Glossary of AI Terms and Definitions

Updated: March 6, 2025

1. Glossary

A

Adversarial AI: Techniques used to manipulate AI models, often by introducing deceptive inputs to exploit vulnerabilities.

Agentic AI: Highly autonomous AI capable of goal-setting, long-term planning, and decision-making with minimal human intervention.

AGI (Artificial General Intelligence): AI systems with human-like ability to learn and perform any intellectual task.

AI Agent: A system that perceives its environment, makes decisions, and takes actions to achieve specific goals.

AI Ethics: The study and application of moral principles in AI development, ensuring responsible use of AI technologies.

Algorithm: A step-by-step procedure or formula for solving a problem or completing a task.

Ambient Intelligence: Electronic environments that are sensitive and responsive to human presence.

Artificial Intelligence (AI): The simulation of human intelligence by machines, including learning, reasoning, and self-correction.

Attention Mechanism: A technique allowing neural networks to focus on specific parts of input data.

AutoML (Automated Machine Learning): AI-driven tools that automate the process of training and optimizing machine learning models.

Autoencoder: Neural networks that learn efficient data representations by attempting to reconstruct their inputs.

Autonomous Agents: AI-driven systems that interact with environments in real-time, learning and adapting without direct human input (e.g., AI trading bots, robotics, digital assistants).

Augmented Intelligence: AI systems designed to enhance human capabilities rather than replace them.

B

Backpropagation: Algorithm for training neural networks by calculating gradients and adjusting weights.

Bias (AI Ethics): The presence of systematic errors in AI models that lead to unfair, skewed, or discriminatory outcomes. Bias in AI can arise from biased training data, flawed algorithms, or human prejudices embedded in decision-making processes.

Bias (Statistical): Systematic error that causes models to favor certain outcomes.

Bias in AI: Systematic errors in AI models caused by biased training data or flawed algorithms, leading to unfair outcomes.

Big Data: Extremely large datasets that may be analyzed to reveal patterns and trends.

Black Box Problem: The inability to explain or interpret how AI systems arrive at their decisions.

BERT (Bidirectional Encoder Representations from Transformers): Language model designed to understand context by considering words bidirectionally.

C

ChatGPT: Conversational AI system developed by OpenAI based on the GPT architecture.

Classification: Task of assigning inputs to specific categories or classes.

Clustering: Grouping similar data points together based on patterns or similarities.

CNN (Convolutional Neural Network): Neural network architecture specialized for image processing.

Computer Vision: AI technology that allows machines to interpret and analyze visual information from the world, such as images and videos.

Contrastive Learning: A self-supervised learning technique that helps models differentiate between similar and dissimilar examples to improve feature representation.

Corpus: A collection of texts used for training language models.

D

Data Mining: Process of discovering patterns and extracting valuable information from large datasets.

Data Science: Interdisciplinary field that uses scientific methods to extract knowledge from data.

Decision Tree: A tree-like model of decisions and their possible consequences.

Deep Learning: Machine learning approach using neural networks with multiple layers.

Deep Learning Iteration: The repeated training of neural networks through multiple layers to improve the model's ability to recognize complex patterns.

Diffusion Models: Generative models that create data by gradually denoising a random signal.

Dimensionality Reduction: Techniques to reduce the number of variables under consideration.

E

Edge AI: AI computation performed on local devices rather than relying on cloud-based processing, reducing latency and improving privacy.

Embeddings: Dense vector representations of data that capture semantic relationships.

Embodied AI: AI systems integrated with physical bodies that can interact with their environment.

Ensemble Learning: Combining multiple models to improve performance and robustness.

Ethical AI: The practice of designing AI systems that are fair, unbiased, and aligned with human values and ethical principles.

Explainable AI (XAI): AI systems designed to make their functioning transparent and interpretable.

F

Federated Learning: A decentralized ML approach where models are trained across multiple devices while keeping data local for privacy.

Feed-Forward Layers: Neural network layers that process token embeddings by applying transformations to refine meaning and improve understanding.

Feature: Individual measurable property or characteristic used as input for machine learning.

Feature Engineering: Process of selecting, transforming, or creating features to improve model performance.

Few-Shot Learning: A technique where AI models learn new tasks with only a few training examples.

Fine-tuning: Process of further training a pre-trained model on a specific dataset or task.

