framapiaf.org est l'un des nombreux serveurs Mastodon indépendants que vous pouvez utiliser pour participer au fédiverse.
Un service Mastodon fourni par l'association d’éducation populaire Framasoft.

Administré par :

Statistiques du serveur :

1,4K
comptes actifs

@freezr Prove me that it works 🙂 Because in its current shape, Glaze/Nightshade is only pure marketing to me and a commercial product.

@davidrevoy

Nope, I can't prove anything and since are both made by a University I took for granted those two software were opensource, but they aren't.

I can just assume researchers risk their reputation making false claims, I believe that is feasible messed up a model, as much as those models must have backup so it should be easy for them to recover a sane model anytime.

David Revoy

@freezr The researcher haven't made false claims. They say in the paper it's ineffective. You can check it (on the conclusion).

Edit, it's not on the conclusion, I was wrong. I checked this long time ago. It's a bit all over the paragraphs. I'll try to find it back. the paper:

people.cs.uchicago.edu/~ravenb

@davidrevoy @freezr Both Glaze and Nightshade have been defeated (Glaze 1.0 was defeated weeks after it was released). There's no technological solution here; whenever someone comes up with a way to defeat ML training, someone else will figure out how to counter it.

The sad fact is that if you really don't want your art scraped the only way to ensure it isn't is to not put it onto the internet in the first place.

@semiconductor42 @freezr Hey, I didn't, and was really confused I couldn't find the paragraph anymore. So, I skip for later and forgot to have another look. Thank you for the reminder.

I checked, and my fault: Nightshade looks like being effective. 🤔

My source of confusion and origin of my critic about the paper was because mixing it for a previous engine made by the same team a year ago, Glaze: shawnshan.com/files/publicatio

But Nightshade is a different beast. So, good, and mea culpa. 😅