# Conversational Prompt Engineering

**Authors:** Liat Ein-Dor, Orith Toledo-Ronen, Artem Spector, Shai Gretz, Lena Dankin, Alon Halfon, Yoav Katz, Noam Slonim  
**arXiv:** [2408.04560](https://arxiv.org/abs/2408.04560)  
**Date:** 2024-08-08  
**Summary page:** https://www.summarizepaper.com/en/paper/2408.04560v1/

## Key points

- Conversational Prompt Engineering (CPE) is a user-friendly tool developed by a team of researchers to streamline prompt engineering for Large Language Models (LLMs).
- The CPE process involves two main stages: generating data-driven questions based on user-provided data and refining prompts based on user feedback.
- A recent study demonstrated the effectiveness of CPE in creating personalized and high-performing prompts for summarization tasks.
- Zero-shot prompts obtained through CPE were shown to be comparable in performance to longer few-shot prompts, indicating efficiency gains for repetitive tasks with large text volumes.
- CPE offers a promising solution for enhancing communication with LLMs through tailored prompts that meet individual preferences and task requirements.

## Summary

Conversational Prompt Engineering (CPE) is a user-friendly tool developed by Liat Ein-Dor, Orith Toledo-Ronen, Artem Spector, Shai Gretz, Lena Dankin, Alon Halfon, Yoav Katz, and Noam Slonim to streamline the process of prompt engineering for communicating with Large Language Models (LLMs). The tool utilizes a chat model to interact with users and understand their output preferences. The CPE process involves two main stages: first, the model uses user-provided unlabeled data to generate data-driven questions and incorporates user responses to shape the initial instruction. Subsequently, the model shares the outputs generated by the instruction and uses user feedback to further refine both the instruction and outputs. The final outcome is a few-shot prompt where the approved outputs serve as examples. A recent study on summarization tasks demonstrated the effectiveness of CPE in creating personalized and high-performing prompts. Results showed that zero-shot prompts obtained through CPE were comparable in performance to longer few-shot prompts. This suggests significant efficiency gains for repetitive tasks with large text volumes. In conclusion, offers a promising solution for enhancing communication with LLMs through tailored prompts tailored to individual preferences and task requirements.