Why RAG won’t solve generative AI’s hallucination problem

Retrieval Augmented Generation (RAG) can address the limitations of generative AI models. RAG is a technique where a model retrieves possibly related documents to a question and generates answers based on the additional context. While RAG can help improve the factuality of the generated answers, it does not prevent models from hallucinating and has limitations in reasoning-intensive tasks like coding and math. RAG is expensive in hardware and requires ongoing efforts to improve the models’ use of retrieved documents.

Source: Techcrunch