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.
The really exasperating thing is that Wikipedia makes all of their data available in compressed database form, and I suspect that other MediaWiki wikis could do so too. I checked one of the wikis being complained about, the Minecraft wiki, and it’s a MediaWiki wiki.
This isn’t even a case where the server operators have any problem with the data being available to the bots. In fact, people are going out of their way to package up all the data in a form that is optimal for computer processing (better for the bots) and which doesn’t create additional load on the wiki servers (better for the wiki operators).
But the bots aren’t using them! And on top of that, they aren’t even written to follow traditional conventions for being polite in scraping. We’ve had lots of software spider the Web — that’s not new — but normally that spidering software has followed basic politeness to try to avoid excessive disruption to the servers being spidered.
It’s as if someone has set up a table out front of their house with a big sign reading “Free cookies!” with boxes of pre-packaged cookies. Instead of taking the damn cookies, the bots are obtaining the largest trucks possible and then ramming holes in the house with a truck and scavenging through the ruins, desperately trying to find cookies.
It’s a weird phenomenon I encountered as a data engineer inside corporations. Teams wanted to run endless queries off our API instead of using the snapshots we provided. Most of the apps didn’t have a justifiable reason for wanting up to the minute information. 🤷♂️