Responsible AI

A security risk in which hidden or malicious text embedded in a document or webpage tricks an AI system into ignoring its real instructions and following the attacker's instead.

A technique where an AI model first retrieves relevant information from a specific set of documents before generating an answer, instead of relying only on what it was originally trained on.

The number of examples given to a model within a prompt to guide its response; "few-shot" means the model is shown only a handful of examples before producing new output.

A setting on a language model that controls the randomness of its output; a higher temperature produces more varied, less predictable responses, and a lower one produces more consistent, focused ones.

An individual unit of text, such as a word or piece of punctuation, that an AI language model processes; tokenization is the act of breaking text into these units.

The body of data used to train an AI model, shaping what patterns and language it learns to reproduce.