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.
And so the Tragedy of the Commons plays out, yet again.
There’s no cost to being a selfish asshole, so it’s sadly not surprising that many individual actors are destroying the public Internet. Like, how can we align incentives to stop this? Regulations/laws are mostly pointless since the very same tactics used to dodge bot detection also make it incredibly hard to identify the originator.
The only other disincentive with a real cost, that I can think of, would be to poison the data fed to scrapers, so they get bad data? That seems expensive to set up, though.
I think TFA has the best solution idea: make it easy to scrape all the useful data using a low-cost standardized system. Then there’s no incentive to scrape the website using a stupid, expensive crawler in the first place.
Edit: actually, LLMs make poisoning the data fairly reasonable… When there’s a high volume of requests for outdated pages/edit pages/other rarely accessed pages, have the server serve a pre-cached parody version of the root page instead. Pre-build one parody copy of each page with a standardized prompt, like “rewrite this page like it comes from an academic journal of medicine or economics with APA citations for every fact.”