TCT - Transformer Circuits Tracker

An unofficial RSS feed generator for Transformer Circuits, a research publication from Anthropic's Interpretability team.

Overview

This project automatically scrapes the Transformer Circuits website and generates an RSS feed, making it easy to stay updated with the latest mechanistic interpretability research without manually checking the website.

The RSS feed is automatically updated daily via GitHub Actions and hosted on GitHub Pages.

Features

• Automatic Updates: RSS feed refreshes daily at 06:17 UTC via GitHub Actions
• Complete Coverage: Captures all articles including research papers and circuit updates
• Clean Content: Handles UTF-8 encoding properly, avoiding mojibake characters
• Metadata Rich: Includes publication dates, descriptions, and proper titles
• Sorted by Date: Articles are ordered newest-first for easy consumption

RSS Feed URL

Subscribe to the feed at: https://dschaehi.github.io/TCT/index.xml

You can use any RSS reader to subscribe to this feed.

Local Usage

Prerequisites

• Python 3.11 or higher
• pip package manager

Installation

1. Clone the repository:

git clone https://github.com/dschaehi/TCT.git
cd TCT

2. Install dependencies:

pip install -r requirements.txt

Generate the RSS Feed

Run the feed generator script:

python scripts/generate_feed.py

This will create/update the RSS feed at docs/index.xml.

How It Works

The script (scripts/generate_feed.py):

1. Fetches the homepage of transformer-circuits.pub
2. Extracts all article links (both .note cards and .paper elements)
3. For each article, retrieves metadata including publication dates
4. Generates a properly formatted RSS 2.0 feed with all discovered articles
5. Saves the output to docs/index.xml

The GitHub Actions workflow (.github/workflows/rss.yml):

• Runs daily on a schedule
• Executes the feed generator
• Commits and pushes any updates to the feed

Dependencies

• beautifulsoup4 - HTML parsing
• requests - HTTP requests
• feedgen - RSS feed generation
• lxml - XML processing

Contributing

Contributions are welcome! Feel free to:

• Report bugs or issues
• Suggest improvements
• Submit pull requests

License

MIT License - see LICENSE file for details.

Copyright (c) 2025 Jae Hee Lee

Disclaimer

This is an unofficial RSS feed. The content is sourced from transformer-circuits.pub, which is maintained by Anthropic's Interpretability team. This project is not affiliated with or endorsed by Anthropic.

Acknowledgments

• Thanks to the Anthropic Interpretability team for publishing their research at transformer-circuits.pub
• This project helps researchers and enthusiasts stay updated with the latest developments in mechanistic interpretability