@vv @cwebber Isn't the difference a little blurred? There are definitely the spiders that collect training material from the Web, which is used for creating the models. But most chat and coding systems now have RAG and MCP and similar tool infrastructure, so they can do live web searches and API calls at query time, not training time. I haven't seen anyone who talks about the effect from excess traffic on LLMs on Web sites and services tease those two apart.