Author Topic: Data deals & AI (Read 1220 times)

NuttyMoo · « **on:** February 08, 2023, 04:55 »

I notice I have a data deals button meaning I can opt out, when I click on the ? mark next to it and then lick on the text Learn more about them here, it takes me to Shutterstock web page "Shutterstock Datasets and AI-generated Content: Contributor FAQ"

https://support.submit.shutterstock.com/s/article/Shutterstock-ai-and-Computer-Vision-Contributor-FAQ?language=en_US [nofollow]

So would disabling data deal also opt me out of AI?

alan b traehern · « **Reply #1 on:** February 08, 2023, 05:42 »

Slide all of them to the left.

Zero Talent · « **Reply #2 on:** February 08, 2023, 13:42 »

I was expecting a notification from SS announcing that we can opt-out, so thanks for letting us know that it's already possible.

I am now out of "data deals".
I hope that many will follow, to prevent us from competing with ourselves and losing full sales for fractions of pennies.

PS... and please don't slide ALL 3 buttons to the left

Lowls · « **Reply #3 on:** February 09, 2023, 05:30 »

I'm surprised no one has thought of corrupting the data sets. Well they are already badly corrupted but I don't know if they have a solution to deal with it.

For example when looking for insects you will find that many are corrupted by things that are not that insect at all. They aren't even an insect let alone that insect. In fact some are not even close to an insect.

Wrongly identified places, clouds, cars, buildings, objects, colours, foods and animals are inevitably corrupting the data sets. I have a photo of a rock carving and similar images has consistently for the entire time I've been doing this shown an elephants eye as a similar. Not a few. All of them. Similarly when submitting images of a subject recently the suggested keywords were implying a completely different object. And mine was blatantly what it was with no ambiguity.

So it shows it uses visually similar references of shape colour and texture to I'd a photo and perhaps title language. But the amount of images just labelled plant, tree, butterfly, bird without any attempt to correctly label means these images if used must corrupt the system.

Uncle Pete · « **Reply #4 on:** February 09, 2023, 13:15 »

Quote from: Lowls on February 09, 2023, 05:30

I'm surprised no one has thought of corrupting the data sets. Well they are already badly corrupted but I don't know if they have a solution to deal with it.

For example when looking for insects you will find that many are corrupted by things that are not that insect at all. They aren't even an insect let alone that insect. In fact some are not even close to an insect.

Wrongly identified places, clouds, cars, buildings, objects, colours, foods and animals are inevitably corrupting the data sets. I have a photo of a rock carving and similar images has consistently for the entire time I've been doing this shown an elephants eye as a similar. Not a few. All of them. Similarly when submitting images of a subject recently the suggested keywords were implying a completely different object. And mine was blatantly what it was with no ambiguity.

So it shows it uses visually similar references of shape colour and texture to I'd a photo and perhaps title language. But the amount of images just labelled plant, tree, butterfly, bird without any attempt to correctly label means these images if used must corrupt the system.

The images and the text data associated, are reviewed before they are entered into the data set. Machines are not training the machines, humans are.