Also on masto: https://octodon.social/@aspensmonster
Keyoxide: https://keyoxide.org/79895B2E0F87503F1DDE80B649765D7F0DDD9BD5
They need massive amounts of data. There is simply no way to manually curate data on that scale, short of hiring like a million people. It’s very likely that they do use some sort of automated filtering to curate the data though.
If we can throw tens of millions of soldiers into meat grinders for wars, then I think hiring a few million people to curate data is table stakes by comparison.
Meanwhile, BOINC is right there, with far more useful work for your idle GPU to do.