What is C-RAG: Certified Generation Risks

C-RAG (Certified Generation Risks for Retrieval-Augmented Language Models) is the first framework to certify generation risks for RAG models. C-RAG provides conformal risk analysis for RAG models and certifies an upper confidence bound of generation risks, which is referred to as conformal generation risk. The framework addresses whether RAG can indeed lead to low generation risks, how to provide provable guarantees on the generation risks of RAG and vanilla LLMs, and what sufficient conditions enable RAG models to reduce generation risks. C-RAG provides theoretical guarantees on conformal generation risks for general bounded risk functions under test distribution shifts and proves that RAG achieves a lower conformal generation risk than that of a single LLM when the quality of the retrieval model and transformer is non-trivial.