AI Glossary

Operational speed metrics: latency expresses the time to the first response, throughput the number of tokens processed per unit of time; they affect usability and cost.

License terms determining how a model may be used, distributed, and further modified, including any restrictions on commercial use.

A numerical value inside a model that is set during training; the number of parameters (often in the billions) tends to be a rough indicator of the model's size and capacity.

A technique that automatically routes each query to the most suitable model based on its complexity, cost, or required capabilities.

The process by which a model learns to recognize patterns and relationships in data.

A set of learned parameters that determines a model's behavior. Making them available distinguishes open-weight models from closed ones.

The rights of an author connected with their personality, such as the right to attribution and to the integrity of the work, which cannot be transferred to another person.

The author's moral rights (e.g., the right to attribution of authorship and to the integrity of the work) and their application in connection with content created or modified by AI.

A collective term for consolidated proceedings bringing together multiple lawsuits against providers of large language models, in which fundamental rulings on fair use are expected.

An arrangement in which multiple AI agents communicate with one another and share tasks in order to jointly solve more complex assignments.

A system composed of multiple AI agents that cooperate or compete with one another to jointly accomplish tasks or solve complex problems.

A security method requiring multiple independent factors to verify identity, for example a password and a one-time code.

A model capable of processing and combining multiple types of inputs and outputs, such as text, image, audio, and video.