# RecurrentGPT: Interactive Generation of (Arbitrarily) Long Text

**Authors:** Wangchunshu Zhou, Yuchen Eleanor Jiang, Peng Cui, Tiannan Wang, Zhenxin Xiao, Yifan Hou, Ryan Cotterell, Mrinmaya Sachan  
**arXiv:** [2305.13304](https://arxiv.org/abs/2305.13304)  
**Date:** 2023-05-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.13304v1/

## Key points


## Summary

In this paper, the authors introduce RecurrentGPT, a language-based model that enables the generation of arbitrarily long text without forgetting. Unlike hierarchical methods that prompt LLMs for long story generation in a plan-and-write fashion, RecurrentGPT overcomes limitations via recurrent generation, enabling effective human-LM collaboration and improving flexibility and controllability for long text generation. The authors also discuss the limitations of existing AI writing assistants and automatic generation systems and how RecurrentGPT can improve the quality of AI-generated long texts while increasing the productivity of human writers. They further demonstrate the possibility of using RecurrentGPT as an interactive fiction that directly interacts with consumers, which they call "AI As Contents" (AIAC). The authors provide evaluations of RecurrentGPT's performance in generating novel genres such as sci-fi and fantasy, as well as discussing potential social impacts and promoting responsible usage of generative AI.