AI_Gen_Novel

Exploring the limits of AI's ability to write online articles

Introduction

In recent years, AI has made significant progress in the field of literary creation. From the AI ​​micro-fiction competition to the wonderful writing of literature to Midreal AI, these cases have proved the great potential of AI in literary creation. As an online literature enthusiast, I hope to develop an application that can automatically generate online novels through large language models and multi-agent technology.

The creation of online articles can apply the cognitive process model of writing , which regards writing as a goal-oriented thinking process, including nonlinear cognitive activities: planning, transformation and review.

Literature and practice show that LLM performs better in conversion and review, but has flaws in the planning stage. Specifically reflected in

• Limited understanding and reasoning skills
• Unable to remember and generate long texts
• Lack of originality and diversity

My solution to these problems is

• Using the ability of LLM to compress long text into memory composed of a few sentences
• Optimize Prompt, collaborate with multiple agents, stimulate LLM's capabilities and improve its originality
• Drawing on the core idea of ​​RecurrentGPT , iteratively creates text of any length based on language loop calculations.
• Combining the prior knowledge of online novel creation, optimize the creative process

After some exploration, I think the current large language model does not have enough ability to create long online novels .

The possible future development directions are:

Requirements for large-scale models

• Longer context
• Stronger understanding and reasoning skills
• Reinforcement learning training for online articles based on human judgment criteria

Optimization of program architecture

• Try multiple times, discuss more intelligently, and improve originality
• Introduce more prior knowledge, such as structured plot frameworks

Online experience

https://modelscope.cn/studios/cjyyxn/AI_Gen_Novel/summary

Quick guide to get started

Step 1: Environment Configuration

First, install the required dependencies:

pip install -r requirements.txt

Step 2: Implement the LLM interface

The project relies on a large language model. You need to implement the chatLLM function in LLM.py

Step 3: Run the project

• Run demo.py directly, the novel will be automatically created and the results will be saved in the novel_record.md file.

• Run app.py to start a gradio-based application. By opening the displayed link, you can experience the visualization process of AI novel generation.

Star History