But I do know he created an enormous dataset of anime images used to train machine learning and generative AI models [1]. Hosting large datasets is moderately expensive - and it's full of NSFW stuff, so he's probably not having his employer or his college host it. Easy for someone on a six-figure salary, difficult for a person on $12k/year.
Also, I thought these lesswrong folks were all about "effective altruism" and "earning to give" and that stuff.
Hosting large datasets can be expensive but the hosting for the danbooru datasets was not.
It's "only" a few terabytes in size. A previous release was 3.4TB, so the latest is probably some hundreds of GB, to a TB~, in size larger.
The download was hosted on a hetzner IP, which is a provider known for cheap servers. You can pay them $50/m for a server with "unmetered" 1gigabit up/down network + 16TB of disks.
$600 a year would not be difficult.
