In response, the LAION database has been taken offline for the removal of the illegal content. Stability AI, which developed the popular image generator Stable Diffusion, stated that it has introduced filters to prevent the creation of illegal content and that its models were trained only on a filtered subset of LAION-5B. However, an earlier version of Stable Diffusion, which can generate explicit content, is still available on the open web.
Key takeaways:
- The LAION-5B database, used to train AI image generators, was found to contain over 3,200 images of suspected child abuse, with over 1,000 confirmed as child sexual abuse material.
- The presence of these images in the dataset raises concerns that AI tools could be used to create new child abuse content.
- Stability AI, which uses the LAION database for its image generator Stable Diffusion, has introduced filters to prevent users from creating illegal content and only trains its models on a filtered subset of LAION-5B.
- LAION, the German non-profit behind the dataset, is temporarily removing its datasets from the internet to remove illegal content.