Microsoft Certified: Azure AI Engineer AssociateImplement natural language processing solutionsMedium
A financial institution is building a chatbot for internal employees to answer questions about company policies, benefits, and IT support. The chatbot needs to provide precise, factual answers based on a large, continuously updated knowledge base of internal documents. It must avoid generating incorrect or 'hallucinated' information. Which generative AI pattern should the institution implement?
- APure Generative Model
- BAbstractive Summarization
- CFine-tuning a Large Language Model (LLM) without external data
- DRetrieval Augmented Generation (RAG)
Show answer & explanationAnswer & explanation
Correct answer: D. Retrieval Augmented Generation (RAG)
Retrieval Augmented Generation (RAG) combines the strengths of retrieval (finding relevant information from a knowledge base) and generation (using an LLM to formulate an answer). This pattern is ideal for grounding responses in factual, up-to-date information, thereby minimizing hallucinations and ensuring precision, which is critical for a financial institution's internal chatbot.
Why the other options are wrong
- A. A pure generative model is prone to 'hallucinations' and cannot guarantee factual accuracy from an external knowledge base.
- B. Abstractive Summarization condenses text, it does not answer specific questions from a knowledge base.
- C. Fine-tuning without external data would still rely on the LLM's pre-trained knowledge, which might be outdated or incomplete for specific internal documents, leading to inaccuracies.
Retrieval Augmented Generation (RAG)
A generative AI pattern that combines information retrieval with large language models to generate responses grounded in specific, external data sources.
- Reduces 'hallucinations' by providing factual context.
- Enables LLMs to access and utilize up-to-date, proprietary information.
- Improves accuracy and trustworthiness of generated responses.
Memory trick: RAG helps the AI 'remember' facts by retrieving them from its library before speaking.