And we've got the first copyright issue on text to image AIs. Some text to image AIs are being trained on datasets provided by LAION e.V., which is a German non profit organization.
These datasets do provide deep links to images scattered all across the net, as well as descriptions on what's being depicted on it in English. The AI companies/projects then download the images, pair it with the descriptions and train their networks on that.
Now a German stock photographer named Robert Kneschke is suing LAION e.V., and wants them to remove his images of the training data of those models.
LAION itself states they are just providing the deep link lists with a notice that everybody should check it for issues before using it commercially. Kneschke on the other hand thinks that the non profit organizations are just there in reality to divert legal trouble from the AI companies itself, telling that e.g. the founder of LAION is also the head of machine learning operations at Stability AI.
A stock photographer sues LAION, while the non-profit AI association argues having complied with applicable law. The dispute raises questions beyond the case.
www.heise.de
apnews.com
/cdn.vox-cdn.com/uploads/chorus_asset/file/24676983/IMG_20230323_133713_volcano.jpg)
