METHODS OF IMPLEMENTING RETRIEVAL-AUGMENTED GENERATION IN COMBINATION WITH MODERN LARGE LANGUAGE MODELS

DOI: 10.31673/2786-8362.2025.012843

Abstract

The article presents a comprehensive study of stateof-the-art Retrieval-Augmented Generation (RAG) methods integrated with large language models such as
GPT-4 and open GPT-4-equivalent models (GPT-4o). We analyze experimental results from 2023-2025
(including open sources like Arxiv, HuggingFace, PapersWithCode) that demonstrate the advantages of
new approaches: RAG 2.0 with joint end-to-end training of components, adaptive dynamic retrieval,
generative prompts for retrievers, self-refining pipelines with feedback, and modular architectures for APIbased models. We explain how these innovations overcome the shortcomings of traditional RAG by
reducing hallucination rates and improving answer faithfulness. The mechanisms of the models are
described, along with system workflow diagrams and a comparative table of metrics (Hallucination Rate,
Answer Faithfulness, Retrieval Precision). We substantiate the potential of advanced RAG in information
systems, chatbots, and analytics platforms. A clear vision for further development is formulated –
integrating RAG with continuously updated knowledge bases and real-time search APIs to ensure up-todate responses.
Keywords: large language models, retrieval-augmented generation, RAG 2.0, dynamic retriever, selfreflection, hallucinations, factual accuracy

References
1. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks / Lewis M. et al.
NeurIPS, 2020. URL: https://medium.com/@adnanmasood/contextually-enriched-knowledgeenhanced-and-externally-grounded-retrieval-models-for-fun.
2. RAG 101: Demystifying Retrieval-Augmented Generation Pipelines / Wolff H. NVIDIA
Technical Blog, 2023. URL: https://developer.nvidia.com/blog/rag-101-demystifying-retrievalaugmented-generation-pipelines.
3. Self-RAG: Self-Reflective Retrieval-Augmented Generation / Asai A. et al.
arXiv:2310.11511, 2023. https://doi.org/10.48550/arXiv.2310.11511.
4. Contextual AI. Introducing RAG 2.0 / Kiela D. Medium blog post, 2024. URL:
https://medium.com/@jackdrummond_16745/introducing-rag-2-0-revolutionising-ai-and-llmdevelopment-71a1f85cc0dd.
5. Active Retrieval-Augmented Generation / Mallen E. et al. arXiv: 2409.11136, 2023. URL:
https://doi.org/10.48550/arXiv.2409.11136.
6. Contextually Enriched, Knowledge-Enhanced, and Externally Grounded Retrieval Models /
Masood A. Medium, 2025.
7. Promptriever: Promptable Retrieval Models / Weller O. et al. EMNLP, 2024 (preprint). URL:
https://medium.com/@techsachin/promptriever-first-zero-shot-promptable-instruction-trainedretriever-model-72e9f2eecbb2.
8. Speculative RAG: Enhancing RAG through Drafting / Wang Z. et al. arXiv:2407.08223, 2024.
URL: https://doi.org/10.48550/arXiv. 2407.08223.