Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-3-7-sonnet-20250219· AnthropicHugging Face LMArena Human Preference 140K

Tell me a few websites (apart from Wikipedia) where the articles could be used to…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Metallama-4-maverick-17b-128e-instruct’s answer.
  1. User

    Tell me a few websites (apart from Wikipedia) where the articles could be used to construct a public LLM evaluation benchmark. This may imply relatively loose Licensing.

  2. Claude Assistant

    Some websites with content that could be used for LLM evaluation benchmarks due to their relatively permissive licensing include:

    Stack Exchange network (including Stack Overflow) - uses CC BY-SA 4.0 licensing which allows reuse with attribution

    Project Gutenberg - contains public domain books and texts

    arXiv - academic preprints with various open licenses

    Common Crawl - web data archive widely used in AI training

    HackerNews - discussions under Creative Commons licensing

    Would you like me to explain more about any of these sources or their specific licensing considerations for benchmark creation?

  3. User

    I'd prefer options that are constantly updated with new articles. Of course arXiv and Stack Exchange are okay, but what else?

  4. Claude Assistant

    For constantly updated sources with permissive licensing that could work for LLM benchmarks:

    The Conversation - academic articles under CC BY-ND 4.0

    Internet Archive blogs - frequently updated with CC licensing

    GitHub README files and documentation - many under MIT or Apache licenses

    Wikimedia Commons - beyond Wikipedia, includes media with various open licenses

    OpenStreetMap wiki and documentation - regularly updated under open licenses

    Government data portals (like data.gov) - public domain content that's frequently refreshed

    These sources provide fresh content while maintaining licensing terms that could potentially allow for benchmark creation.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.