The meta training surveillance leak is why i stopped pretending the cloud parts were fine
The meta monitoring engineers to train internal ai story, and then the leak exposing tens of thousands of records, landed for me less as a news item and more as the thing the selfhosted instinct has been right about all along. If the training method is surveillance then the data boundary is the whole game. I self host most things for exactly this reason. But i am going to be honest about the awkward corner, the heavy coding models are not something you realistically run on your own metal, and pretending oth
评论
?
参与讨论