AI-Safety Reading List

🌐 Live Site: Check out the running instance at papers.unsearch.org

An automated system that creates and maintains a curated list of AI safety research papers, including summaries and multi-channel distribution. The system automatically fetches papers from Semantic Scholar, generates summaries using Claude, and distributes content through a static website, with planned support for Telegram and Twitter distribution.

Features

• 🤖 Automated paper discovery using Semantic Scholar API
• 📝 AI-powered paper summarization using Claude
• 🖼️ Automatic figure extraction and analysis
• 📊 Google Sheets integration for paper tracking and management
• 🌐 Static website generation using Hugo
• 🔄 Continuous deployment with Netlify
• 📱 Planned support for Telegram and Twitter distribution

Components

1. Backend

   • Paper discovery and fetching via Semantic Scholar API
   • Paper caching and management
   • Google Sheets integration for tracking
   • Claude-powered summarization
   • Figure extraction and processing

2. Static Website

   • Hugo-based static site
   • Automatic deployment via Netlify
   • Responsive design for all devices
   • Full paper archive with summaries

3. Planned Features

   • Telegram bot for paper notifications
   • Twitter bot for sharing summaries
   • User feedback mechanism
   • Paper categorization system

Setup

1. Clone the repository:

   git clone https://github.com/yourusername/ai-safety-reading-list.git
   cd ai-safety-reading-list

2. Install dependencies:

   pip install -r backend/requirements.txt
   pip install -r ai-safety-site/site-requirements.txt

3. Configure API keys and settings:

   • Create secrets/config.yaml with your API keys:

     semantic_scholar:
       api_key: "your_api_key"
     
     google_sheets:
       credentials_file: "your_credentials.json"
       spreadsheet_id: "your_spreadsheet_id"
       range_name: "Sheet1!A1:K"
     
     anthropic:
       api_key: "your_claude_api_key"

4. Set up Google Sheets:

   • Create a service account and download credentials
   • Share your tracking spreadsheet with the service account email
   • Place credentials JSON in the secrets directory

Usage

1. Run the backend processor:

   cd backend/src
   python main.py

2. Build the Hugo site:

   cd ai-safety-site
   hugo server

Repurposing for Other Topics

This system can be easily adapted for tracking research papers on any topic. To repurpose:

1. Update the search queries in main.py:

   queries = ["Your Topic 1", "Your Topic 2"]

2. Modify the Claude prompts in paper_summarizer.py to focus on your topic

3. Update the website configuration in ai-safety-site/config.toml

4. Adjust the Google Sheets structure if needed

Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

License

MIT License

Acknowledgments

• Semantic Scholar for their API (and for providing a free API key!)
• Anthropic for Claude API (unfortunately no free API key usage for us yet 😢)
• Hugo for static site generation
• Netlify for hosting
• Monochrome - A clean, fast and responsive Hugo theme by @kaiiiz