247|AI – RAG Model Basics

Challenges
, , ,
1

BYU Student Author: @Kiya_Smith
Reviewers: @ Kylie_Larsen17 @Kaden_Sandberg
*Estimated Time to Solve: *20 Minutes

We provide the solution to this challenge using:

  • ChatGPT

Need a program? Click here.

Overview
This challenge will help you understand what a RAG (Retrieval -Augmented Generation) model is and why they are used. It will also give a basic overview of how to create a RAG model. Then you will be quizzed on what you have learned. After completing this challenge, you will have a better understanding of the basics of RAG modeling.

Instructions
Follow the link to a site that explains what RAG models are and take the quiz. You can use the information provided as well as AI or other sources to answer the questions.

Make sure you get questions 1-3 correct on the quiz and paste your answer to all 6 questions in the comments as your solution. NOTE: questions 4-6 will not say if you got the answer correct or incorrect. They are open ended questions to test understanding. Still attempt to answer the questions to the best of your ability because the solution below will give an example of potential correct answers.

Data Files

Suggestions and Hints

Pay attention to key words in your answers.

Solution

Challenge247_Solution.docx

THE FIRST THREE QUESTIONS HAVE SPECIFIC CORRECT ANSWERS

  1. What are the two key components of a RAG model?
    Retrieval and generation

  2. How does the Retrieval component of a RAG model work?
    It retrieves specific information from a database

  3. What are three benefits of using a RAG model?
    Up to date data, accurate data, and specific information

THE NEXT THREE QUESTIONS ARE OPEN ENDED AND WILL NOT SHOW AS CORRECT OR INCORRECT ON THE QUIZ

  1. How does a RAG model improve accuracy compared to a regular AI model?
    It is based off of a database that is specific to whatever the RAG model is focused on. This means the answers are limited to relevant information. (This answer requires understanding how retrieval prevents hallucinations by grounding responses in real documents.)

  2. Why might a business choose to use a RAG model instead of training a custom AI model from scratch?
    The RAG model is business specific and can only look in the database to retrieve answers to questions. This means any information it comes up with will be pulled from what is already known. (This question requires consideration of cost, flexibility, and the ability to update information dynamically.)

  3. What challenges might arise when implementing a RAG model, and how can they be addressed?
    When there is not an answer to a question in the database the RAG model could potentially make something up. This can be addressed by training the model to provide an answer that says something like “the is no information related to this question available” instead of making something up if it’s not in the database. (This question requires thinking about retrieval errors, slow response times, and data quality issues.)

(Any answers that are along these same lines are valid as well)

Solution Video: Challenge 247AI RAG_Model_Basics