Learn Before
Choosing an Information Integration Strategy
A startup is developing a customer support AI for its software, which is updated weekly with new features and bug fixes. They are using a powerful, general-purpose language model. The engineering team is debating two approaches to ensure the AI provides up-to-date answers:
- Approach A: Periodically retrain the entire language model with the latest documentation. This process is computationally expensive and takes several days to complete.
- Approach B: For each customer question, build a system that first finds the most relevant sections from the latest documentation and then provides those sections to the existing, unchanged language model along with the question.
Based on the startup's need for the AI to handle weekly updates, evaluate the two approaches. Which approach is more suitable and why? Justify your answer by explaining the key operational difference between how each approach incorporates new information.
0
1
Tags
Ch.3 Prompting - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Evaluation in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
A company wants to build a chatbot that uses a pre-existing, general-purpose Large Language Model to answer questions about its new product line, whose documentation was just finalized. The company has a very tight deadline and does not have the computational resources to modify the underlying model. Which of the following statements best explains the primary advantage of using a system that retrieves relevant documentation to add to the model's input for each user query?
Choosing an Information Integration Strategy
Fine-Tuning LLMs to Enhance RAG Performance
To integrate a new set of documents into a system that uses a pre-existing Large Language Model to answer questions, the standard approach involves modifying the model's internal parameters to learn the new information.