AI Glossary

A massive AI model trained on vast amounts of text data. The foundational technology behind ChatGPT, Claude, and other AI assistants.

The art and science of crafting effective instructions for AI to produce desired outputs. The key to unlocking AI's full potential.

A technique that improves AI response accuracy by retrieving relevant information from external databases before generating answers.

The phenomenon where AI generates plausible-sounding but factually incorrect information. The most critical concern when using AI.

The smallest unit of text that an AI model processes. Used as the basis for pricing and context window limits.

The process of further training an existing AI model on specific data to specialize it for particular tasks.

A technique that converts text or images into numerical vectors. The foundational technology for similarity search and RAG.

An AI's ability to understand and generate across multiple data types — text, images, audio, and video.

An AI system that autonomously sets goals, makes plans, uses tools, and executes tasks.

An open standard protocol for AI models to securely connect with external tools and data sources. Proposed by Anthropic.

The practice of optimizing content so that AI-powered search engines cite it in their generated responses. The new SEO for the AI era.

Google's feature that displays AI-generated summaries at the top of search results. A transformative shift in the search experience.

The neural network architecture that underpins modern AI models. Self-attention is its core innovation.

A machine learning approach using multi-layered neural networks. The core technology driving modern AI.

The field of AI focused on enabling computers to understand and generate human language. The foundation for translation and chatbots.

The maximum amount of text an AI model can process at once, measured in tokens.

OpenAI's series of large language models. The foundational technology behind ChatGPT, which brought AI to the mainstream.

An AI model that gradually generates data from noise. The core technology behind image and video generation AI.

A reinforcement learning method that uses human feedback to improve AI model outputs. Essential for AI safety.

AI with human-level or greater intellectual capabilities across all domains. One of the ultimate goals of AI research.

The umbrella term for AI technologies that automatically create new content — text, images, audio, video, and more.

The input text or instructions given to an AI model. A critical factor that shapes the quality of AI outputs.

A program that automatically converses with humans through text or voice. Dramatically enhanced by AI technology.

The field of AI that enables computers to understand and analyze images and video. Applied in autonomous driving and medical imaging.

An AI training method that learns optimal behavior through trial and error. Used in game AI and robotics control.

A technique that generates realistic data by pitting two neural networks against each other. A foundational milestone in AI image generation.

A database specialized for storing and searching high-dimensional vector data. A foundational technology for RAG.

An open-source framework that streamlines the development of LLM-powered applications.

An interface that enables software systems to exchange data and functionality. The foundation for integrating AI capabilities into your own services.

AI models whose weights and code are publicly available for anyone to use, modify, and redistribute.

Technology that runs AI processing directly on devices rather than in the cloud. Offers low latency and strong privacy protection.

A technique that reduces AI model size and computational cost by lowering numerical precision.

A method for efficiently fine-tuning large models by training only a small number of additional parameters.

The file format used by llama.cpp for AI models. The standard format for running LLMs locally.

A collection of techniques and strategies for reducing the problem of AI generating factually incorrect information.

A general-purpose AI model pre-trained on large-scale data, adaptable to a wide variety of downstream tasks.

The field concerned with ethical principles and social responsibility in the development and deployment of AI.

The frameworks, rules, and structures ensuring the proper management and deployment of AI within organizations and society.

A technique that transfers knowledge from a large teacher model to a smaller student model. Enables lightweight AI deployment.

The process by which a trained AI model generates predictions or responses for new inputs.

The component that converts text into a sequence of tokens that an AI model can process.

A neural network mechanism that dynamically computes the relevance between elements of input data. The core innovation of the Transformer.

A technique that teaches AI new tasks with just a handful of examples. Used by providing examples within the prompt.

An AI's ability to perform unfamiliar tasks without any prior examples. A key indicator of LLM versatility.

Legal issues surrounding AI-generated content and copyright law. Concerns arise from both training data and generated outputs.

A lightweight language model with fewer parameters. Excels in efficiency and cost-effectiveness.

An AI architecture that efficiently processes inputs by routing them to specialized sub-networks.

The end-to-end processing flow in a RAG system, from data retrieval to answer generation.

The technology that enables AI to automatically produce natural, human-like text. The most fundamental capability of LLMs.

AI technology that automatically generates images from text descriptions. Revolutionizing design and creative work.

Tools that use AI to assist with code completion, generation, and debugging. Dramatically boosts developer productivity.

Next-generation search services where AI synthesizes and summarizes information to answer questions directly.

Tools that let you build AI-powered apps and workflows without any programming knowledge.

