Seven Failure Points When Engineering a Retrieval Augmented Generation System
Семь критических точек отказа при разработке системы генерации с дополнением извлечённой информацией
2024-04-14
SCID: 54.1/tadw4umf
Discuss with AI
hallucinated responsesinformation retrievallarge language modelsretrieval-augmented generationsemantic search
Figures from the paper
Abstract (AI)
Software engineers are increasingly adding semantic search capabilities to applications using a strategy known as Retrieval Augmented Generation (RAG). A RAG system involves finding documents that semantically match a query and then passing the documents to a large language model (LLM) such as ChatGPT to extract the right answer using an LLM. RAG systems aim to: a) reduce the problem of hallucinated responses from LLMs, b) link sources/references to generated responses, and c) remove the need for annotating documents with meta-data. However, RAG systems suffer from limitations inherent to information retrieval systems and from reliance on LLMs. In this paper, we present an experience report on the failure points of RAG systems from three case studies from separate domains: research, education, and biomedical. We share the lessons learned and present 7 failure points to consider when designing a RAG system. The two key takeaways arising from our work are: 1) validation of a RAG system is only feasible during operation, and 2) the robustness of a RAG system evolves rather than designed in at the start. We conclude with a list of potential research directions on RAG systems for the software engineering community.
Key Findings
1
RAG aims to reduce hallucinations, provide source references, and avoid requiring document metadata annotation, but these goals are not guaranteed.
2
RAG robustness evolves over time and must be continuously developed rather than fully designed at system launch.
3
RAG systems inherit limitations from both information retrieval components and their reliance on large language models.
4
The paper identifies seven failure points in Retrieval Augmented Generation systems based on three case studies spanning research, education, and biomedical domains.
5
Validation of a RAG system is feasible only during real-world operation rather than exclusively during initial development.
Research Object
Retrieval Augmented Generation (RAG) systems
Research Subject
Failure points, operational validation, and evolving robustness of RAG systems across research, education, and biomedical domains
Publication Details
Publication Date
2024-04-14
Journal
Publisher
ISSN
Cited by
155
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai5
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks2020
Lost in the Middle: How Language Models Use Long Contexts2024
Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering2021
Benchmarking Large Language Models in Retrieval-Augmented Generation2024
Improving the Domain Adaptation of Retrieval Augmented Generation (RAG) Models for Open Domain Question Answering2023