Learn Before
Analyzing Feedback for a Multi-Step Reasoning Task
A language model is given the prompt: 'A bakery sells muffins for $3 each and cookies for $2 each. If they sold 15 muffins and 20 cookies on Monday, what was their total revenue?' The model generates the multi-step response below. A reward system that evaluates the entire response with a single score gives it a low rating because the final answer is incorrect. Analyze the primary limitation of this single-score feedback approach for improving the model's reasoning. Then, explain how evaluating each step of the response as a separate unit would provide more useful training data.
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
Notation for a Set of Output Segments
Input Formulation for Segment-Based Reward Computation
Difficulty of Obtaining Segment-Level Human Preference Data
Applying Pointwise Methods for Segment-Level Reward Modeling
Alignment as a Segment Classification Problem
Strategies for Segmenting Output Sequences in Reward Modeling
Analyzing Feedback for a Multi-Step Reasoning Task
A team is training a language model to generate detailed, multi-paragraph explanations of complex scientific phenomena. They observe that while the final conclusions are often correct, the intermediate steps in the explanations frequently contain subtle inaccuracies or logical gaps. Which of the following feedback strategies would be most effective for identifying and correcting these specific intermediate errors during training, and why?
Reward Model as an Imperfect Proxy for the Environment
Evaluating Reward Modeling Strategies for Creative Writing