Resource

AI Glossary

The terms that keep coming up in meetings, articles, and proposals — explained clearly. No unnecessary jargon.

69 terms · updated June 2026

A

Adversarial Attacks

Techniques used to fool AI models using manipulated data, causing them to produce incorrect results.

AGI

Artificial General Intelligence — A type of artificial intelligence capable of performing any intellectual task that a human can do. Does not yet exist in practice.

AI Agents

Systems that can understand information, reason, and execute actions autonomously.

AI Orchestrator

Software or framework designed to coordinate multiple AI tools, models, or agents to execute complex, automated workflows.

Algorithmic bias

Systemic errors in an AI model that generate unfair or prejudiced outcomes due to flawed training data or design.

Artificial Intelligence (AI)

Field of technology focused on creating machines capable of performing tasks that normally require human intelligence, such as learning, reasoning, or making decisions.

Artificial Neural Network

Computational model inspired by the biological structure of the human brain, composed of interconnected nodes organized in layers.

C

Chain-of-thought Prompting

Technique that guides a model to solve problems step by step, showing its intermediate reasoning.

Computer vision

Field of AI that enables computers to extract meaningful information and understand digital images, videos, or other visual inputs.

Context window

The maximum amount of information (measured in tokens) that a model can retain and process within a single interaction or prompt.

Conversational Agents

AI designed to interact with people through natural language, understanding intent and context.

D

Data Accessibility

Availability of data in the right format and location to be used by AI models.

Data Ingestion and Preparation

Process of collecting, cleaning, and transforming raw data into a format suitable for use by AI models.

Data Quality

How correct, complete, consistent, and useful the data is.

Deep Learning

Type of machine learning that uses neural networks with many layers to detect complex patterns.

E

Embeddings

Numerical representations in the form of vectors that capture the semantic meaning and context of words, phrases, or data.

Ethical AI

Use of artificial intelligence in a responsible way, avoiding harm or bias.

Explainable AI (XAI)

Approaches and methods that ensure the outputs and decision-making processes of an AI system are understandable and transparent to humans.

F

Fine-tuning

Process of adapting an already-trained AI model to a specific task using new data.

Foundation Models

AI models trained on large volumes of data that can be adapted to multiple tasks.

G

Generative AI

Type of AI capable of creating new content, such as text, images, code, or audio.

Google Gems

Customized AI assistants that help with specific tasks within a defined context.

GPUs

Graphics Processing Units. Hardware originally designed for graphics, now widely used to train AI models due to their parallel processing capacity.

Grounding

Process of connecting AI responses to real or verifiable sources to increase accuracy.

Guardrails

Safety rules and systems enforced on an AI model to ensure its outputs are appropriate, safe, and aligned with company policies.

H

Hallucination

A phenomenon where a generative AI model generates plausible-sounding but completely false or ungrounded responses.

Humans in the Loop (HITL)

Approach where humans supervise or intervene in AI systems to improve results or avoid errors.

Hypercomputer

Infrastructure that connects multiple computers (with GPUs/TPUs) to process large AI workloads.

I

Inference

The process of using a trained AI model to evaluate new data and generate a prediction, decision, or output.

L

Labeled Data

Data that has additional information (such as categories or names) that helps train models.

Large Language Models (LLMs)

AI models specialized in understanding and generating human language (such as ChatGPT or Claude).

LLM API

An application programming interface that allows developers to connect and integrate external language models into their own software or systems.

M

Machine Learning (ML)

Subfield of AI where machines learn from data without being explicitly programmed.

ML Learning Approaches

Types of machine learning: Supervised (uses labeled data), Unsupervised (finds patterns), Reinforcement (learns from feedback).

MLOps

A set of practices combining Machine Learning, DevOps, and data engineering to standardize and streamline the lifecycle of models in production.

Model Deployment

Process of putting a trained model into production so it can be used.

Model distillation

A technique that transfers knowledge from a large, complex model (teacher) to a smaller, faster, and more efficient one (student).

Model Management

Administration of the model lifecycle: monitoring, updating, and maintenance.

Model Training

Process of teaching a model using data.

Multimodal

The ability of an AI system to process, understand, and combine different types of data inputs simultaneously, such as text, images, audio, and video.

N

Natural Language Processing (NLP)

The branch of AI focused on the interaction between computers and human language, allowing machines to read, understand, and interpret native speech.

O

Overfitting

A modeling error that occurs when an AI algorithm adapts too closely to the training data, failing to predict or generalize accurately on new data.

P

Parameters

Internal variables within an AI model that are adjusted during the training process, determining how it handles information and generates outputs.

Prompt Chaining

Use of multiple prompts connected to each other to solve more complex tasks.

Prompt Engineering

Art of designing instructions (prompts) to get better results from AI models.

Prompt injection

A type of vulnerability or exploit where a user maliciously manipulates inputs to bypass the safety guardrails and system instructions of an LLM.

Q

Quantization

The process of reducing the precision of the weights of an AI model to decrease its memory footprint and accelerate inference speed.

R

ReAct (Reason + Act)

Framework that allows a model to reason and take actions step by step when facing a problem.

Reasoning Loop

Internal cycle of an AI agent where it analyzes information, reasons, and decides what to do next.

Reinforcement Learning

Type of learning where the model improves through trial and error, receiving rewards or penalties.

Responsible AI

Principles and practices to ensure that AI is safe, fair, and trustworthy.

Retrieval-Augmented Generation (RAG)

Technique that combines external information retrieval with response generation to improve accuracy.

Role Prompting

Technique where a role is assigned to the model to guide its behavior.

S

SAIF (Secure AI Framework)

Framework for developing and operating AI systems securely.

Secure AI

Practices to protect AI systems from attacks or malicious use.

Structured Data

Data organized in clear formats, such as tables or databases.

Structured vs Unstructured Data

Structured data is organized and easy to analyze; unstructured data requires more advanced techniques.

Supervised Learning

Training with labeled data where the model learns the relationship between input and output.

Swarm Intelligence

An AI approach inspired by collective biological systems where multiple decentralized agents cooperate to solve complex problems.

T

Temperature

A configuration setting in language models that controls the randomness, creativity, and predictability of the generated output.

Token

The basic unit of data (such as a word, syllable, or character fragment) into which text is split for a language model to process.

TPUs (Tensor Processing Units)

Chips designed specifically for artificial intelligence tasks, developed by Google.

Transfer learning

A Machine Learning technique where a pre-trained model for one task is repurposed as the starting point for a related task.

Transformer

A neural network architecture based on self-attention mechanisms that revolutionized language processing and serves as the foundation for modern LLMs.

U

Unlabeled Data

Data without classification or labels, used in unsupervised learning.

Unstructured Data

Data without a defined format, such as text, images, or videos.

Unsupervised Learning

Model that finds patterns in data without labels.

V

Vector database

A specialized type of database designed to store and query high-dimensional vector embeddings, crucial for semantic search and RAG systems.

W

Workflow Agents

AI designed to automate tasks and processes within a workflow.

Missing a term?

The glossary grows. If there's a concept you use that isn't here yet, write to us and we'll add it.

Suggest a term →