RecurrentGPT offers an interactive framework for generating arbitrarily long text by simulating RNN recurrence with natural language. It's designed for writers and researchers exploring long-form content generation with LLMs, enabling human-AI co-creation and flexible story development.

How It Works

RecurrentGPT replaces traditional RNN components like cell state and hidden state with natural language descriptions. At each step, it processes a text paragraph and a plan for the next, generated in the previous step. It leverages a long-term memory (summaries of past paragraphs, retrievable via semantic search) and a short-term memory (natural language summaries of recent steps). These are combined with current inputs into a prompt for a backbone LLM to generate the next paragraph, a plan, and update memories. This approach allows for more intuitive and flexible long-form generation compared to purely vectorized methods.

Quick Start & Requirements

• Install via pip install -r requirements.txt.
• Requires an OpenAI API key.
• Local deployment involves editing utils.py for API keys and proxy settings, then running python gradio_server.py.
• Online demos are available at https://www.aiwaves.org/recurrentgpt and https://www.aiwaves.org/interactivefiction.

Highlighted Details

• Simulates RNN recurrence using natural language prompts and memory management.
• Supports interactive co-writing, allowing human intervention in planning and editing.
• Utilizes semantic search for retrieving relevant information from long-term memory.
• Offers flexibility in choosing topics, genres, and iteration counts via configuration scripts.

Maintenance & Community

• Project is associated with the paper "RecurrentGPT: Interactive Generation of (Arbitrarily) Long Text".
• A Discord community is available at 🔥 Discord.
• Local deployment with open-source LLMs is supported via a separate repository: Recurrent-LLM.

Licensing & Compatibility

• The README does not explicitly state a license. The citation format angular2 is unusual and may indicate a non-standard or missing license declaration.

Limitations & Caveats

• The project heavily relies on OpenAI's API, potentially incurring costs and limiting offline use unless alternative LLMs are integrated.
• The license is not specified, which may pose compatibility issues for commercial or closed-source applications.