> How can they draw something without having seen the exact thing before? There has, in fact, been CSAM found in a version of the training set that was used to

> How can they draw something without having seen the exact thing before? There has, in fact, been CSAM found in a version of the training set that was used to make Stable Diffusion 1.5, for example. https://www.theregister.com/2023/12/20/csam_laion_dataset/ https://www.theregister.com/2023/12/20/csam_laion_dataset/