The actual LLM benchmark - make a site as wonderful both in visual presentation and in code, while being as factually accurate as ciechanow.ski.

If the LLM can create something that could be mistaken for that work, that's awesome.