Bots are currently scraping the internet for LLM training data at unprecedented rates[1][2][3], driving up costs and destabilizing public-facing websites. I want to talk about how this has been particularly difficult for wikis, and has gotten much worse in the last few months.
I wonder if it would be possible to filter traffic by limiting the rate at which links can be followed after the initial connection.
I suppose one problem with that is it would also limit the crawlers you want, like search engine indexing. Maybe if enough sites were doing this it would generate some pressure on the AI companies to behave better.
If they have 10k URLs and in parallel find out all links via different IPs, they get new links.
When they do that again with the new links, there is no connection between visits, because the IPs differ. It looks like someone else is requesting what is behind the link.
If they have 10K IPs, then you need a DDOS filter, which would require a huge infrastructure no matter what.
The issue with wiki’s and source forges is that there is a maze of links to all past versions of everything, each generated on demand from a cpu-expensive database query. You basically have to limit the pages anonymous users can spider into. Forgejo has a setting to block expensive pages from non-logged in users for example.
Probably can configure anubis to require challenge that is proportional to the CPU time needed to render each page?
That seems to be a lot of peoples approach, but if they cared about time or bandwidth they wouldn’t be spidering Dow into your commit history multiple times a day. They have more patience and resources than your human readers.
I could see maybe caching that and providing it to a not-clearly-human user if it is in cache. That lets someone do something like link to a particular version of a file in a discussion here on the Threadiverse. The first user loading it will cause it to be cached.
Sure. Its just the thousands of obscure page edit history pages that ai crawlers hit every hour that cause the problem.