A new term?
Start here.

A system that can use a model to choose actions and tools toward a goal.

The broad field of building systems for tasks associated with intelligence.

The model-specific capacity for the information handled in a request; check input and output limits.

Machine learning with neural networks containing multiple layers.

A numerical representation used to compare or process items such as text.

A structured way to test behavior against defined examples and criteria.

Further training a pretrained model on examples to adapt its behavior.

AI methods that produce content such as text, images or audio.

Connecting a generated response to supplied evidence or a defined context.

Generated content that is false or unsupported by the available evidence.

Combining retrieval signals such as keyword and vector similarity.

Using a trained model to produce outputs for an input.

The time an operation takes to return a useful result.

Giving a component only the permissions needed for its task.

A large language model trained to process and generate language.

Fitting a model from data so it can perform a task on new inputs.

Model Context Protocol: a standard connection between AI applications and external capabilities.

Using more than one kind of information, such as text and images.

Learning patterns that fit training examples but generalize poorly.

Instructions and context supplied to guide a model’s response.

Untrusted content attempting to redirect an AI system’s behavior.

Retrieval-augmented generation: retrieve relevant evidence and supply it for generation.

A unit a model processes, often a word fragment or punctuation.

Adjusting a model using examples and a learning objective.

Finding representations that are nearby under a similarity measure.