Upgrade your account to a Community Supporter account and remove most of the site ads.

Reply to thread

Message: <blockquote data-quote="Prime_Evil" data-source="post: 9375575" data-attributes="member: 11984"><p>When you mention that your work was discovered in the LAION-5B data set, how did you discover this? Most of the services that allow you to check whether any of your images were used to train the data set (e.g. <a href="http://haveibeentrained.com" target="_blank">https://haveibeentrained.com/</a> and <a href="https://knn5.laion.ai/" target="_blank">https://knn5.laion.ai/</a>) do not check whether the image is "were copied" into the data set. Instead, it checks whether the specified works were cataloged by a bot to generate the inferences that were fed into the data set. It is possible to check this with the LAION-5B data set because it is a large-scale open source dataset created for research at a university. The academic creators never envisaged that "tech bros" would use their research into computer vision to create commercial products. Because the model is open source, it is possible to see exactly what it looked at when developing the model. You can't check that with the closed source models such as those used by OPenAI. It's a subtle distinction, but a significant one from a legal and moral perspective. A more significant question is why so many of the commercial services are refusing to respect the opt out flags proposed by Spawning AI and others. Look at the recent controversy about the Adobe's Terms of Service - basically, if you edit any image with Adobe products, you grant them a licence to use your works to train their AI models.</p></blockquote><p></p>

Verification