Comparison of LLM-based benchmarks using auxiliary systems such as Knowledge Graphs (KGs), Retrieval Augmented Generation (RAG) systems and external dataset sources

This comparison includes the most representative state-of-the-art (SOTA) benchmarks up-to-date, focusing and including on each respective benchmark architecture, large language models (LLMs) aided by knowledge graph (KG), retrieval augmented generation (RAG) systems, and other external sources such as document datasets and search engines. We present the following research questions associated with this benchmark comparison:

1) RQ1: Do the diverse LLM-based benchmarks increase accuracy metrics significantly by incorporating external sources such as auxiliary knowledge graphs (KGs), retrieval augmented generation (RAG) systems, external document datasets, and search engines?
2) RQ2: Do the diverse LLM-based benchmarks decrease LLM-based hallucination metrics significantly by incorporating external sources such as auxiliary knowledge graphs (KGs), retrieval augmented generation (RAG) systems, external document datasets, and search engines?
3) RQ3: Do the diverse LLM-based benchmarks significantly increase accuracy metrics by fine-tuning the respective LLMs compared to querying pre-trained LLMs via zero-shot prompting?