# DirectGPT: A Direct Manipulation Interface to Interact with Large Language Models

**Authors:** Damien Masson, Sylvain Malacria, Géry Casiez, Daniel Vogel  
**arXiv:** [2310.03691](https://arxiv.org/abs/2310.03691)  
**Date:** 2023-10-05  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2310.03691v1/

## Key points

- The paper explores using direct manipulation principles to improve interaction with large language models.
- Strategies proposed include continuous representation of objects, reuse of prompt syntax in a toolbar, manipulable outputs, and undo mechanisms.
- DirectGPT is introduced as a user interface layer on top of ChatGPT for translating direct manipulation actions into prompts.
- Participants using DirectGPT were able to edit text, code, and vector images 50% faster than those using the baseline ChatGPT model.
- Users relied on 50% fewer and 72% shorter prompts when utilizing DirectGPT.
- The research validates integrating Large Language Models (LLMs) into software applications through direct manipulation techniques.
- Incorporating direct manipulation principles improves efficiency and enhances user experience in text editing tasks across various domains.

## Summary

, , , , 

The paper titled "DirectGPT: A Direct Manipulation Interface to Interact with Large Language Models" by Damien Masson, Sylvain Malacria, Géry Casiez, and Daniel Vogel explores the use of direct manipulation principles in improving interaction with large language models. The authors propose various strategies such as continuous representation of generated objects, reuse of prompt syntax in a toolbar of commands, manipulable outputs for composing or controlling prompts' effects, and the implementation of undo mechanisms. They introduce DirectGPT as a user interface layer built on top of ChatGPT that translates direct manipulation actions into engineered prompts. Through a study conducted by the authors, it was found that participants using DirectGPT were able to edit text, code, and vector images 50% faster than those using the baseline ChatGPT model. Additionally, users relied on 50% fewer and 72% shorter prompts when utilizing DirectGPT. This research contributes a validated approach to integrating Large Language Models (LLMs) into traditional software applications through direct manipulation techniques. Overall, the study demonstrates the effectiveness of incorporating direct manipulation principles in interacting with large language models like ChatGPT. The findings suggest that this approach not only improves efficiency but also enhances user experience when working with text editing tasks across various domains.