The concept of adblock isn’t useless though, the client can always deny rendering a portion of the page. The final control of rendering is in the browser.
Glaze is useless because it only serves to teach AI how to be more agnostic to countermeasures. There is no mathematically secure way to poison an image without ruining the image for human eyes too.
In other words, a sufficiently smart adblocker will always block every ad, and a sufficiently smart AI model will always bypass Glaze.
I think I’m thinking further ahead than you are. Right now Glaze works, but the end result is that it becomes useless. There is no “keeping on top of it” in the long run - the entire concept is a mathematical dead end.
Adblock is a mathematically sound principle though.
Glaze is a mathematical dead end due to the Analog Gap. As long as the image is visually unchanged to humans, an algorithm is mathematically capable of replicating it.
This is the same reason why DRM is fundamentally pointless - if a person can play the game, the DRM can be circumvented. We’ve seen this even with extraordinarily sophisticated DRM (eg Denuvo). Crackers can get through it just fine, and often with far less effort than it took to implement the DRM in the first place.
Adblock is fundamentally different however. It operates by changing the client side rendering of a web page. Since the rendering code is owned by the end user, you can modify the final web page in any way you want and the owner of the website can’t do a damn thing about it. There has been work lately on signing a page to verify it hasn’t been tampered with, but a client browser can simply forge that response.
The concept of adblock isn’t useless though, the client can always deny rendering a portion of the page. The final control of rendering is in the browser.
Glaze is useless because it only serves to teach AI how to be more agnostic to countermeasures. There is no mathematically secure way to poison an image without ruining the image for human eyes too.
In other words, a sufficiently smart adblocker will always block every ad, and a sufficiently smart AI model will always bypass Glaze.
The concept of glaze, as in poisoning ai, isn’t useless either.
I think you don’t understand what I am saying. I don’t compare addblock with a smart ai, I compare addblock with glaze.
You said that because something can be trained on glaze it makes it useless and I disagree with that because glaze can react to that adoption.
Like it is happening with addblock vs YouTube.
I think I’m thinking further ahead than you are. Right now Glaze works, but the end result is that it becomes useless. There is no “keeping on top of it” in the long run - the entire concept is a mathematical dead end.
Adblock is a mathematically sound principle though.
Why is it a mathematical dead end? And why is addblock not?
Glaze is a mathematical dead end due to the Analog Gap. As long as the image is visually unchanged to humans, an algorithm is mathematically capable of replicating it.
This is the same reason why DRM is fundamentally pointless - if a person can play the game, the DRM can be circumvented. We’ve seen this even with extraordinarily sophisticated DRM (eg Denuvo). Crackers can get through it just fine, and often with far less effort than it took to implement the DRM in the first place.
Adblock is fundamentally different however. It operates by changing the client side rendering of a web page. Since the rendering code is owned by the end user, you can modify the final web page in any way you want and the owner of the website can’t do a damn thing about it. There has been work lately on signing a page to verify it hasn’t been tampered with, but a client browser can simply forge that response.