Retrieval-Augmented Generation

The core idea of RAG is to retrieve relevant information from an external knowledge base as context to inject into the prompt before the large model generates a response.

Strategies for mitigating LLM hallucination include retrieval-augmented generation (RAG), chain-of-thought prompting, and requiring the model to output confidence scores.

Retrieval-Augmented Generation (RAG) retrieves relevant document fragments from an enterprise's private knowledge base and injects them into the prompt before generating a response, causing internal document fragments to be sent to the model provider's servers with every query.

检索增强生成（RAG）通过基于真实知识库内容回答来解决大模型的幻觉问题

For systems using RAG, retrieval quality and answer quality should be monitored separately, because poor answers may stem from poor retrieval rather than weak generation capability

RAG（检索增强生成）是一种结合信息检索与生成式大模型的架构范式，模型生成答案前先根据用户查询从外部知识库检索最相关的文档片段，再将其作为上下文注入模型

This feature uses a Retrieval-Augmented Generation (RAG) technical architecture, retrieving external documents provided by the user in real time as contextual basis when generating content.

Retrieval-Augmented Generation (RAG) reduces the probability of hallucinated content and enables traceable citations by anchoring model outputs to a verified document database.

检索增强生成（RAG）是主流幻觉抑制技术，在模型生成回答前从外部知识库检索相关文档片段作为上下文注入提示词

目前解决AI知识截止问题的方案包括RAG（检索增强生成）、在提示中粘贴最新文档内容，或使用支持更长上下文窗口的模型

The mechanism of retrieving relevant memories and injecting them into context is commonly referred to in academia as a memory-extended form of Retrieval-Augmented Generation (RAG).

Common techniques in AI-assisted decompilation include using retrieval-augmented generation (RAG) to bring in source code from similar projects as references, using compiler feedback to form closed-loop iterative corrections, and using unit tests or differential testing to verify that the logic is equivalent before and after rewriting.

Successful experience retrieval relies on a retrievable library of historical trajectories, where each trajectory is typically indexed by the task description, the sequence of executed actions, and the final result, and is recalled at retrieval time based on task semantic similarity or execution path similarity.

Fine-tuning is better suited to stable, enumerable rules; for information that changes frequently or requires real-time verification, it often needs to be combined with retrieval-augmented generation (RAG) and other methods, otherwise the model tends to memorize outdated rules or produce answers that appear compliant but are actually wrong.

Fine-tuning suits high-frequency, stable rules that need to become a conditioned reflex; RAG suits regulatory provisions that change frequently, require citations, and are easy to audit manually.

In practice, a combined approach is common, using fine-tuning to teach the model the process and format for rule-based judgments, and retrieval to ensure the timeliness and traceability of specific provisions.

"Read it and believe it" is a typical failure mode of retrieval-augmented systems, where the model writes retrieved text directly into its output as fact without distinguishing source credibility.

Having models explicitly output confidence or evidence-strength labels, introducing uncertainty-aware loss functions, and attaching verifiable source spans to statements through retrieval-augmented generation (RAG) are related research directions for handling uncertainty.

Legal AI products usually need to incorporate mechanisms such as retrieval-augmented generation (RAG), citation tracing and human review workflows to suppress hallucinations.

In the Wikipedia context, RAG reduces completely fabricated citations, but citations with "real sources, wrong interpretation" are instead more subtle and harder to investigate.