AI Glossary

Activation Function

A mathematical function in neural networks that determines whether a neuron should be activated, introducing non-linearity to enable complex pattern learning.

Adversarial Training

A training technique that strengthens AI models by exposing them to deliberately crafted malicious inputs during the training process.

Agentic AI

AI systems that can autonomously plan, make decisions, and take actions to achieve goals, going beyond simple response generation to act as independent digital agents.

Agentic Workflow

A structured sequence of tasks where AI agents autonomously collaborate to complete complex objectives with minimal human intervention.

AI Agent Framework

A software architecture that enables autonomous AI systems to perceive, reason, and act in environments using tools, memory, and planning capabilities.

AI Alignment

The challenge of ensuring AI systems pursue intended goals and human values, preventing misalignment between what we want AI to do and what it actually does.

AI Democratization

The process of making AI tools, technologies, and knowledge accessible to everyone, regardless of technical background or organizational resources.

AI Devices

Hardware devices with built-in artificial intelligence capabilities that can process data, make decisions, and perform tasks autonomously without constant cloud connectivity.

AI Ethics

The field of study examining moral principles, values, and guidelines for responsible AI development and deployment to ensure fairness and prevent harm.

AI Governance

The frameworks, policies, and practices that guide the responsible development, deployment, and oversight of artificial intelligence systems.

AI in Climate Modeling

The application of artificial intelligence techniques to predict, simulate, and analyze climate patterns and environmental changes.

AI Income Bot

Automated software that uses AI to generate revenue through trading, content creation, or business automation tasks.

AI Maturity Model

A framework that assesses an organization's AI capabilities across different levels, from basic experimentation to advanced enterprise-wide implementation.

AI Orchestration

Coordination and management of multiple AI models and services working together to accomplish complex tasks efficiently.

AI Pair Programming

A collaborative coding approach where AI assistants work alongside human developers to write, debug, and improve code in real-time.

AI Policy Compliance

The process of ensuring AI systems adhere to regulatory requirements, ethical guidelines, and organizational policies throughout development and deployment.

AI Readiness

An organization's capability to successfully adopt, implement, and scale artificial intelligence technologies across their operations.

AI Slop

Low-quality, generic AI-generated content that floods digital platforms, often characterized by poor accuracy, repetitive patterns, and lack of human oversight.

AI Watermarking

A technique for embedding invisible digital signatures into AI-generated content to identify its artificial origin and combat deepfakes and misinformation.

AI-Assisted Coding

Using artificial intelligence tools to help write, debug, and improve code by providing suggestions, completions, and explanations.

Algorithmic Bias

Unfair or discriminatory outcomes produced by AI systems due to biased training data, flawed algorithms, or prejudiced design decisions.

API Gateway for AI

A centralized interface that manages, secures, and routes requests between applications and AI services, providing unified access to multiple AI models and APIs.

Deep Learning Intermediate

Artificial General Intelligence (AGI)

Hypothetical AI that matches or exceeds human cognitive abilities across all domains, representing the next major milestone in AI development.

Attention Mechanism

A neural network component that allows models to focus on relevant parts of input data when making predictions, enabling better understanding of context and relationships.

Automatic Speech Recognition (ASR)

Technology that converts spoken language into text, enabling voice-controlled applications and transcription services through machine learning models.

AutoPrompt Optimization

Automated techniques that iteratively improve AI prompts through algorithmic refinement to achieve better model outputs and performance.

2 terms

Backpropagation

The core learning algorithm in neural networks that calculates gradients and updates weights by propagating errors backward through the network layers.

Batch Normalization

A deep learning technique that normalizes inputs to each layer during training, accelerating convergence and improving model stability.

11 terms

NLP Intermediate

Chain-of-Thought Prompting (CoT)

A prompting technique that guides AI models to break down complex problems into step-by-step reasoning, improving accuracy on logical tasks.

Code Interpreter / Advanced Data Analysis

AI-powered tools that execute code, analyze data, and generate insights through natural language commands, combining programming capabilities with conversational interfaces.

Cognitive Computing

AI systems that simulate human thought processes using machine learning, natural language processing, and pattern recognition to understand and interact naturally.

Consistency Models

A class of generative models that enable fast, high-quality sampling by learning to map noise directly to data in a single step or few steps.

