Learn Before
Improving Model Output Consistency
A data scientist is creating few-shot demonstrations to train a language model to solve multi-step physics problems. The model is expected to show its work and then provide a single numerical answer. However, during testing, the model's final answers are often embedded within explanatory sentences, making them difficult to parse programmatically. For example, an output might be '...and so, after calculating the final velocity, we find that the object is moving at 19.6 m/s.'
Describe a specific, simple modification you could make to the structure of the demonstrations to train the model to present its final answer in a more consistent and easily extractable format. Explain why this modification would be effective.
0
1
Tags
Ch.3 Prompting - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Application in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
A developer is creating few-shot demonstrations to teach a language model to solve word problems. They notice the model's outputs are often verbose and fail to clearly state the final numerical answer, even when the reasoning steps are correct. Review the following demonstration from their prompt:
Q: A grocery store had 50 cans of soup. They sold 15 on Monday and received a new shipment of 25. How many cans do they have now? A: The store started with 50 cans. They sold 15, so 50 - 15 = 35. Then they received 25 more, so 35 + 25 = 60. The store now has 60 cans.
Which of the following critiques best identifies the primary weakness in this demonstration that is likely causing the model's inconsistent output format?
Improving Model Output Consistency
Refining a CoT Prompt for Programmatic Extraction