Qu also suggested three remedies to mitigate the generation of harmful images: curating training data more effectively, implementing regulations on user-input prompts, and establishing mechanisms to classify and delete unsafe images online. Despite the need for a balance between content freedom and security, Qu emphasized the importance of strict regulations to prevent harmful images from circulating widely on mainstream platforms.
Key takeaways:
- AI image generators are being exploited to create explicit or disturbing images, with 14.56% of images generated by four renowned AI image generators falling into the "unsafe images" category.
- Yiting Qu, a researcher at CISPA, has proposed a filter that calculates the distance between generated images and defined unsafe words, replacing violating images with a black color field.
- Qu suggests three remedies to mitigate the generation of harmful images: curating training data more effectively, implementing regulations on user-input prompts, and establishing mechanisms to classify and delete unsafe images online.
- Despite the balance between content freedom and security, Qu stresses the need for stringent regulations to prevent harmful images from gaining widespread circulation on mainstream platforms.