Constitutional AI

An AI training approach that teaches models to follow a set of principles or 'constitution' to behave helpfully, harmlessly, and honestly.

Context Collapse

Loss of contextual information when AI models encounter situations outside their training data or when context gets compressed beyond useful limits.

Context Window

The maximum amount of text (measured in tokens) that an AI model can process and remember at once during a conversation or task.

Continual Learning

Machine learning approach enabling models to learn new tasks sequentially without forgetting previously learned information.

Conversational Interface Design

The practice of designing user interfaces that enable natural language communication between humans and AI systems through voice or text.

Convolutional Neural Network (CNN)

A deep learning architecture designed for processing grid-like data such as images, using convolutional layers to detect features like edges, shapes, and patterns.

Cross-Market AI

AI systems designed to operate across multiple market sectors or geographic regions, adapting to different regulatory, cultural, and business environments.

8 terms

Data Curation

The systematic process of collecting, cleaning, organizing, and maintaining high-quality datasets for AI and machine learning applications.

Data Poisoning

A cyberattack where malicious data is injected into training datasets to compromise AI model behavior and performance.

Decentralized AI

AI systems distributed across multiple nodes without central control, enabling privacy, resilience, and collaborative intelligence.

Deep Learning

A subset of machine learning using multi-layered neural networks to automatically learn complex patterns from data without explicit programming.

Diffusion Model

A generative AI model that creates images by gradually removing noise from random data, like sculpting a statue by slowly revealing it from marble.

Direct Preference Optimization (DPO)

A training method that directly optimizes AI models based on human preferences without requiring a separate reward model, making alignment training simpler.

Distillation Training

A machine learning technique where a smaller 'student' model learns to mimic a larger 'teacher' model, achieving similar performance with reduced computational requirements.

Dropout (Regularization)

A regularization technique that randomly sets neurons to zero during training to prevent overfitting and improve model generalization.

5 terms

Edge AI

Running AI inference directly on local devices like smartphones and IoT sensors rather than sending data to cloud servers for processing.

Edge Computing for AI

Processing AI workloads on local devices or nearby servers instead of centralized cloud infrastructure to reduce latency and improve privacy.

NLP Intermediate

Embedding

Dense vector representations that capture semantic meaning of words, sentences, or other data in high-dimensional space.

Emergent Behavior

Complex, unpredictable behaviors that arise from AI systems beyond their original training, often appearing when systems interact or scale up.

Explainable AI (XAI)

AI systems designed to provide clear, understandable explanations for their decisions and predictions to human users.

5 terms

Federated Learning

Machine learning approach where models are trained across decentralized devices without sharing raw data, preserving privacy while enabling collaborative learning.

Few-Shot Learning

A machine learning approach where models can learn new tasks or recognize new patterns using only a few training examples.

Flash Attention

A memory-efficient attention algorithm that reduces GPU memory usage and speeds up transformer training without sacrificing accuracy.

Foundation Model

Large-scale AI model trained on broad data that serves as a base for adapting to multiple downstream tasks and applications.

Function Calling

A feature that allows AI models to execute specific functions or tools by generating structured calls based on natural language requests.

5 terms

Generative Adversarial Network (GAN)

AI architecture using two competing networks to generate realistic synthetic content like images, videos, and audio through adversarial training.

Generative Pre-trained Transformer (GPT)

A type of large language model that uses transformer architecture to generate human-like text by predicting the next word in a sequence.

Gradient Boosting

An ensemble machine learning technique that builds models sequentially, with each new model correcting errors from previous ones to improve predictions.

Graphics Processing Unit (GPU)

Specialized computer chips designed for parallel processing, essential for training and running AI models efficiently at scale.

Guardrails

Safety mechanisms that constrain AI system behavior to prevent harmful, inappropriate, or unintended outputs while maintaining functionality.

4 terms

Hallucination Suppression

Techniques to reduce AI's tendency to generate false or fabricated information by improving training methods, validation processes, and output filtering.

Human-in-the-Loop (HITL)

AI approach where humans actively participate in training, validating, and refining machine learning models throughout the development cycle.

Hybrid Search (Keyword + Vector)

A search technique that combines traditional keyword matching with vector similarity search to provide more accurate and contextually relevant results.

Hyperparameter

Configuration settings that control how machine learning algorithms learn, set before training begins and not learned from data.

