Running summarizing tools on a new article

Abstract: The fast-paced progressions in Natural Language Processing (NLP) have led to extensive use in obtaining state-ofthe-art responses by leveraging the parametric memory of pre-trained Large Language Models (LLM). The generated responses of LLMs still have limitations where the data generated is inconsistent and often factually incorrect, giving rise to hallucinations. Even with the power of comprehensible generation, hallucinations are a major drawback for LLM models. Retrieval Augmented Generation (RAG) overcomes this trade-off by generating highly accurate responses that are both creative and factually correct using knowledge bases. It combines non-parametric memory and pretrained parametric memory of LLMs for response generation. It uses both the retrieval and generation models and augments the response to be precise and reliable. This article is a study of the working of RAG models using a simplified approach for QA tasks. The RAG model developed for this study is evaluated specifically on QA tasks and compared with state-of-the-art techniques (Bidirectional Encoder Representations from Transformers (BERT) and Bidirectional Auto-Regressive Transformers (BART)) and the obtained results show that the RAG model outperforms both (BERT and BART) in terms of factual accuracy and semantics. The study finds a suitable RAG attachment with existing LLMs that provides better consistency and factually more correct responses.