Foundation Models: Large-scale pre-trained models (e.g., GPT, BERT, DALL·E) that serve as a base for fine-tuning on specialized tasks.

G

GANs (Generative Adversarial Networks): Architecture using two neural networks competing against each other.

GPT (Generative Pre-trained Transformer): Large language model architecture using transformers.

Generative AI: AI systems that create new content, such as text, images, music, or videos, often using models like GANs and transformers.

Gradient Descent: A fundamental optimization algorithm used in training deep learning models by adjusting model weights to minimize error.

Graph Neural Networks: Neural networks that operate on graph-structured data.

H

Hallucination (AI): When AI systems generate false or misleading information that appears plausible but is incorrect.

Hidden Layer: Layers in neural networks between input and output layers.

Human-in-the-Loop (HITL) AI: AI systems that incorporate human oversight and feedback to improve accuracy and decision-making.

Hyperparameter: Parameter whose value is set before the learning process begins.

I

Image Generation: Creating new images using AI, often with text prompts.

Image Recognition: Technology that identifies objects, people, or scenes in digital images.

Inference: Process of using a trained model to make predictions.

Information Retrieval: Systems for obtaining relevant information from large collections.

Intent Recognition: Identifying the purpose or goal behind user inputs or queries.

K

K-means Clustering: Algorithm that partitions data into k distinct clusters.

Knowledge Distillation: Process of transferring knowledge from a large model to a smaller one.

Knowledge Graph: Network representing relationships between entities.

L

Large Language Model (LLM): A deep learning model trained on vast text datasets to generate human-like text based on statistical probabilities.

Latent Space: Compressed representation of data captured by models like VAEs.

Learning Rate: Hyperparameter controlling how much to adjust model weights during training.

LSTM (Long Short-Term Memory): Recurrent neural network architecture designed to remember long-term dependencies.

M

Model Context Protocol (MCP): MCP is aimed at enabling AI systems to better understand and adapt to the user’s intent and maintain continuity across conversations or interactions. It is particularly useful in applications like conversational AI, where maintaining context is crucial for providing coherent and relevant responses.

Machine Learning (ML): A subset of AI that enables computers to learn patterns from data and make predictions or decisions without explicit programming.

Metadata: Data that provides information about other data.

Metaverse: Collective virtual shared spaces, often incorporating AI technologies.

Multimodal AI: AI models that process and understand multiple types of data (e.g., text, images, and audio) simultaneously.

N

Natural Language Processing (NLP): A field of AI that enables machines to understand, interpret, and generate human language.

Natural Language Understanding (NLU): Subset of NLP focused on comprehension of language.

Neural Network: A computing system inspired by the human brain, consisting of interconnected layers of nodes (neurons) that process information.

Neuro-Symbolic AI: AI that combines neural networks (learning-based AI) with symbolic reasoning (rule-based AI) to improve explainability and logic-driven decision-making.

Normalization: Process of rescaling input data to improve model performance.

O

One-Hot Encoding: Representation of categorical variables as binary vectors.

Ontology: Formal naming and definition of categories, properties, and relations.

Overfitting: A modeling error in ML where the algorithm performs well on training data but poorly on new data due to excessive complexity.

P

Perceptron: Simplest type of artificial neural network.

Precision: Proportion of positive identifications that were actually correct.

Prediction & Sampling: The technique of generating outputs by estimating the most likely next elements based on probability distributions learned during training.

Prompt Engineering: Designing and refining inputs to guide AI systems toward desired outputs.

Pruning: Technique to reduce model size by removing unnecessary connections.

Q

Q-learning: Reinforcement learning algorithm to learn quality of actions.

Quantization: Technique to reduce model size by using fewer bits to represent weights.

Quantum AI: The application of quantum computing principles to enhance AI performance, particularly in solving complex optimization problems.

R

RAG (Retrieval-Augmented Generation): Technique combining information retrieval with text generation.

Recall: Proportion of actual positives that were correctly identified.

Recommendation System: System that suggests items or content to users based on their preferences.

Regression: Technique for predicting continuous values.

Reinforcement Learning (RL): A type of ML where an agent learns by interacting with an environment and receiving rewards or penalties based on its actions.

RNN (Recurrent Neural Network): Neural network architecture designed for sequential data.

