Skip to content
TheChatArchiveThe Chat Archive
Qwenqwen-max-2025-01-25· Alibaba (Qwen)Hugging Face LMArena Human Preference 140K

You have to provide a list of those most relevant internet domains names (top and…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with o4-mini-2025-04-16’s answer.
  1. User

    You have to provide a list of those most relevant internet domains names (top and second-level domain, e.g. “example.com”) used for scraping the content for training of the AI model, by listing the top 10 % of all domain names determined by the size of the content scraped (in a representative manner across all modalities where applicable). Tell me how many (number) of domain names would be in case of current large-scale AI models (meaing 10^23 FLOPs or above according to https://epoch.ai/blog/tracking-large-scale-ai-models).

  2. Qwen Assistant

    To address your question, we need to break it down into several components:

    1. Understanding the context of large-scale AI models:

      • Large-scale AI models are typically trained on vast amounts of data scraped from the internet. These models often reach computational scales of 102310^{23} FLOPs or more, which corresponds to extremely large training datasets.
      • The training data for such models is usually drawn from a diverse range of internet domains, including text, images, videos, and other modalities.
    2. Determining the "top 10% of domain names by size of content scraped":

      • When scraping content for training AI models, the amount of data collected from each domain varies significantly. Some domains (e.g., Wikipedia, Reddit, or news websites) contribute disproportionately large amounts of data due to their high volume of content, structured formats, and open licensing.
      • To identify the top 10% of domain names, one would first need to aggregate all domains used for scraping, rank them by the size of the content scraped, and then select the top 10%.
    3. Estimating the total number of domains used in training:

      • Large-scale AI models are trained on datasets that are often derived from extensive web crawls. For example, Common Crawl, a widely used dataset for training AI models, indexes billions of web pages across millions of domains.
      • While the exact number of domains used for training is not always disclosed, it is reasonable to estimate that the total number of unique domains could range from hundreds of thousands to millions, depending on the breadth of the crawl and filtering criteria.
    4. Calculating the number of domain names in the top 10%:

      • If we assume that the total number of unique domains used for training is approximately 1 million (a reasonable estimate for large-scale models), then the top 10% would consist of 10%×1,000,000=100,00010\% \times 1,000,000 = 100,000 domain names.
    5. Applying this to current large-scale AI models:

      • Models at the 102310^{23} FLOPs scale or above (e.g., GPT-4, PaLM, etc.) are likely trained on datasets derived from similarly extensive web crawls. Therefore, the number of domain names in the top 10% would still be on the order of 100,000, assuming a total pool of 1 million domains.

    Final Answer:

    100,000\boxed{100,000}

    This represents the estimated number of domain names in the top 10% by size of content scraped for training current large-scale AI models. Note that this is an approximation and the actual number could vary depending on the specific dataset and model training methodology.

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