Learn Before
Combining Predictions from Diverse Prompts
A core principle in prompt ensembling is that utilizing different prompts, such as 'Condense and simplify this text' and 'Rewrite for easy reading,' will generate varied predictions from a model. These distinct outputs are then considered together to produce a final, aggregated prediction.
0
1
Tags
Ch.3 Prompting - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Related
A team wants to improve the robustness of its text summarization system. Their strategy involves creating several different instructions (e.g., 'Summarize the key points,' 'Provide a one-paragraph summary,' 'Extract the main conclusion') for the same input text. They plan to run each instruction through their single, fine-tuned language model and then use a method to combine the resulting summaries into a single, higher-quality final summary. Which of the following descriptions accurately represents the workflow for this strategy?
A technique is used to improve the reliability of a language model's output by using several different instructions for the same task and combining the results. Arrange the following steps to accurately represent the workflow of this technique.
Combining Predictions from Diverse Prompts
Analyzing a Flawed Workflow
Learn After
A research team is using a single large language model to generate a balanced and comprehensive summary of a complex, multi-perspective news article. To improve the quality of the final output, they decide to generate several different responses and combine them. Which set of prompts is most likely to produce the varied outputs needed for a high-quality, aggregated summary?
Diagnosing Ineffective Prompt Combination
Evaluating a Prompt Ensemble Strategy