Retrieval-Augmented Generation
Tags: Tools & Platforms
Role in Accuracy and Reliability
The Guideline recommends Retrieval-Augmented Generation (RAG) to enhance factual accuracy and reduce model hallucinations. During generation, a model retrieves relevant information from external knowledge sources and integrates it into its responses, improving alignment with real-world facts.
Developer and Provider Practices
- Technology Developers: Employ RAG and knowledge bases to ensure the timeliness and accuracy of generated content during pre-training, fine-tuning, and inference workflows.
- Service Providers: Offer user-friendly fact-checking tools, including authoritative data retrieval interfaces and intelligent comparison utilities, to assist manual verification of outputs in context-specific services.
Risk Considerations
- In the usage and maintenance stage, the Guideline cautions that generative AI may access data used during the RAG process, creating potential data exposure risks that require careful governance, permissions, and transparency.
Impact
By combining retrieval with generation, RAG supports the Guideline’s principle of Accuracy & Reliability, underpinning safer deployment of generative AI in settings where factual correctness is critical.
Sources:- HK_Generative_AI_Technical_and_Application_Guideline_en.pdf
← Back to Knowledge Base