Automatic Prompt Engineering: How?

Automatic Prompt Engineering (APE) automates the creation of prompts^[1]. It involves a model generating multiple prompts, evaluating them, and refining the good ones, repeating the process^[1].

Here's how it works^[1]: First, a model is prompted to generate output variants^[1]. Next, evaluate the generated prompts by scoring them based on a chosen metric such as BLEU (Bilingual Evaluation Understudy) or ROUGE (Recall-Oriented Understudy for Gisting Evaluation)^[1]. Finally, select the prompt with the highest evaluation score for use; this can be further tweaked and re-evaluated^[1].