The AI Glossary
Master the language of the future. From AGI to Zero-Shot Learning, we've decoded the most essential terms in Artificial Intelligence.
AI Search
AI search lets users look up information through natural language queries, not rigid keywords, making search results feel more intuitive and conversational.
Adapters
Adapters are a powerful way to adapt pre-trained AI models for new tasks without full retraining. They save time, cost, and resources by reusing existing models across NLP.
Agentic AI
Agentic AI describes systems built to act autonomously, handling complex goals and workflows efficiently with minimal human guidance or ongoing supervision.
Annotation
Data annotation means tagging or labeling raw information with context so machine learning systems can interpret, learn, and improve their understanding of data.
AGI
Artificial General Intelligence (AGI) is an advanced AI capable of human-like cognition, enabling it to learn, and create solutions across diverse tasks rather than performing specific functions.
Artificial Intelligence
AI is the simulation of human intelligence in machines designed to think and learn like people. Example: A self-driving car that uses AI to navigate roads and make decisions independently.
Associative Memory
Associative memory is a system’s capability to store, link, and retrieve connected information, allowing efficient recognition and use of relevant data to enhance decision-making.
ASR
Automatic speech recognition (ASR) is a technology that converts spoken words into written text, enabling real-time voice-to-text communication and improving accessibility across digital platforms.
AI Assistant
An AI assistant is a conversational tool powered by large language models that helps users handle diverse tasks and make informed decisions across areas.
Automation
Automation is the use of technology to complete tasks efficiently with little human involvement, streamlining processes and improving accuracy across operations.
AI Plugin
AI plugins are dedicated software modules that let AI systems connect and interact seamlessly with external tools, platforms, or services to extend their functionality.
Attention Mechanism
Attention mechanisms are AI methods that let models concentrate on the most important parts of input data, improving how they process and interpret information efficiently.
Benchmarking
Benchmarking involves assessing and comparing products or systems through standardized tests to measure their performance, efficiency, and overall capabilities.
Big Data
Big data represents the massive amounts of structured and unstructured information created every day from diverse sources such as social media, sensors, online transactions, and digital activities.
Chatbot
A simple interface where users can ask questions and get answers. Depending on the backend, it can range from preset replies to a dynamic AI that resolves issues.
Controllability
Controllability means understanding, managing, and guiding an AI system’s decisions to maintain accuracy, ensure safety and prevent unintended or harmful outcomes.
ChatGPT
A chat interface powered by GPT‑3.5, a large language model from OpenAI trained on vast internet text and fine‑tuned for tasks like translation, summarization, and Q&A.
Collective Learning
Collective learning is an AI training method that combines the strengths, data, and expertise of multiple models to create more capable, adaptive, and resilient intelligence systems.
CLLM
Large language models are costly due to their vast size and complexity, requiring heavy compute power, storage, and resources for training and usage.
Conversational AI
A specialized branch of AI focused on creating systems that understand and generate human language, enabling conversations like chatbots that handle customer interactions smoothly.
Computer Vision
Computer vision enables machines to analyze, interpret, and understand visual data from their surroundings, allowing them to recognize objects and make intelligent decisions.
Data Augmentation
Data Augmentation expands the training dataset by creating altered versions of existing data. It applies tweaks like flipping, resizing to boost variety and reduce overfitting.
Deep Learning
A branch of machine learning that employs multi-layered neural networks to learn complex patterns. Example: A deep learning model identifies objects in images.
Deterministic Model
A deterministic model relies on fixed rules and defined conditions, ensuring a predictable outcome through clear cause-and-effect relationships and consistent logical reasoning.
Discriminative Model
Discriminative models are algorithms that learn and define the decision boundary separating various classes or categories within a given dataset to enable accurate and efficient classification.
Extraction
Extraction allows generative models to examine vast datasets and identify meaningful patterns, and precise information relevant to a given goal or context.
Explainability
Explainability means using methods that help humans clearly interpret and understand how an AI model makes its decisions and predictions across different scenarios.
Enterprise AI
Enterprise AI is the purposeful adoption of artificial intelligence across an organization to optimize operations, strengthen decision-making, and drive higher efficiency in business.
Extensibility
Extensibility in AI means enabling systems to grow their abilities across new domains, tasks, or datasets without complete retraining or significant modifications.
Fine Tuning
Fine-tuning uses a pre-trained model and adapts it for a specific task with a smaller dataset. A model trained on intersections can learn to detect red-light violations.
Few-Shot Learning
Few-shot learning is a machine learning technique that enables models to understand new concepts from only a handful of labeled samples, fewer for each category.
Foundation Model
Foundation models are versatile AI systems that include large language, computer vision, and reinforcement learning models. They’re termed “foundation”.
