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 $\mathbf{x}_i$, 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?