The complete AI vocabulary

A standard dataset, task, or measurement used to compare the performance of AI systems.

A systematic tendency in data or model behavior that can produce skewed, unfair, or inaccurate outcomes.

An internal or generated sequence of intermediate reasoning steps used to reach an answer or decision.

A conversational interface that responds to user messages using rules, retrieval, generative AI, or a combination of methods.

The process of splitting large documents or datasets into smaller units that can be searched or processed more effectively.

A machine learning task that assigns an input to one or more predefined categories.

A field of AI that enables systems to interpret and act on information from images and video.

The maximum amount of text, images, or other tokens a model can consider during a single interaction.

A change in real-world input data over time that can reduce the reliability of a deployed model.

A branch of machine learning that uses neural networks with many layers to learn complex patterns.

A generative model that learns to create data by reversing a gradual noise-adding process.

A numerical representation that places semantically similar items close together in a multidimensional space.

A capability or pattern that appears in a complex AI system without being explicitly programmed as a rule.

A curated set of examples used to measure how well an AI system performs against expected outcomes.

Methods that make an AI system’s outputs, influences, and decision process more understandable to people.

The ability to perform a task after seeing only a small number of examples in the prompt or training data.

Additional training that adapts a pretrained model to a specific domain, style, or task.

A large model trained on broad data that can be adapted to many downstream tasks.

AI that creates new content such as text, images, audio, video, software, or structured data.

Connecting a model’s response to verified context, source material, business data, or real-world constraints.

A technical or policy control that constrains AI behavior, inputs, outputs, access, or actions.

A plausible-sounding AI output that is unsupported, inaccurate, or fabricated.

A workflow design where people review, approve, correct, or guide important AI decisions and actions.

The process of running a trained model to generate a prediction, classification, or response.

Training a model on instruction-and-response examples so it follows natural-language requests more reliably.

A structured network of entities and relationships that represents how information is connected.

A neural network trained on large amounts of language data to understand and generate text and other tokenized content.

The time between sending a request to an AI system and receiving its response or completed action.

A field of AI in which systems learn patterns from data rather than relying only on explicitly written rules.

An open protocol for connecting AI applications to tools, data sources, and reusable context through a standard interface.

A decline or change in model performance as users, data, processes, or real-world conditions evolve.

AI that can understand or generate more than one data type, such as text, images, audio, and video.

The field of AI focused on enabling computers to understand, analyze, and generate human language.

A machine learning architecture made of connected computational units that learn patterns from examples.

The ability to understand an AI system through traces, events, metrics, costs, decisions, and quality signals.

A condition where a model learns training examples too closely and performs poorly on new data.

A learned numerical value inside a model that shapes how inputs are transformed into outputs.

The instructions and context supplied to a generative model to guide its response or action.

The practice of designing instructions, context, examples, and constraints to improve model outputs.

A model optimization technique that reduces numerical precision to lower memory use and improve inference speed.

A model optimized to spend additional computation on planning, problem-solving, and multi-step tasks.

A learning method where a system improves its behavior through rewards or penalties associated with its actions.

A method that retrieves relevant information from an external source and supplies it to a model before generation.

Search that uses meaning and context rather than relying only on exact keyword matches.

A relatively compact language model designed for lower cost, faster inference, or specialized deployment environments.

A model response constrained to a predictable format such as JSON, a schema, or defined fields.

Artificially generated data used to supplement, simulate, anonymize, or rebalance real-world datasets.

A generation setting that adjusts output variability, with lower values usually producing more consistent responses.

A unit of content, often a word fragment or symbol, that a language model processes and generates.

The ability of a model or agent to select and invoke an external function, API, database, or application.

The examples used to teach a machine learning model how to recognize patterns and perform tasks.

A neural network architecture based on attention mechanisms and widely used in modern generative AI.

A condition where a model is too simple or insufficiently trained to capture useful patterns in the data.

A database optimized to store embeddings and find items based on numerical similarity.

The design of connected, rule-based or AI-assisted steps that move work from trigger to completion.

The ability to perform a task without seeing a task-specific example in the current prompt or training set.