Robotics Process Automation (RPA): The use of AI-driven bots to automate repetitive business processes and workflows.

S

Self-Attention Mechanism: A component of transformer models that determines how different parts of input data relate to each other, allowing for context-aware predictions.

Self-Improving AI: AI systems that can autonomously refine their models and adapt without explicit retraining, improving their performance over time.

Self-Supervised Learning: A form of ML where the model generates its own training labels from raw data, reducing the need for human-labeled datasets.

Semi-Supervised Learning: A hybrid ML approach that uses a small amount of labeled data combined with a large amount of unlabeled data.

Semantic Search: Search method based on meaning rather than literal matches.

Sentiment Analysis: Process of determining emotional tone behind text.

Stable Diffusion: Type of diffusion model for generating high-quality images.

Supervised Learning: A type of ML where the model is trained on labeled data, meaning the input data is paired with the correct output.

Support Vector Machine (SVM): Algorithm that finds the hyperplane that best divides a dataset.

Swarm Intelligence: A decentralized AI approach where multiple agents collaborate to solve complex problems, inspired by natural systems like ant colonies but also applied in robotics, traffic optimization, and financial modeling.

Synthetic Data: Artificially generated data used to train AI models when real-world data is scarce or sensitive.

T

TensorFlow: Open-source machine learning framework developed by Google.

Text-to-Image: AI systems that generate images based on text descriptions.

Tokenization & Embeddings: The process of breaking text into smaller units (tokens) and converting them into numerical representations in a multi-dimensional space.

Training Data: The dataset used to train an AI model, helping it learn and recognize patterns before making predictions.

Transfer Learning: The practice of applying knowledge learned from one task or domain to improve performance on a different but related task.

Transformer Model: A deep learning architecture, such as GPT and BERT, used in NLP for tasks like translation, summarization, and text generation.

Turing Test: Test of a machine's ability to exhibit intelligent behavior indistinguishable from a human.

U

Underfitting: When an ML model is too simple and fails to capture patterns in the training data, resulting in poor performance.

Unsupervised Learning: A type of ML where the model learns patterns and structures from unlabeled data without explicit instructions.

User Intent: The purpose or goal behind a user's query or action.

V

Validation Set: Subset of data used to tune hyperparameters.

VAE (Variational Autoencoder): Generative model that learns latent representations.

Vector Database: Database designed to store and query vector embeddings.

Vision Transformer (ViT): Transformer model adapted for computer vision tasks.

W

Weight (Neural Network): Parameter determining the strength of connection between neurons.

Word Embedding: Representation of words as vectors capturing semantic relationships.

Z

Zero-Shot Learning: An AI model's ability to make predictions about unseen data without direct prior training on that specific category.

2. Easily Confused Terms in AI

1. Bias (Statistical) vs. Bias (AI Ethics) vs. Bias in AI

These three related terms address different aspects of bias. Statistical bias relates to systematic errors in model predictions, AI Ethics bias refers to the broader moral implications, and Bias in AI combines both aspects.

2. Agentic AI vs. AI Agent vs. Autonomous Agents

The differences are meaningful but subtle:

An AI agent refers to any AI system that perceives its environment, makes decisions, and takes actions to achieve a goal. AI agents can range from simple rule-based systems to complex, autonomous learning models.
Agentic AI is a more advanced concept, referring to AI systems that demonstrate high levels of autonomy, adaptability, and goal-directed behavior. These systems are not just executing predefined tasks but can self-improve, plan, and act with minimal human intervention.
Autonomous Agents emphasizes real-time interaction and adaptation

Key Differences:

Feature	AI Agent	Agentic AI
Definition	Any AI system that perceives and acts to achieve a goal.	AI that exhibits high autonomy, adaptability, and long-term decision-making.
Autonomy Level	Varies (can be rule-based or learning-based).	High (capable of self-directed learning and decision-making).
Complexity	Can be simple (e.g., rule-based bots).	More advanced, often involving multi-step planning.
Human Oversight	Often requires supervision or pre-programmed logic.	Less human intervention; can generate and execute its own tasks.
Examples	Chatbots, recommendation systems, autonomous drones.	AI researchers, autonomous AI-driven businesses, advanced AI assistants.

