is a method of finding information based on the intended meaning of words, rather than matching exact terms.…

Semi-Structured Data

Semi-structured data is information that is not organized using traditional structures such as tables or relational databases. It does not…

Sequence Packing

Sequence Packing Sequence packing is a training optimization technique in which multiple variable-length input sequences are concatenated into a single…

State-Space Model (SSM)

State-Space Model (SSM) A state-space model (SSM) is a class of neural network architecture that processes sequences by maintaining a…

Stemming is a natural language processing method that reduces words to a simpler form, known as a stem. It allows…

Structured Data

Structured data is information that follows a predefined format or schema. It is highly organized and easy to understand by…

Supervised Learning

Supervised learning is a type of machine learning that uses labeled data to train algorithms. It is commonly applied in…

Synthetic data is artificially generated information that reproduces the patterns, structure, and statistical properties of real-world data — such as…

Task Decomposition

Task decomposition is a process in artificial intelligence (AI) that involves dividing a complex objective into smaller, logically defined tasks.…

Training data is the information used to develop artificial intelligence (AI) models and machine learning algorithms. It enables the model…

Transfer Learning

Transfer learning is a machine learning technique where knowledge gained from one task is used to improve performance on a…

Tree of Thoughts Prompting

Tree of thoughts (ToT) prompting is a technique in artificial intelligence (AI) in which a language model generates and evaluates…

Trusted Execution Environments (TEE)

A trusted execution environment (TEE) is a secure area inside a computer processor that runs sensitive code and handles private…

Underfitting is a common challenge in developing a machine learning (ML) model. It occurs when the model is too simplistic…

Unlabeled data is raw data used in machine learning (ML) that has no annotations, tags, categories, or labels. It is…

Unstructured Data

Unstructured data is information that does not follow a predefined format. It can be found in various formats such as…

Unstructured RAG

Unstructured RAG is an approach that retrieves and uses information from unstructured data to support AI outputs. It refers to…

Unsupervised Learning

Unsupervised learning is a type of machine learning in which algorithms explore and learn from data without labeled input, similar…

Vector Database

A vector database is a specialized type of database that stores data as numerical vectors — structured lists of numbers…

A vector index is a data structure that organizes and stores numerical representations of data — called embeddings, or vectors…

Verifiable Rewards

Verifiable Rewards Verifiable rewards are a class of reward signals used in reinforcement learning for LLMs where the correctness of…

Vertical LLMs are large language models (LLMs) tailored for specific industries, functions, or domains. Unlike general-purpose models trained on broad…

Vision-Language Model (VLM)

What it is A vision-language model (VLM) is a type of multimodal AI system that can process both visual inputs…

Workflow Automation

Workflow automation is the process of using software to perform business processes and tasks without human involvement. It can complete…

Zero-Shot Learning

Zero-shot learning is a machine learning (ML) approach that allows a model to recognize, classify, or understand new types of…

Zero-Shot Prompting

Zero-shot prompting is a method in artificial intelligence in which a model is given a prompt to complete a task…