Both Akari et al. 2023 and Matsumoto et al. 2024 warn that naive retrieval creates its own reliability problems—unhelpful output from indiscriminately injected passages, and difficulty selecting appropriate knowledge from large, noisy sources—motivating their respective self-reflection and graph-of-thoughts mechanisms.
E2 warns that indiscriminately retrieving and incorporating fixed passages can lead to unhelpful response generation and introduces Self-RAG's self-reflection, while E17 warns of RAG's difficulty selecting appropriate knowledge from a large, noisy source and proposes KRAGEN's graph-of-thoughts to overcome these limitations.
Written by Kimi K3 via Ollama Cloud · checked by GLM-5.3 via Ollama Cloud · 1 Oct, 04:39
Source chain
AEvery quote below was checked, without a model, to appear verbatim in its source.
- 01
“indiscriminately retrieving and incorporating a fixed number of retrieved passages, regardless of whether retrieval is necessary, or passages are relevant, diminishes LM versatility or can lead to unhelpful response generation”
Full-text passage · p. 1 · evidence E2
Preprint. SELF -RAG: L EARNING TO RETRIEVE , G ENERATE , AND CRITIQUE THROUGH SELF -R EFLECTION Akari Asai†, Zeqiu Wu†, Yizhong Wang†§, Avirup Sil‡, Hannaneh Hajishirzi†§ †University of Washington §Allen Institute for AI ‡IBM Research AI {akari,zeqiuwu,yizhongw,hannaneh}@cs.washington.edu, avi@us.ibm.com ABSTRACT Despite their remarkable capabilities, large language models (LLMs) often produce responses containing factual inaccuracies due to their sole reliance on the parametric knowledge they encapsulate. Retrieval-Augmented Generation (RAG), an ad hoc approach that augments LMs with retrieval of relevant knowledge, decreases such issues. However, indiscriminately retrieving and incorporating a fixed number of retrieved passages, regardless of whether retrieval is necessary, or passages are relevant, diminishes LM versatility or can lead to unhelpful response generation. We introduce a new framework called Self-Reflective Retrieval-Augmented Generation (SELF -R AG) that enhances an LM’s quality and factuality through retrieval and self-reflection.
Passage read from arxiv.org, which may be a preprint rather than the published version.
- 02
“the difficulty of selecting the most appropriate knowledge points from a large and noisy knowledge source through methods like vector similarity search”
Full-text passage · no page number · evidence E17
Passage read from www.ebi.ac.uk, which may be a preprint rather than the published version.