AISafetyGraph
AI Safety Graph
Install / Use
/learn @ai-safety-graph/AISafetyGraphREADME
AI Safety Graph
<img src="/docs/assets/ai_graph_3_sec.gif" width="100%"> <div align="center">📅 Recent Updates
[August 2024] 🥇 First Place at Research Augmentation Hackathon
[January 2025] 🎉 Now part of Apart Lab Studio!
👉 Check out the live version here: AI Alignment Research Graph
👋🏻 Getting Started & Join Our Community
Whether you're a newcomer or a seasoned researcher, we have a place for you in our community. Here are some ways to get started:
| | Visit the website |
| :---------------------------------------- | :----------------------------------------------------------------------------------------------------------------- |
|
| Join our Discord community! |
[!IMPORTANT]
Star Us, to get updates from GitHub ~ ⭐️
Features
- Interactive graph visualization of AI Alignment Research
- High-Quality LLM-based segementation of papers
- Search for papers by title, author, or abstract
- Click on a node to view a summary of the topic
Local Build
To build locally, first clone with submodules:
# Clone repo with submodule
git clone --recurse-submodules https://github.com/ai-safety-graph/AISafetyGraph.git
# Or if you've already cloned without submodules:
git submodule update --init --recursive
# node lts/iron
npm i
npx quartz build --serve
Development Notes:
- the aisgraph_quartz submodule points to aisgraph which is a fork of quartz
- the content folder stores the markdown files that represents the graph strucuture
- the generate_md folder contains the code to generate the .md files
- generating .md files requires ai-alignement-dataset-jsonl-file to be placed under generate_md/dataset folder
- generating .md files requires a anthropic api key to be stored in
/generate_md/.envfile
Troubleshooting
<details> <summary>If you get ADDRINUSE: address already in use :::8080</summary>npx kill-port 8080
</details>
Contributors
Acknowledgements
- Thanks to alignment-research-dataset for the dataset
- Built using Quartz v4
- Commits prior to commit hash #5c7cb55 come from the quartz v4 web framework, this is to allow easier updates of the web-framework using
git pull upstream
| <img src="docs/assets/LISA.svg" width="120" alt="LISA"> | <img src="docs/assets/Apart.png" width="120" alt="Apart Research"> | <img src="docs/assets/SAIL.png" width="120" alt="SAFE AI London"> | |:---:|:---:|:---:| | LISA | Apart Research | SAFE AI London |
TODO:
- [ ] Refactor the code in explore_ds.py to output in json and not yaml format
- [ ] run llm_cluster.py on the entire source
Improvements:
- [ ] explore_ds.py currently filters by arxiv papers, could also support other sources