3 terms

Image-to-Image (img2img)

AI technique that transforms one image into another while preserving structural elements, enabling style transfer, enhancement, and guided image generation.

NLP Intermediate

In-Context Learning

Ability of large language models to learn and perform new tasks using only examples provided in the input prompt, without parameter updates.

Inference Optimization

Techniques for accelerating AI model predictions and reducing computational costs during deployment, focusing on speed and efficiency rather than training performance.

1 terms

Jailbreaking (AI)

Techniques used to bypass AI safety guardrails and content filters, making AI models produce outputs they were designed to refuse or avoid generating.

1 terms

Knowledge Distillation

A technique where a smaller 'student' model learns from a larger 'teacher' model, transferring knowledge to create efficient models with similar performance.

5 terms

Large Language Model (LLM)

AI models trained on massive text datasets to understand and generate human-like text, powering chatbots and content creation tools.

Deep Learning Intermediate

Latent Space

A compressed mathematical representation where AI models encode high-dimensional data into lower-dimensional vectors that capture essential features.

Long Short-Term Memory (LSTM)

A type of recurrent neural network designed to remember information over long sequences, solving the vanishing gradient problem in traditional RNNs.

Loss Function

A mathematical function that measures how far a machine learning model's predictions are from the correct answers, guiding the training process.

Low-Rank Adaptation (LoRA) / Quantized Low-Rank Adaptation (QLoRA)

Memory-efficient techniques for fine-tuning large language models using low-rank matrix decomposition, with QLoRA adding quantization for further optimization.

9 terms

Machine Learning Operations (MLOps)

The practice of deploying, monitoring, and maintaining machine learning models in production environments, combining ML, DevOps, and data engineering practices.

Deep Learning Advanced

Mixture of Agents (MoA)

An AI architecture where multiple specialized AI agents collaborate to solve complex tasks, each contributing their unique expertise to achieve better overall performance.

Mixture of Experts (MoE)

A neural network architecture that uses multiple specialized sub-models (experts) and a gating mechanism to route inputs to the most relevant experts.

Model Context Protocol (MCP)

A standardized protocol for AI systems to access and share data across different platforms, enabling seamless interoperability between AI models and applications.

Model Deployment

The process of making trained AI models available for real-world use in production environments and applications.

Model Monitoring

The practice of continuously tracking ML model performance, data drift, and behavior in production to ensure reliable AI system operation.

Model Quantization

Optimization technique that reduces AI model size and computational requirements by representing weights and activations with fewer bits while maintaining performance.

Model Router / LLM Routing

Intelligent system that automatically selects the most appropriate AI model or LLM for each query based on task complexity, cost, and performance requirements.

Multimodal AI

AI systems that can understand and generate content across multiple types of data like text, images, audio, and video simultaneously.

6 terms

Natural Language Processing (NLP)

A branch of AI that enables computers to understand, interpret, and generate human language through computational techniques and machine learning algorithms.

Deep Learning Advanced

Neural Architecture Search (NAS)

Automated method for designing optimal neural network architectures using AI, eliminating manual trial-and-error in network design.

Audio & Speech Advanced

Neural Codec

AI-powered compression system that uses neural networks to encode and decode audio, video, or data more efficiently than traditional codecs.

Neural Radiance Fields (NeRF)

A technique that creates photorealistic 3D scenes from 2D photos using neural networks to represent how light interacts with objects.

Computer Vision Advanced

Neural Rendering

A computer graphics technique that uses neural networks to generate photorealistic images and 3D scenes, combining traditional rendering with AI-powered synthesis.

Computer Vision Intermediate

Neuromorphic Computing

Computing architecture that mimics the structure and function of biological neural networks, using brain-inspired chips for efficient AI processing.

5 terms

Object Detection

A computer vision technique that identifies and locates multiple objects within images or videos, providing both classification and precise bounding box coordinates.

Open Weights vs Open Source AI

The distinction between AI models with publicly available parameters (open weights) versus complete open development with accessible code and training data (open source).

Optical Character Recognition (OCR)

Technology that converts images of text into machine-readable text, enabling computers to recognize and extract written or printed characters.

Optical Flow

A computer vision technique that tracks the movement of objects and pixels between consecutive video frames to understand motion patterns.

Overfitting

When a machine learning model memorizes training data too closely, performing well on training data but poorly on new, unseen data.

