A research team is refining a language model using two distinct methods. In Method A, they train the model on a large dataset of specific commands paired with ideal, human-written responses that perfectly execute those commands (e.g., Command: 'List three benefits of solar power.' Ideal Response: A list of exactly three benefits). In Method B, they show human raters two different model-generated responses to the same open-ended prompt (e.g., 'Write a short, encouraging note') and ask the raters to choose which response they prefer. The model is then updated based on these preferences. What fundamental difference in goals do these two methods represent?
0
1
Tags
Ch.4 Alignment - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Analysis in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
A research team is refining a language model using two distinct methods. In Method A, they train the model on a large dataset of specific commands paired with ideal, human-written responses that perfectly execute those commands (e.g., Command: 'List three benefits of solar power.' Ideal Response: A list of exactly three benefits). In Method B, they show human raters two different model-generated responses to the same open-ended prompt (e.g., 'Write a short, encouraging note') and ask the raters to choose which response they prefer. The model is then updated based on these preferences. What fundamental difference in goals do these two methods represent?
Diagnosing Chatbot Performance Issues
A development team is working on two separate improvement goals for their language model. Match each goal with the alignment methodology it primarily represents.