Definition
A RAG implementation where the LLM is prompted to synthesize an answer from retrieved context without being provided any prior examples of the task or expected output format. It relies on the model's innate reasoning and instructions alone to process retrieved documents.
Zero-shot focuses on instruction following without examples, whereas Few-shot RAG includes 'demonstrations' of the desired Q&A behavior.
"A student taking an open-book exam for a subject they have never studied, relying entirely on the provided textbook and their general literacy."
Conceptual Overview
A RAG implementation where the LLM is prompted to synthesize an answer from retrieved context without being provided any prior examples of the task or expected output format. It relies on the model's innate reasoning and instructions alone to process retrieved documents.
Disambiguation
Zero-shot focuses on instruction following without examples, whereas Few-shot RAG includes 'demonstrations' of the desired Q&A behavior.
Visual Analog
A student taking an open-book exam for a subject they have never studied, relying entirely on the provided textbook and their general literacy.