4. Self-Supervised Learning vs. Unsupervised Learning

These learning paradigms sound similar but operate differently. Self-supervised learning generates its own labels from the data structure itself, while unsupervised learning identifies patterns without any labels. The "self" prefix suggests independence in both cases, making them easy to confuse.

5. XAI (Explainable AI) vs. Ethical AI

Both terms relate to responsible AI development but address different aspects. XAI focuses on making AI decisions interpretable and transparent, while Ethical AI encompasses broader moral considerations including fairness and societal impact.

6. Foundation Models vs. Large Language Models

Foundation Models is the broader category that includes LLMs but also encompasses other modalities like vision. People often use these terms interchangeably, especially when discussing text-based systems, which can lead to confusion about their scope.

7. Transfer Learning vs. Few-Shot Learning vs. Zero-Shot Learning

These terms represent a progression of related capabilities in AI systems, but the distinctions can be subtle. They all involve applying knowledge across contexts, but with varying amounts of example data.

8. Reinforcement Learning vs. Self-Improving AI

Both involve systems that get better over time, but through different mechanisms. Reinforcement learning uses explicit reward signals, while self-improving AI can encompass broader approaches to autonomous improvement.

9. Embeddings vs. Word Embeddings vs. Tokenization & Embeddings

These related terms describe different aspects of representing language in vector space. The overlap in terminology can make it difficult to understand where one concept ends and another begins.

10. Generative AI vs. Text-to-Image vs. Image Generation

Generative AI is the broader category, while the others are specific applications. The hierarchical relationship might not be clear from the current definitions.

11. Computer Vision vs. Image Recognition

Image recognition is a subset of computer vision, but the terms are sometimes used interchangeably, which can cause confusion about their scope.

12. Edge AI vs. IoT and AI

These terms have significant overlap since many IoT implementations leverage edge computing for AI processing. The connection between these concepts might not be evident from separate definitions.

To address these potential confusions, the glossary could benefit from:

Cross-references between related terms
Hierarchical organization that shows parent-child relationships between concepts
Comparison tables for closely related term sets (like you provided for AI Agent vs. Agentic AI)
Visual concept maps showing how different terms relate to each other

3. Comprehensive List of Generative AI Products

1. Text Generation and Conversational AI

OpenAI Products:

ChatGPT: Conversational AI assistant based on the GPT models, with web browsing and image recognition capabilities
GPT-4o: Latest multimodal language model with capabilities for text, image, and audio processing
GPT-4: Advanced large language model with enhanced reasoning capabilities
GPT-3.5 Turbo: Efficient version of GPT-3.5 optimized for chat applications
DALL-E 3: Text-to-image generation model integrated with ChatGPT
Sora: AI video generation system (announced but limited access)

Anthropic Products:

Claude 3.7 Sonnet: Current state-of-the-art conversational AI assistant
Claude 3.5 Sonnet: Conversational assistant optimized for reasoning
Claude 3.5 Haiku: Faster, more efficient version optimized for everyday tasks
Claude 3 Opus: Flagship model designed for complex reasoning and tasks
Claude Code: An agentic command line tool (in research preview)

Google Products:

Gemini: Family of multimodal AI models (formerly Bard)
Gemini Ultra: Google's most advanced AI model
Gemini Pro: Balanced model for various tasks
Gemini Nano: Lightweight model for on-device applications
PaLM 2: Large language model powering many Google products
MedPaLM: Specialized medical knowledge model
Med-PaLM 2: Enhanced medical AI model

Meta AI Products:

Llama 3: Open-source large language model
Llama 2: Previous version of Meta's open-source LLM
Llama 3.1: Enhanced version with improved capabilities
Meta AI Assistant: Consumer-facing chatbot available on WhatsApp, Instagram, etc.
Code Llama: Specialized model for programming tasks

Microsoft Products:

Copilot: Integrated AI assistant in Microsoft products (formerly Bing Chat)
Microsoft Copilot Studio: Tool for building custom copilots
Azure OpenAI Service: Cloud platform offering various AI models
Orca: Research model focused on reasoning

Other Text Generation Products:

