AI Glossary — Every artificial intelligence term explained

An autonomous software entity that perceives its environment, makes decisions, and takes actions to achieve specific goals without continuous human…

A finite set of well-defined instructions or rules designed to solve a specific problem or perform a computation, forming the backbone of all AI…

Systematic errors in AI outputs that arise from biased training data or flawed model design, potentially leading to unfair or discriminatory outcomes…

Application Programming Interface — a set of protocols enabling software applications to communicate. In AI, APIs are the primary way to integrate…

Hypothetical AI capable of understanding and performing any intellectual task a human can. AGI remains a theoretical goal. Current systems, however…

The simulation of human cognitive processes by computer systems, encompassing learning, reasoning, problem-solving, perception, and language…

A neural network component that learns to focus on the most relevant parts of input data, enabling models like Transformers to capture long-range…

A neural network trained to compress input data into a compact latent representation and then reconstruct it, used for dimensionality reduction and…

A standardized test or dataset used to evaluate and compare the performance of different AI models, algorithms, or systems. Benchmarks like MMLU…

Extremely large and complex datasets that exceed the capacity of traditional tools, characterized by high volume, velocity, and variety. Big Data is…

An AI application that simulates human conversation through text or voice, powered by natural language processing and often backed by large language…

A supervised learning task where the model assigns input data to predefined categories or classes, such as spam detection, image recognition, or…

Using remote internet-hosted servers for data storage, processing, and computation. Cloud computing (AWS, Google Cloud, Azure) is the essential…

An unsupervised learning technique that groups similar data points together without predefined labels, revealing hidden structures in data. Used for…

A specialized neural network architecture designed for processing grid-like data such as images, using convolutional layers to automatically detect…

A field of AI enabling machines to interpret and understand visual information from images, videos, and the physical world. Used in facial…

An AI field enabling machines to analyze and understand visual content in images and videos. Used in facial recognition, autonomous vehicles, and…

The maximum amount of text (in tokens) a language model can process in a single interaction. Larger context windows allow processing more text at…

A large, structured collection of text documents used for training and evaluating natural language processing models. The quality and diversity of…

NVIDIA's parallel computing platform and API that enables developers to harness GPU power for accelerating AI computations and deep learning…

Techniques for artificially expanding training datasets by applying transformations such as rotation, flipping, or noise injection to existing data…

A structured collection of data used for training, validating, or testing machine learning models, typically organized as rows of samples and columns…

A subset of machine learning based on artificial neural networks with many layers, capable of learning complex patterns from large amounts of data…

A generative model that creates data by learning to reverse a gradual noise-adding process. Diffusion models (Stable Diffusion, DALL·E 3, Midjourney)…

A model compression technique where a smaller student model is trained to replicate the behavior of a larger teacher model, preserving performance…

A dense vector representation of data (words, images, or other entities) in a continuous space, capturing semantic relationships and enabling…

One complete pass through the entire training dataset during model learning. The number of epochs is a crucial hyperparameter: too few leads to…

The study and application of moral principles to the design, development, and deployment of AI systems, addressing fairness, transparency…

A machine learning paradigm where models learn to perform tasks with only a handful of labeled examples, mimicking human ability to generalize…

The process of taking a pre-trained model and further training it on a smaller, task-specific dataset to adapt it to a particular domain or…

A large-scale AI model pre-trained on broad data that can be adapted to many downstream tasks, such as GPT-4, Claude, Llama, and Gemini. Foundation…

A development environment providing tools, libraries, and conventions for building AI applications. Major frameworks include PyTorch, TensorFlow…

A framework of two competing neural networks — a generator and a discriminator — that train together to produce realistic synthetic data. GANs were…

AI systems capable of creating new content including text, images, code, music, and video. Generative AI is the major technological revolution of…

Graphics Processing Unit — a specialized processor designed for parallel computation, essential for training and running deep learning models. NVIDIA…

An iterative optimization algorithm that minimizes the loss function by updating model parameters in the direction of the steepest descent of the…

A technique anchoring AI-generated responses in verifiable facts and reliable sources. Grounding is essential for reducing LLM hallucinations by…

The phenomenon where AI models generate plausible-sounding but factually incorrect or fabricated information. Hallucinations are a major challenge…

A leading open-source platform and community for sharing, discovering, and deploying machine learning models, datasets, and NLP tools. Hugging Face…

Configuration settings external to the model that control the training process, such as learning rate, batch size, and number of layers…

The phase where a trained model is applied to new, unseen data to generate predictions or outputs. Inference is the production phase. Optimizing…

Techniques to bypass safety restrictions and guardrails implemented in an AI model. Jailbreaks exploit system prompt vulnerabilities to obtain…

A structured representation of real-world entities and their relationships, enabling machines to reason about interconnected information and power…

