cannedtuna@lemmy.world to Microblog Memes@lemmy.worldEnglish · 1 month agoI wonder if the AI would actually recognize the images. Doubtfullemmy.worldimagemessage-square76linkfedilinkarrow-up1680arrow-down125file-text
arrow-up1655arrow-down1imageI wonder if the AI would actually recognize the images. Doubtfullemmy.worldcannedtuna@lemmy.world to Microblog Memes@lemmy.worldEnglish · 1 month agomessage-square76linkfedilinkfile-text
minus-squarenomad@infosec.publinkfedilinkEnglisharrow-up2arrow-down2·1 month agoJup, people need to read up on adversarial training. I’ve seen examples of lightly sprinkled noise over pictures that completely breaks any object recognition and you wouldn’t see any difference in the picture with a human eye.
minus-squarehuppakee@lemmy.worldlinkfedilinkEnglisharrow-up2·1 month agoI think if you have access to the digital content it is much easier to trick the machines, here is a cool example of that: https://www.dezeen.com/2026/07/31/decoy-font-ai-mixfont/ Hardest is when you don’t and don’t even know what the data is exactly, how much footage there is of you, where camera’s are etc.
Jup, people need to read up on adversarial training. I’ve seen examples of lightly sprinkled noise over pictures that completely breaks any object recognition and you wouldn’t see any difference in the picture with a human eye.
I think if you have access to the digital content it is much easier to trick the machines, here is a cool example of that: https://www.dezeen.com/2026/07/31/decoy-font-ai-mixfont/
Hardest is when you don’t and don’t even know what the data is exactly, how much footage there is of you, where camera’s are etc.