Learn Before
Determining Individual Predictions in Prompt Ensembling
A data scientist is using a language model to classify a user comment: 'The interface is a bit clunky, but the features are powerful.' They use two different prompts to get a prediction.
Prompt 1 (x₁): 'Classify the sentiment of this comment: ...' Prompt 2 (x₂): 'Is the following user feedback positive, negative, or neutral? ...'
The model returns the following conditional probabilities for each prompt:
- For x₁: Pr(Positive|x₁)=0.6, Pr(Negative|x₁)=0.3, Pr(Neutral|x₁)=0.1
- For x₂: Pr(Positive|x₂)=0.5, Pr(Negative|x₂)=0.2, Pr(Neutral|x₂)=0.3
Based on the mathematical principle of maximizing the conditional probability for each prompt, what is the individual prediction (ŷᵢ) for Prompt 1 and Prompt 2?
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
Formula for Combining Predictions in Prompt Ensembling
Bayesian Interpretation of Prompt Ensembling
A developer uses a technique involving three distinct prompts to classify a customer review as 'Positive', 'Negative', or 'Neutral'. The prompts are sent to a language model to get a response for each. According to the mathematical formulation of this technique, what is the immediate next step in the process?
Determining Individual Predictions in Prompt Ensembling
An engineer is implementing a system that uses a set of K distinct prompts to improve the reliability of a text summarization task. They notice that the final, combined summary is often incoherent. Upon investigation, they discover that for each individual prompt , the system is not selecting the single summary with the highest conditional probability, but is instead randomly choosing one from the top five most likely summaries. Which specific component of the mathematical formulation for this technique is being incorrectly implemented?