Notes by Lex Toumbourou

Zero-shot learning refers to the ability of a model to perform tasks or make predictions on classes or categories it has never explicitly seen during training. Instead of relying on labeled examples, the model leverages its understanding of relationships, features, or instructions to generalize to new tasks.

Zero-Shot Prompting

In zero-shot prompting for language models, the model generates accurate responses based purely on an instruction or context without requiring specific examples, showcasing its capacity to generalize knowledge to unfamiliar situations. This approach can be especially powerful for handling diverse and unseen challenges.