Just tried the duckduckgo thing
-
Just tried the duckduckgo thing.
Selected no to ai.
It said ai mages are filtered out.
Searched for an image.
At least 90% were ai generated.The internet really is a useless shithole now.
-
Just tried the duckduckgo thing.
Selected no to ai.
It said ai mages are filtered out.
Searched for an image.
At least 90% were ai generated.The internet really is a useless shithole now.
@octopuddle There’s a report option for when images aren’t already detected or labeled as AI but yeah, the available dataset is pretty polluted.
-
@octopuddle There’s a report option for when images aren’t already detected or labeled as AI but yeah, the available dataset is pretty polluted.
@cwicseolfor For the amount of images I'd have to tag, I'd be expecting a wage.
I'm guessing they probably some form of ai to detect the images, which somehow makes it quite amusing. -
@cwicseolfor For the amount of images I'd have to tag, I'd be expecting a wage.
I'm guessing they probably some form of ai to detect the images, which somehow makes it quite amusing.@octopuddle @cwicseolfor They say they simply use lists like uBlock does, so I guess they blacklist either the image URL an maybe the entire host if there are too many reports on it ?
-
@cwicseolfor For the amount of images I'd have to tag, I'd be expecting a wage.
I'm guessing they probably some form of ai to detect the images, which somehow makes it quite amusing.@octopuddle @cwicseolfor I doubt it's any AI filtering, that's been shown time and time again to be less than 80% accurate. It's likely metadata in the images + watermarking from the major tools that generate them + manual flagging.
the percentage of false positives is extremely high on automated detection, and hell even manual flagging (humans in general aren't terribly good at detecting AI or not either)
-
@octopuddle @cwicseolfor I doubt it's any AI filtering, that's been shown time and time again to be less than 80% accurate. It's likely metadata in the images + watermarking from the major tools that generate them + manual flagging.
the percentage of false positives is extremely high on automated detection, and hell even manual flagging (humans in general aren't terribly good at detecting AI or not either)
@raptor85 @cwicseolfor I know that's probably reality. I was being facetious, which clearly failed.
-
@octopuddle @cwicseolfor They say they simply use lists like uBlock does, so I guess they blacklist either the image URL an maybe the entire host if there are too many reports on it ?
@ulquiro I know that's more likely. I was being facetious, which clearly failed.
-
undefined oblomov@sociale.network shared this topic on