AI Test Checks If Agents Can Rebuild Websites From Browsers
A new open test called WebsiteBench asks AI agents to rebuild websites by exploring them through a web browser. It measures how well such agents can understand a site's layout and content just by clicking through it. The project is shared openly on GitHub, so anyone can try it.
Quick glossaryTIGER-AI-Lab, browser exploration, GitHub
- TIGER-AI-Lab
- The research group behind WebsiteBench, which shares its work publicly on GitHub, a site where developers post code.
- browser exploration
- Letting an AI click through and read web pages in a browser to learn how they work.
- GitHub
- A website where developers store and share code projects, often openly for anyone to use.
Burning Potato score: 71%