Applying Retrieval-Augmented Generation (RAG) to Combat Hallucinations in GenAI

phenomenon of "hallucinations," where models generate plausible-sounding but incorrect or nonsensical information. This presentation delves into the innovative technique of Retrieval-Augmented Generation (RAG) as a solution to this problem. By integrating retrieval mechanisms with generative models, RAG significantly enhances the accuracy and reliability of AI outputs. Attendees will learn about the principles of RAG, its implementation strategies, and practical applications, gaining insights on how to effectively reduce hallucinations in their own GenAI applications.

1. Applying Retrieval-Augmented Generation (RAG) to Combat Hallucinations in GenAI Auckland AWS User Group - July 2024 Geethika Guruge Lead Consultant @ Mantel Group AWS Ambassador | AWS Community Builder

2. Agenda GenAI Beyond the Buzz Customizing Foundation Models What is RAG The complicated workflow Amazon Bedrock to the Rescue Demo

3. Remember this???

4. GenAI Beyond the Buzz

5. Why Customize GPT = Generative Pre-trained Transformers, Model doesn't know your domain

6. Why Customize Adapt to domain specific language Health care terminology Medical devices Enhance Performance Teach the model about the specialised tasks your organization does Improve context awareness in responses Provide the model with external data Company intranet Your code base

7. Customization Options Prompt Engineering ( Craft and Iterate upon the input) Priming Weigiting Chaining RAG Leveraging external knowledge sources NOT changing anything in the foundation model Fine-tune Adapt a foundation model on a specialized task set Training the foundation model on labeled examples of tasks Specifying the expected output and outcome Re-train

8. Customization Effort

9. When to Customize

10. What is RAG Retrieval Fetches the relevant content from the external knowledge base Argumentation Argument the retrieved context to the user prompt Generation Response from the foundation model based on the augmented prompt

11. Types of Retrieval Rule based Unstructured data Keyword searches Structured data Retrieval from a database i.e Select cases where resolution like reboot Semantic search Get relevant documents based on text embedding

12. What are Embeddings Numerical representation of text (vectors) Captures semantics and relationships between words. Embedding models capture features and nuances of the text. Rich embeddings can be used to compare text similarity. Multilingual Text Embeddings can identify meaning in different languages.

13. What are Embeddings

14. Importance of Embedding in RAG Powers text retrieval based on semantic meaning. Used to augment prompts with more accurate context from vector stores High-accuracy embeddings leads to improved context and higher quality responses Less Hallucinations !!!!!!

15. RAG in Action

16. Knowledge Bases for Amazon Bedrock

17. Retrieval and Generate

18. RetrieveAndGenerate API

19. Demo Time!!!!!