GPT-3
GPT-3 is the third model in the GPT-n series, built with 175 billion parameters that help it make predictions. ChatGPT runs on GPT-3.5, an advanced version designed for smoother, smarter responses.
Generation
Generation refers to how a generative model produces completely new and original content—like text, images, audio, or video—entirely from scratch using learned data patterns.
Grounding
Grounding connects AI systems to real-world knowledge and data, helping them understand context better, interpret user input accurately.
GPT-4
GPT-4 marks a major leap in OpenAI’s deep learning journey, representing a milestone in model scaling. It’s the first large multimodal model that processes both images and text.
Generative AI
Generative AI models produce fresh content by identifying patterns in training data. They can craft an original short story after studying numerous published ones.
GPT
Generative pre-trained transformers (GPT) are advanced neural models trained on massive datasets using unsupervised learning to produce human-like text.
GANs
GANs are advanced neural networks that create realistic, never-before-seen data by learning patterns from existing training datasets and mimicking them.
Hallucination
Hallucination happens when an AI, especially in language tasks, produces irrelevant or incorrect outputs due to unclear context, overreliance on training data, or limited subject understanding.
Instruction-tuning
Instruction-tuning adapts a pre-trained model to handle specific tasks by supplying clear guidelines or directives that define how the model should perform, interpret input, and generate responses.
Intelligence Amplification
Intelligence amplification means enhancing human abilities by blending AI systems with conventional tools to create a powerful, cooperative form of capability expansion.
Intelligence Augmentation
Intelligence augmentation means enhancing human potential by merging artificial intelligence with traditional tools to create a more capable, adaptive, system for better decision-making.
Interpretability
Interpretability describes how clearly an AI model’s design, logic, and decisions can be understood or explained based on its internal structure and behavior.
K-Shot Learning
K-shot learning is a machine learning method where a model learns to recognize each class using only k labeled examples, enabling efficient learning from limited data.
Knowledge Generation
Knowledge generation uses large datasets to train models that analyze information, identify patterns, and produce fresh insights from existing data.
Knowledge Graph
Knowledge graphs are structured networks that link related data points, enabling AI systems to analyze, navigate, and interpret complex information through connected relationships.
Latency
Latency is the time gap between when an AI model gets a user’s input and when it produces the related output, reflecting how quickly the system processes.
Low-code
Low-code is a visual method for building software that accelerates application creation, improves efficiency, and reduces development time with minimal manual coding.
Large Language Model
A deep learning model trained on massive text data for natural language understanding and generation. Popular LLMs include BERT, PaLM, and GPT series, each differing in training data.
Machine Learning
A branch of AI focused on creating algorithms and models that help machines learn from experience and improve over time. Example: A machine learning model that predicts customer churn.
Model Chaining
Model chaining is a data science approach where multiple machine learning models are connected in sequence, allowing outputs from one model to serve as inputs for the next.
Multi-hop Reasoning
Multi-hop in NLP and reading comprehension means an AI connects multiple information pieces from texts or sources to find answers, instead of relying on one direct passage.
MLM
Multimodal Language Models are advanced deep learning systems trained on massive datasets that combine text with images, audio, or other data types to understand.
N-Shot Learning
Zero, single, and few-shot learning all follow one principle—training models with minimal data to classify new inputs. Each “shot” is one example.
NLA
Natural language ambiguity happens when a word, phrase, or sentence carries multiple interpretations, causing difficulty for both people and AI systems to understand.
NLG
A specialized branch of AI focused on generating human-like written or spoken language that sounds natural, coherent, and contextually relevant across different communication tasks.
NLP
A branch of AI focused on teaching computers to interpret and organize large amounts of language data, turning unstructured text into a clear, machine-readable format.
NLU
A branch of NLP that interprets text to uncover its semantic meaning, helping machines grasp context, sentiment, intent, and the deeper nuances of written language.
No-code
No-code is a method for building and managing applications without writing code or learning programming languages, making app creation faster and easier for anyone to use.
Neural Network
A machine learning model inspired by the human brain, built from layers of connected nodes or neurons. Example: A neural network trained to accurately identify numbers.
OpenAI
OpenAI, the organization behind ChatGPT, is a research company focused on creating and advancing safe AI. Its GPT-3 model capable for natural language processing.
Optimization
Tuning a model’s parameters to reduce a skip function that quantifies the gap between predictions and values, using gradient descent to optimize neural network performance.
Overfitting
Overfitting happens when a model learns the training data too precisely, capturing noise instead of true patterns. It performs well on training data but fails to generalize.
PEFT
Parameter-Efficient Fine-Tuning (PEFT) improves large AI models by updating only select parameters instead of retraining the entire model, saving time, energy, and computational resources.
Prompting
Prompting is the practice of writing clear, detailed instructions that help AI tools understand tasks and produce accurate, high-quality results aligned with your goals.