An AI design philosophy where systems autonomously understand goals, plan, make decisions, and execute tasks iteratively. The biggest trend of 2026.

A system where multiple AI agents collaborate and divide tasks to accomplish complex objectives.

An AI model that can simulate physical laws and causal relationships of the real world. Seen as the next frontier beyond LLMs.

A prompting technique that improves answer accuracy by guiding AI through step-by-step reasoning.

The technique of anchoring AI outputs to reliable external data sources to ensure factual accuracy.

Artificially generated training data produced by AI. Used for privacy protection and overcoming data scarcity.

The knowledge and ability to understand how AI works and use it effectively and responsibly. An essential skill for today's professionals.

A security attack that manipulates AI behavior through malicious inputs. A major threat in AI deployment.

Technology that uses AI to realistically synthesize or replace a person's face or voice. Misuse for disinformation and fraud is a growing social concern.

An open protocol for safe communication and collaboration between AI agents from different vendors. Proposed by Google.

A development approach where you describe the desired outcome in natural language and AI generates the code interactively. Named 2025 Word of the Year by Collins Dictionary.

An AI model that executes step-by-step logical reasoning before answering. Dramatically improves accuracy on complex problems.

A term for the flood of low-quality digital content mass-produced by AI. Named Merriam-Webster's 2025 Word of the Year.

AI technology that can perceive, reason, and act in the real world. Gaining momentum in 2026 as the foundation for robotics and autonomous driving.

The strategic design and optimization of context information sent to AI models. An evolution of prompt engineering.

The degradation of AI output quality that occurs when models are repeatedly trained on their own generated data. Related to the so-called '2026 problem.'

Anthropic's approach of giving AI a set of ethical principles — a 'constitution' — to guide safe, value-aligned behavior.

The technical and theoretical research field dedicated to ensuring AI systems operate in accordance with human values and intentions.

An advanced RAG approach that combines retrieval with autonomous AI agent decision-making for multi-step information gathering.

The mechanism that enables AI agents to invoke external APIs and functions to execute real-world tasks. The core technology behind agentic AI.

Standardized tests and metrics used to objectively compare and evaluate AI model performance.

A technique that transfers knowledge from large AI models to smaller ones, creating lightweight yet high-performing models.

A capability that allows LLMs to automatically invoke external functions and APIs to perform real-world tasks.

A parallel computing processor essential for AI training and inference. NVIDIA's H100 is the industry standard.

A method of training models on diverse instruction-response pairs so they can properly follow human instructions.

A high-dimensional internal space where AI models represent the essential features of data in compressed form.

A technique that combines multiple AI models to leverage each one's strengths and produce higher-quality outputs.

The most fundamental computational model in AI, inspired by the neural circuits of the human brain. The foundation of deep learning.

An AI model whose trained weights (parameters) are publicly released, allowing anyone to download and use it.

The technology for efficiently searching and extracting relevant information from large datasets. The foundation of RAG.

The umbrella term for techniques and processes that ensure AI operates safely and doesn't generate harmful outputs.

A predictable relationship between model size, data volume, compute, and performance that follows power-law patterns.

A feature that constrains LLM outputs to specific formats like JSON. Essential for application integration.

A parameter that controls the diversity and creativity of AI outputs. Lower values produce certainty, higher values produce creativity.

The dataset used to train AI models. It fundamentally determines model performance and quality.

A machine learning technique that applies knowledge learned from one task to another. Dramatically improves AI development efficiency.

A technique for artificially increasing training data through transformations. Improves model accuracy via image rotations, text paraphrasing, and more.

When an AI model becomes too closely fitted to training data, causing poor performance on new, unseen data.

A normalization technique that stabilizes and accelerates neural network training. A standard technique in deep learning.

A function that introduces non-linearity into neural networks. ReLU, Sigmoid, and GELU are common examples.

The fundamental optimization algorithm for training AI model parameters. The core of the deep learning training process.

An algorithm for efficiently computing gradients in neural network training. The foundational technique of deep learning.

Technology for automatically detecting and identifying misinformation (hallucinations) generated by AI. Key to AI reliability.

A distributed learning approach where models are trained across multiple devices without centralizing data. A privacy-preserving AI technology.

Technology for embedding invisible identification information in AI-generated content. Essential for combating deepfakes.

A safety evaluation method that deliberately probes AI models for vulnerabilities and harmful outputs.

An efficient AI model design that deactivates most parameters and computes only the necessary parts for each input.

An AI technique for learning new tasks and data without forgetting previous knowledge. Essential for keeping AI up to date.

A next-generation RAG technology that retrieves and uses not just text but also images, tables, and diagrams to generate answers.