Cohere Command: Enterprise-focused LLM
Mistral AI: Family of open-source language models
Mixtral 8x7B: Mixture-of-experts model from Mistral AI
Perplexity AI: AI search engine with generative responses
Character.AI: Platform for creating conversational AI characters
Pi: Personal AI assistant from Inflection AI
Claude.ai: Consumer-facing application for Anthropic's Claude
Khanmigo: Educational AI tutor from Khan Academy
Poe: Platform offering access to multiple AI models
Anthropic API: API access to Claude models
HuggingChat: Open-source chatbot by Hugging Face
LLaMA Chat: Meta's interface for Llama models

2. Open source

DeepSeek Products

Large Language Models:

DeepSeek LLM: Their flagship large language model series, with various sizes including 7B, 67B parameters
DeepSeek Coder: Specialized coding models designed specifically for programming tasks
DeepSeek-V2: Their second-generation language model with improved capabilities
DeepSeek-Math: Specialized model for mathematical reasoning

3. Image Generation

Text-to-Image Models:

Midjourney: AI image generation via Discord
DALL-E 3: OpenAI's advanced text-to-image model
DALL-E 2: Earlier version of OpenAI's image generator
Stable Diffusion: Open-source image generation model
Stable Diffusion XL: Enhanced version with improved quality
Stable Diffusion 3: Latest version with advanced capabilities
Adobe Firefly: Creative image generation integrated with Adobe products
Canva Text to Image: Integrated image generation in Canva
DreamStudio: Interface for Stability AI's models
Leonardo.AI: AI image generator for creative professionals
Ideogram: Text-to-image with strong typography handling
Imagen: Google's text-to-image model
Bing Image Creator: Microsoft's image generation tool
Craiyon (formerly DALL-E mini): Lightweight image generator

4. Image Editing/Enhancement:

Adobe Generative Fill: AI-powered content generation in Photoshop
Luminar Neo: AI photo editor with generative capabilities
Runway Gen-2: Image and video generation and editing
Lensa AI: Portrait enhancement and stylization
Topaz Photo AI: AI-powered photo enhancement

5. Video Generation:

Runway Gen-2: AI video generation and editing platform
Synthesia: AI video creation with virtual avatars
Pika Labs: Text-to-video generation
HeyGen: AI video generation platform
Sora: OpenAI's text-to-video model (limited access)
D-ID: AI-powered talking avatars from images
Descript: AI video and podcast editing tools
ElevenLabs: Voice cloning and video dubbing

6. Audio and Music Generation

ElevenLabs Voice Lab: AI voice generation and cloning
MUBERT: AI music generation
Suno: AI music creation from text prompts
Udio: AI music generation platform
SoundDraw: AI music composition tool
AudioCraft: Meta's audio generation model
MusicLM: Google's music generation system
Voicemod: Real-time voice modification
Play.ht: AI voice generation
Resemble.ai: Voice cloning technology
VALL-E: Microsoft's text-to-speech model

7. Code and Development

GitHub Copilot: AI pair programmer from GitHub/Microsoft
Replit Ghostwriter: AI coding assistant
Amazon CodeWhisperer: AWS coding assistant
Tabnine: AI code completion tool
Codeium: AI coding assistant
Cursor: AI-first code editor
DeepMind AlphaCode: AI system for competitive programming
Anthropic Claude Code: Code-focused AI tool

8. Specialized Industry Tools

Design and Creative:

Figma AI: AI features in the Figma design platform
Framer AI: Website building with AI
Jasper: AI content creation platform
Runway: Creative tools for video editing and generation

Business and Productivity:

Notion AI: Integrated AI in Notion workspace
Microsoft 365 Copilot: AI integrated across Microsoft Office
Google Workspace AI: AI features in Google's productivity suite
Anthropic Claude for Slack: Claude integration in Slack

Research and Education:

Elicit: AI research assistant
Consensus: AI-powered scientific search engine
ChatPDF: Conversational interface for PDFs
Khan Academy Khanmigo: Educational AI tutor

Customer Service and Sales:

HubSpot ChatSpot: CRM-integrated AI assistant
Intercom Fin: Customer service AI
Salesforce Einstein: AI for CRM and customer insights
Drift: Conversational marketing platform with AI

9. Multi-Modal Systems

GPT-4o: OpenAI's omni-modal system (text, image, audio)
Gemini: Google's multimodal model
Claude 3 models: Multimodal capabilities across text and images
Anthropic API: Multimodal capabilities through API
Fuyu-8B: Adept's open-source multimodal model