Pre-training
Training a model on vast data before refining it for a focused task. Example: Pre-training a language model like ChatGPT on massive text data, then fine-tuning it for tasks like translation.
Prompt Engineering
Prompt engineering focuses on crafting inputs that shape meaningful LLM outputs. These models blend layered algorithms with limited control, guided by templates and wizards for precision.
Probabilistic Model
A probabilistic AI model relies on probability and likelihood to make predictions or decisions, evaluating multiple possible outcomes based on data patterns and uncertainty levels.
Quantum Computing
Quantum computing is an emerging technology that can vastly boost processing power, offering immense potential to elevate the performance and advanced AI systems.
Reasoning
AI reasoning enables machines to analyze data, solve problems, and generate insights by processing information logically, helping them make accurate.
Recursive Prompting
Recursive prompting helps guide AI models like GPT‑4 toward better results. It works by using a sequence of prompts that build on each other, refining context.
Reinforcement Learning
Reinforcement learning is a machine learning approach where a model learns through interactions and feedback in the form of rewards or penalties.
Responsible AI
Responsible AI means designing, deploying, and using AI systems that benefit employees, businesses, and society while prioritizing ethics, trust, and confident.
RAG
Retrieval augmented generation (RAG) boosts large language models (LLMs) by linking them with external data sources to deliver richer, more informed, and context-aware responses.
Sequence Modeling
A branch of NLP that deals with understanding and modeling sequential data like text, speech, or time series. Example: A sequence model predicting the next word or generating fluent sentences.
Speech-to-text
The process of turning spoken language into written text, capturing every word accurately and converting speech into readable, well-structured form for clear understanding.
Stable Diffusion
Stable Diffusion is an AI model that applies deep learning to convert text prompts into visually detailed images, enabling creative image generation through natural language descriptions.
Stacking
Stacking is an AI technique that merges multiple models to boost performance. By combining their strengths, it offsets individual weaknesses, and give reliable results.
Steerability
AI steerability means guiding an AI’s actions and outputs toward human goals. It involves building models that follow user intent, avoid errors, and improve feedback.
Strong AI
Strong AI describes machines with human-level cognitive abilities, capable of reasoning, learning, and adapting across diverse tasks similar to the way humans think and understand.
Structured Data
Structured data is information arranged and tagged in a consistent, standardized way, making it easier for systems to read, interpret, and organize efficiently.
Summarization
Summarization enables generative models to process extensive text and create brief, focused versions that retain the essential meaning and highlight the most important information.
Supervised Learning
A machine learning approach where a model learns from labeled data to predict outcomes for unseen inputs. Example: supervised learning used to recognize handwritten digits from labeled samples.
Stochastic Parrot
Stochastic parrots are AI models that rely on statistical patterns to produce human-like language, but they do so without genuine understanding or awareness of the meaning behind.
Text-to-speech
Text-to-speech (TTS) is a technology that transforms written text into audible speech, enabling users to listen to content read aloud through realistic, computer-generated voices.
Tokenization
Tokenization is the process of splitting text into smaller units like words or subwords for language model. For instance, “I am ChatGPT” becomes “I,” “am,” “Chat,” “G,” and “PT.”
Transformer
A neural network architecture built to handle sequential data like text. Example: Transformers, as used in ChatGPT, power many natural language understanding.
Unstructured Data
Unstructured data refers to information that lacks a predefined format or organization, making it harder to gather, process, and analyze effectively.
Unsupervised Learning
Unsupervised learning trains a model on unlabeled data to identify hidden patterns or features. Example: it can group images of handwritten.
Voice Processing
Voice processing in AI involves converting spoken language into text and then using that text to generate speech, creating a complete speech-to-text and text-to-speech workflow.
Voice Synthesis
Voice synthesis uses artificial intelligence to produce natural, expressive speech by learning from text and audio data, enabling computers to sound more realistic.
Whisper
OpenAI’s Whisper is an advanced AI model designed for automatic speech recognition, enabling accurate transcription of spoken language into written text across various contexts.
Weak AI
Weak AI describes specialized systems built to perform specific tasks within narrow contexts, but lacking broad intelligence or the ability to adapt beyond their scope.
Weak-to-Strong
Weak-to-strong generalization trains AI by using simpler models to guide and shape stronger ones, helping advanced systems generalize better and move beyond the limits of their training data.
X-risk
X-risk, or existential risk, describes the chance that advanced artificial intelligence could endanger humanity’s survival due to unforeseen consequences or goals that conflict with human values.
Zero-Shot Learning
Zero-shot learning enables a model to identify and categorize unseen concepts without using labeled examples, allowing it to generalize knowledge.
Zero-to-One Problem
The zero-to-one problem describes the challenge of creating the first solution to a complex issue, a stage that’s often far tougher than making improvements.