7 terms

Parameter-Efficient Fine-Tuning (PEFT)

Training technique that adapts large pre-trained models for specific tasks by updating only a small subset of parameters rather than the entire model.

Perplexity

A metric measuring how well a language model predicts text sequences, with lower values indicating better performance and understanding.

Principal Component Analysis (PCA)

A dimensionality reduction technique that transforms high-dimensional data into fewer dimensions while preserving maximum variance and information.

Prompt Chaining

A technique that connects multiple AI prompts sequentially, where the output of one prompt becomes the input for the next, enabling complex multi-step reasoning and task completion.

Prompt Engineering

The practice of crafting effective text inputs to guide AI language models toward producing desired outputs, combining art and science to optimize AI interactions.

Prompt Injection

A security vulnerability where malicious instructions are embedded in user input to manipulate AI models into ignoring their original instructions.

Prompt UX

The user experience design of AI prompt interfaces, focusing on making AI interactions intuitive, effective, and user-friendly.

1 terms

Quantum Machine Learning (QML)

A hybrid approach combining quantum computing principles with machine learning algorithms to potentially solve complex problems faster than classical computers.

9 terms

RAG 2.0 / Advanced RAG

Next-generation Retrieval-Augmented Generation systems that enhance traditional RAG with sophisticated retrieval, multi-step reasoning, and adaptive processing techniques.

Random Forest

An ensemble machine learning algorithm that combines multiple decision trees to improve prediction accuracy and reduce overfitting.

Real-Time Voice Conversion (RVC)

AI technology that transforms one person's voice to sound like another person in real-time, maintaining speech content while changing vocal characteristics.

Reasoning Models

AI models specifically designed to perform logical reasoning, multi-step problem-solving, and complex inference tasks beyond simple pattern matching.

Recurrent Neural Network (RNN)

A neural network architecture designed to process sequential data by maintaining memory of previous inputs through recurrent connections.

Red Teaming

A practice of systematically testing AI systems by simulating adversarial attacks to identify vulnerabilities, biases, and safety risks before deployment.

Reinforcement Learning (RL)

A machine learning approach where agents learn optimal behavior through trial-and-error interactions with an environment, receiving rewards or penalties.

Reinforcement Learning from Human Feedback (RLHF)

A training method that uses human preferences to guide AI models toward more helpful, harmless, and honest behavior.

Retrieval-Augmented Generation (RAG)

A technique that combines AI text generation with real-time information retrieval from external databases to provide accurate, up-to-date responses.

16 terms

Self-Consistency

A prompting technique that generates multiple reasoning paths for a problem and selects the most frequent answer to improve accuracy and reliability.

Computer Vision Intermediate

Self-Reflective AI

AI systems capable of examining and analyzing their own reasoning processes, decisions, and performance to improve future outputs.

Semantic Segmentation

Computer vision technique that classifies every pixel in an image, creating detailed maps that identify and outline different objects and regions.

Small Language Model (SLM)

Compact AI language models with fewer parameters than LLMs, designed for efficiency while maintaining strong performance on specific tasks.

Sovereign AI

AI systems developed and controlled entirely within a nation's borders, ensuring data sovereignty, technological independence, and alignment with national values.

NLP Advanced

Speculative Decoding

Optimization technique that speeds up language model inference by predicting multiple tokens ahead and verifying them in parallel.

Speech-to-Text (STT)

AI technology that converts spoken language into written text, enabling voice interfaces, transcription services, and accessibility tools.

Stable Diffusion

Open-source text-to-image AI model that generates high-quality images from text descriptions using diffusion techniques.

State Space Models (SSMs / Mamba)

Neural network architectures that model sequences using state-based representations, offering efficient alternatives to Transformers with linear scaling.

Structured Outputs / JSON Mode

AI model capabilities that generate responses in specific structured formats like JSON, ensuring consistent, parseable data output for applications.

Supervised Fine-Tuning (SFT)

A training method that refines pre-trained AI models using labeled data to improve performance on specific tasks or align behavior with desired outcomes.

Supervised Fine-Tuning + Direct Preference Optimization Pipeline (SFT + DPO Pipeline)

A two-stage training approach that first fine-tunes an AI model on supervised data, then optimizes it using human preferences to create more helpful and aligned AI systems.

Supervised Learning

A machine learning approach where models learn from labeled training data to make predictions on new, unseen data.

