Learn Before
Characteristics of the Starting Model for Alignment
In a process designed to align a language model with human preferences using feedback, what are the two essential training stages the model should have completed before the feedback process begins? Briefly explain the purpose of each stage in preparing the model to serve as a functional baseline.
0
1
Tags
Ch.2 Generative Models - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Comprehension in Revised Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
A development team is preparing to use a human feedback-driven process to improve an AI's helpfulness and safety. They have two candidate models to use as their starting point:
Model A: A raw, pre-trained model that is very good at predicting the next word in a sentence but has not been specifically trained to follow user commands.
Model B: A model that has been pre-trained and then further fine-tuned on a dataset of instructions and high-quality answers, making it proficient at following user commands.
Which statement best evaluates the choice of a starting model for this alignment process?
Diagnosing an Inefficient Alignment Process
Characteristics of the Starting Model for Alignment