This project provides an AI chatbot framework designed for anti-fraud scenarios, targeting developers and researchers interested in conversational AI and NLP. It leverages a large Chinese language model and the Wechaty framework to enable 24/7 real-time WeChat conversations, offering improved fluency over traditional QA models.

How It Works

The core of the chatbot is the YUAN 1.0 model, a 245.7 billion parameter Chinese pre-trained language model. The framework is modular, allowing for the integration of other conversational models like ChatGPT or Wenxin. It employs a prompt-based approach, enhanced by few-shot learning with curated examples derived from online discussions to guide the AI's conversational strategy and improve its ability to recognize scam tactics. A memory mechanism, inspired by LSTMs, is implemented to maintain conversational context over multiple turns.

Quick Start & Requirements

• Installation: pip install requests hashlib json followed by git clone https://github.com/Shawn-Inspur/Yuan-1.0.git. For Wechaty integration, npm install wechaty or use Docker.
• Prerequisites: Python >= 3.7, Tensorflow-gpu 1.15.2, CUDA >= 10.0, CuDNN >= 7.6.0. For Wechaty: a Wechaty token, a dedicated WeChat account, and a Linux server (CentOS 8.0+ recommended).
• Setup: Requires obtaining a YUAN 1.0 API key and potentially a Wechaty token.
• Links: YUAN 1.0 API: YUAN 1.0, Wechaty: wechaty/wechaty.

Highlighted Details

• Utilizes the 245.7 billion parameter YUAN 1.0 model, claiming superior performance on Chinese NLP benchmarks.
• Employs few-shot learning with AI script-killing examples for anti-fraud scenario adaptation.
• Integrates with Wechaty for 24/7 real-time WeChat conversational capabilities.
• Features a memory mechanism to handle multi-turn dialogue context.

Maintenance & Community

The project was initiated in September 2022 with an open-source release in December 2022. Community engagement and further development details are not explicitly detailed beyond the GitHub repository.

Licensing & Compatibility

The project is presented as for technical research and popular science purposes only, with no commercial application authorization provided. The specific license is not explicitly stated in the README.

Limitations & Caveats

The project is primarily focused on Chinese language and anti-fraud scenarios. The effectiveness of the memory mechanism may decrease with low-similarity prompts, potentially leading to repetitive answers. The use of specific Wechaty puppet services may incur costs.