Technology for coordinating and integrating multiple AI models and services to manage complex automated workflows.

An OpenAI multimodal model that learned the relationship between text and images. Foundation technology for image search and generation.

A technology that adds precise control conditions like composition and poses to image generation AI. Enables more accurate image output.

An open-source image generation AI model developed by Stability AI. Can run locally for free.

AI technology that automatically generates images from text descriptions. Used in DALL-E, Midjourney, Stable Diffusion and more.

AI technology that automatically generates video from text descriptions. Sora, Runway, and Kling AI are leading examples.

AI technology that converts text into natural-sounding speech. Foundation technology for narration and voice assistants.

AI technology that automatically converts speech to text. Essential for meeting transcription and subtitle generation.

AI technology that generates or transforms new images based on input images. Used for style transfer and refinement.

AI technology for naturally replacing or restoring parts of an image. Used for removing unwanted objects.

AI technology that naturally extends an image beyond its original borders. Auto-generates content outside the frame.

AI technology that enlarges low-resolution images to high resolution. Enhances quality by filling in detail.

AI technology that applies the artistic style of one image to another. Used for converting photos to painting styles.

AI-powered search that understands meaning and context rather than just matching keywords. Core technology behind RAG.

A method for measuring similarity by computing the angle between two vectors. Standard metric for comparing embeddings.

Technology that automates the machine learning model building process. Handles everything from data preprocessing to model selection and hyperparameter tuning.

Practices and tools for efficiently managing the development, deployment, and monitoring of machine learning models.

An automated workflow for collecting, transforming, storing, and delivering data. The foundation of AI/ML operations.

The technique of designing and creating useful input variables (features) from raw data to improve ML model performance.

A technique that splits data into multiple groups to fairly evaluate model generalization. Effective for detecting overfitting.

The process of optimizing model configuration values to maximize performance. Grid search and Bayesian optimization are common methods.

A development approach where AI acts as a partner, assisting with code creation in real-time.

An AI feature that predicts and auto-completes code as programmers type. Dramatically improves development efficiency.

Technology where AI automatically reviews pull requests and code changes, identifying bugs and improvements.

Technology where AI analyzes source code and automatically generates unit tests and test cases.

Technology where AI analyzes code structure and automatically suggests improvements for readability and maintainability.

A general term for AI assistants that work as a human's 'co-pilot.' Microsoft Copilot and GitHub Copilot are leading examples.

Automated business processes built by connecting multiple AI tools and services in sequence.

A technique that executes multiple prompts in sequence, using each step's output as the next step's input. Breaks complex tasks into manageable steps.

A special instruction that defines an AI model's behavior and role upfront. The design foundation of AI applications.

A prompting technique where you provide a few input-output examples before asking the AI to perform a task. Effective for improving accuracy.

The ability of AI to instantly learn new tasks from examples and context provided in a prompt. A revolutionary LLM capability.

The research and technology field focused on ensuring AI systems operate safely without harming people or society.

Systematic prejudice in AI outputs caused by biased training data or model design. Leads to unfair results across gender, race, age, and other attributes.

Technology that makes AI decision-making understandable to humans. Essential for ensuring trustworthiness and transparency.

Legal regulations and rules governing AI development and use. The EU AI Act is the world's first comprehensive AI regulation.

The world's first comprehensive AI regulation law enacted by the EU. Takes a risk-based regulatory approach.

AI technology that auto-generates 3D models from text descriptions. An innovative technology for game development and 3D design.

AI technology that reconstructs 3D scenes from multiple 2D photos. Generates realistic 3D representations from photographs.

AI technology that learns a specific person's voice and reproduces it for text-to-speech. Can replicate voices from small audio samples.

AI technology that translates speech and text into other languages in real time. Breaks down international communication barriers.

A method for directly optimizing AI models from human preference data. Simpler and more efficient than RLHF.

A framework that has LLMs alternate between reasoning and acting (tool use) to solve complex tasks.

The ability of AI models to call external tools and APIs to retrieve information and perform actions.

A knowledge base that represents relationships between entities in a graph structure. Used for AI knowledge representation and reasoning.

A comprehensive approach to developing and deploying AI while ensuring fairness, transparency, and safety.

Machine learning that runs on ultra-small devices like microcontrollers. Essential for IoT and wearable applications.

Practices for systematically managing the development, deployment, and operation of LLM applications. The LLM-specific evolution of MLOps.

Advanced implementation patterns for Retrieval-Augmented Generation that go beyond basic RAG. Dramatically improve accuracy and practicality.

Automating and optimizing IT operations using AI/ML. Streamlines failure prediction and incident response.