Support Vector Machine (SVM)

A machine learning algorithm that finds the optimal boundary between different classes of data by maximizing the margin between data points.

Synthetic Biology with AI

The integration of artificial intelligence with synthetic biology to design, engineer, and optimize biological systems for practical applications.

Synthetic Data Generation

Creating artificial training data using algorithms and AI models to supplement or replace real-world datasets for machine learning applications.

15 terms

Technological Singularity

A hypothetical future point when AI surpasses human intelligence, leading to unpredictable, exponential technological growth and societal transformation.

Temperature

A parameter that controls randomness in AI text generation, with lower values producing more focused outputs and higher values creating more creative responses.

Test-Time Training (TTT)

A technique where AI models continue learning and adapting during inference, improving performance on specific tasks or data in real-time.

Text-to-Image Generation

AI technology that creates realistic images from written descriptions, using models like Stable Diffusion and DALL-E to generate visual content from text prompts.

Text-to-Speech (TTS)

AI technology that converts written text into natural-sounding spoken audio, enabling applications like virtual assistants and accessibility tools.

Token Limit

Maximum number of tokens (words, characters, or symbols) that an AI model can process in a single input or output, determining conversation length and context size.

Tokenization

The process of breaking down text into smaller units called tokens (words, subwords, or characters) that AI models can understand and process.

Tool Use

AI capability that allows models to interact with external tools and APIs to perform tasks beyond their training, like web search or calculations.

Top-K / Top-P Sampling (Nucleus Sampling)

Text generation techniques that control randomness by selecting from top-K tokens or top-P probability mass to balance creativity and coherence.

Transfer Learning

Machine learning technique where a model trained on one task is adapted for a related task, reducing training time and data requirements.

Transferable Skills in AI

Core competencies and knowledge from traditional fields that can be successfully applied and leveraged in artificial intelligence careers and projects.

Transformer Architecture

A neural network design that revolutionized AI by using attention mechanisms to process sequences of data efficiently and in parallel.

Transparent AI

AI systems designed to provide clear explanations of their decision-making processes and reasoning to users and stakeholders.

NLP Advanced

Tree of Thoughts (ToT)

An advanced prompting technique that explores multiple reasoning paths simultaneously, allowing AI models to backtrack and find optimal solutions to complex problems.

Turing Test

A test proposed by Alan Turing to evaluate whether a machine can exhibit intelligent behavior indistinguishable from a human in conversation.

3 terms

Underfitting

When a machine learning model is too simple to capture the underlying patterns in data, resulting in poor performance on both training and test sets.

Unsupervised Learning

A machine learning approach where algorithms find hidden patterns in data without labeled examples or explicit guidance.

Upscaling / Super-Resolution

AI techniques that enhance low-resolution images to higher resolutions while reconstructing fine details and textures that weren't in the original.

7 terms

Variational Autoencoder (VAE)

A generative AI model that learns to compress and reconstruct data, enabling generation of new samples similar to training data.

Variational Inference (VI)

A machine learning technique that approximates complex probability distributions using simpler, tractable distributions to make inference computationally feasible.

Vector Database

A specialized database designed to store and efficiently search through high-dimensional vector embeddings, commonly used in AI applications.

Vibe Coding

A casual programming approach where developers write code based on intuition and feel rather than rigorous planning or formal methodologies.

Video Diffusion Models

AI models that generate videos by learning to reverse a noise-adding process, creating realistic video content from text prompts or other inputs.

Audio & Speech Intermediate

Vision Transformer (ViT)

Deep learning architecture that applies the transformer model, originally designed for text, directly to image recognition tasks by treating image patches as tokens.

Voice Cloning

AI technology that creates synthetic speech mimicking a specific person's voice using deep learning and audio samples.

2 terms

Wasserstein Distance

A mathematical metric that measures the minimum cost to transform one probability distribution into another, widely used in generative AI models.

Computer Vision Intermediate

World Model

AI system that creates internal representations of environments to predict outcomes and improve decision-making through simulation.

1 terms

YOLO (You Only Look Once)

A real-time object detection algorithm that processes entire images in a single forward pass, achieving fast and accurate object detection for applications.

2 terms

Zero-Knowledge Proofs (ZKP) in AI

Cryptographic protocols that allow AI systems to prove computational integrity or verify data without revealing the underlying information or model details.