AI terms, made clear.
Short explanations for the concepts you meet in AI reporting. Tap an underlined term in a briefing for a quick definition, or explore all 20 here.
AI agent
A system that chooses actions and can use tools toward a goal.
Primary referenceArtificial intelligence
Computer systems performing tasks such as language generation, perception or planning.
Primary referenceBenchmark
A defined evaluation; its score depends on tasks, conditions and measurement.
Primary referenceContext window
The token capacity available to a model for a given interaction.
Primary referenceDistillation
Training a student model from a teacher model’s outputs or behavior, often to reduce size or cost.
Primary referenceEmbedding
A numeric representation that helps models compare or process information.
Primary referenceFine-tuning
Further training an existing model with examples for a particular purpose.
Primary referenceGenerative AI
Models that produce content such as text, images, audio or code.
Primary referenceHallucination
Generated content that sounds plausible but is incorrect or unsupported.
Primary referenceInference
Running a trained model to produce an output from an input.
Primary referenceLarge language model
A large neural model trained on language to process and generate text.
Primary referenceMachine learning
Methods that learn patterns from examples rather than only explicit rules.
Primary referenceMultimodal
Working with multiple input or output types, such as text and images.
Primary referenceOpen weights
Released model parameters. Availability of weights alone does not make a model open source; check its license and accompanying materials.
Primary referenceParameters
Adjustable numbers in a model, usually learned during training.
Primary referenceQuantization
Using lower-precision numbers to reduce model memory needs, sometimes at a cost to accuracy.
Primary referenceRetrieval-augmented generation
Providing a model with relevant retrieved material when it generates a response. Retrieval does not guarantee correctness.
Primary referenceToken
A unit a model processes, often a word fragment or punctuation.
Primary referenceTraining
Adjusting a model using data to improve a defined objective.
Primary referenceTransformer
A neural-network architecture using attention to relate parts of an input.
Primary referencePut the vocabulary into context.
Explore the reading guides