An LLM optimization technique that skips recomputation of identical prompts to reduce costs and response times.

An evolution of RAG that retrieves and understands not just text but also images, tables, and charts to generate answers.

Cloud services with AI as their core functionality. Enables business AI adoption without specialized expertise.

AI technology that automatically generates program code from natural language instructions. Dramatically improves development efficiency.

AI technology that replicates a specific person's voice from a small audio sample. Used for narration and dubbing.

AI models that automatically generate high-quality video from text or images. Revolutionizing video production.

Safety mechanisms that control AI output and prevent harmful or inappropriate content generation.

Standardized tests for objectively comparing and evaluating AI model performance.

An inference acceleration technique that uses a small model to draft tokens and a large model to verify them. Achieves 2-3x speedup without quality loss.

A workflow design pattern where AI agents autonomously plan, execute, and reflect on tasks.

The discipline of optimally designing the context (background information) provided to LLMs. An evolution of prompt engineering.

A specialized model that converts text or images into numerical vectors that preserve semantic meaning. The foundation for search and RAG.

An AI assistant design pattern that supports human work from the side. Decision-making remains with the human.

A feature that forces LLM output into predetermined formats like JSON. Essential for application integration.

The ability of LLMs to process extremely long inputs exceeding 1 million tokens at once. Enables analysis of entire books.

An AI system architecture that combines multiple AI components rather than relying on a single model.

A technique that teaches LLMs a task by providing just a few examples in the prompt. No additional training required.

A virtuous cycle where AI service usage data accumulates and feeds back into model improvement.

The standard file format for running local LLMs. Widely adopted in llama.cpp.

A technique that hierarchically combines multiple LLMs to achieve performance exceeding any single model.

The token-based pricing structure of LLM services and optimization strategies. Essential knowledge for cost management.

Practical patterns for AI tool integration using MCP. Standardizes agent connections to external systems.

Techniques for optimizing content to be cited in AI search engine and LLM responses. The next generation of SEO.

LLMs that run on your own PC or server without cloud services. Advantageous for privacy and cost.

AI models that understand images and respond with text. Integrating visual recognition with language understanding.

A process where third parties evaluate and audit AI systems for fairness, safety, and transparency.

An approach where models use additional computational resources during inference to improve answer accuracy.

A prompting technique that branches LLM reasoning into a tree structure and explores the optimal thinking path.

A collective term for leveraging AI tools to enhance and streamline SEO strategies.

A reusable prompt blueprint with embedded variables for efficient AI utilization.

The design philosophy of AI assistants that support human work from alongside. A collaboration model where humans retain control.

A virtual model that faithfully replicates real-world objects or processes in digital space.

The technological evolution of GANs from image generation to video and 3D applications.

A general term for AI systems that support user tasks through natural language conversation.

Applied techniques for practically designing and optimizing context information passed to LLMs to maximize output quality.

A methodology for building complex AI workflows by chaining multiple prompts in sequence.

Practical techniques for coordinating and integrating multiple AI models and tools to automate complex business processes.

The process of assigning labels (ground truth data) to data for supervised AI model training.

The process of adding semantic information and labels to data. A foundational step that determines AI training data quality.

An inference method that processes large volumes of data through an AI model at once. Superior in cost efficiency and throughput.

An inference method that delivers AI model output in real-time as it is generated. Improves user experience.

Techniques for efficiently utilizing an LLM's context window to maximize output quality.

An automated workflow covering the entire AI process from data collection through model training, inference, and deployment.

A curated collection of prompts organized by use case and task. Contributes to standardizing AI usage across teams.

A technology that enables AI to process multiple modalities — text, images, audio, and video — in an integrated manner.

Educational initiatives for correctly understanding AI and using it safely and effectively.

A system that intelligently routes user queries to the most suitable AI model based on task type, complexity, cost, and latency requirements.

Open protocol proposed by Anthropic to connect LLMs and AI agents to external tools and data sources via a unified interface. The de facto standard in 2026.

AI systems where an LLM autonomously selects tools, plans, executes, evaluates, and iterates to complete complex tasks. The defining concept of 2026.

AI capability to operate a PC directly via screenshots, mouse, and keyboard — pioneered by Anthropic.

A development style of building apps by describing intent in natural language, without reading the code in detail. Coined by Andrej Karpathy in 2025.

Technique where AI models use extra compute at inference time to deepen reasoning and produce better answers. Popularized by OpenAI o1 and Claude Extended Thinking.

Specialized LLM that performs step-by-step reasoning at inference time to excel on math, science, and coding benchmarks. Now standard at the frontier.