A popular open-source framework for developing applications based on large language models. LangChain provides tools for prompt chaining, RAG…

A statistical system modeling word sequence probabilities to predict the next word. Language models are the foundation of chatbots, machine…

A neural network with billions of parameters trained on vast text corpora, capable of understanding and generating human language with remarkable…

A compressed, abstract mathematical representation of data within a generative model. The latent space captures essential data characteristics in a…

Low-Rank Adaptation — a parameter-efficient fine-tuning technique that adds trainable low-rank matrices to frozen model weights, significantly…

A branch of AI where systems learn patterns from data and improve their performance over time without being explicitly programmed for each task…

A branch of AI enabling systems to automatically learn from data without explicit programming. ML encompasses supervised, unsupervised, and…

An architecture where different specialized sub-networks (experts) are selectively activated based on the task. MoE enables highly performant models…

An AI model capable of processing and generating multiple data types simultaneously: text, image, audio, video. Multimodal models like GPT-4V…

A field of AI focused on enabling computers to understand, interpret, and generate human language in both written and spoken forms. NLP powers…

A computing architecture inspired by biological neurons, consisting of interconnected layers of nodes that process data and learn patterns through…

Tools and platforms enabling application creation without or with very little manual coding. No-code/low-code platforms (Bubble, n8n, Make)…

An open-source tool for easily running large language models locally on your computer. Ollama simplifies downloading and running models like Llama…

Software whose source code is freely accessible, modifiable, and redistributable by anyone. Open source is a pillar of the AI ecosystem, with…

An AI model whose source code and weights are publicly accessible and reusable. Open source models (Llama, Mistral, Phi) democratize AI access and…

A condition where a model memorizes training data too closely, including its noise, leading to poor generalization and weak performance on unseen…

An internal numerical value learned and adjusted by the model during training. Parameter count is often used as a proxy for model capacity.

A sequence of automated processing steps where each step’s output feeds the next. In AI, pipelines orchestrate data preprocessing, training…

The input text or instruction given to a generative AI model to guide its output, serving as the primary interface between users and language models…

The practice of crafting and optimizing input prompts to elicit desired responses from AI models. Prompt engineering has become a key skill for…

A model compression technique that reduces numerical precision of weights and activations, enabling faster inference and lower memory usage…

A question, command, or request submitted to an AI system or database to obtain results. Queries are the interface between users and information…

A technique that enhances language model outputs by retrieving relevant information from external knowledge sources before generating a response. RAG…

An algorithm analyzing preferences and behavior to suggest relevant content to the user. Recommendation systems (Netflix, Spotify, Amazon) are among…

A neural network architecture with feedback connections that maintains a hidden state, designed to process sequential data. RNNs have been largely…

A learning paradigm where an agent learns to make optimal decisions by interacting with an environment and receiving reward or penalty signals. The…

Reinforcement Learning from Human Feedback — a technique for aligning language models with human preferences by training a reward model from human…

A distribution model where software is accessible via the internet on subscription without local installation. Most AI tools (ChatGPT, Midjourney…

Empirical rules predicting how model performance improves with size and data. Scaling laws have guided LLM development by showing that performance =…

A search method that understands meaning and intent behind a query, beyond simple keywords. Semantic search uses embeddings to find conceptually…

A parameter controlling the randomness of AI model outputs. Low temperature (0-0.3) produces deterministic, precise responses. High temperature…

Technology for automatic conversion of written text into natural, expressive speech. Modern systems (ElevenLabs, OpenAI TTS) produce voices nearly…

The basic unit of text that language models process, which can be a word, subword, or character. LLMs charge and measure capacity in tokens. In…

The iterative process of feeding data through a model and adjusting its parameters to minimize the loss function. Training a large language model…

A technique where knowledge gained from training on one task is applied to a different but related task, dramatically reducing data and compute…

A neural network architecture based on self-attention mechanisms that processes input data in parallel. Introduced by Google in 2017 (‘Attention Is…

A benchmark proposed by Alan Turing in 1950 to evaluate a machine's ability to exhibit intelligent behavior indistinguishable from that of a human…

A database specialized in storing and quickly searching embedding vectors. Vector databases (Pinecone, Weaviate, ChromaDB) are essential for RAG and…

The process of converting data (text, images) into numerical representations as vectors. Vectorization is the essential prerequisite for semantic…

Numerical values adjusted during training that determine a neural network’s behavior. Weights are the model’s ‘learned knowledge,’ stored as matrices…

An open-source speech recognition model by OpenAI, capable of transcribing and translating speech in 90+ languages. Whisper has become the de facto…

A paradigm where a model performs tasks it was never explicitly trained on, leveraging its general knowledge to handle novel classes or instructions…