ginigen/Dokdo · about

From the moment we launched the Dokdo demo service, I observed a small but significant number of users generating content that dramatically deviates from standard social norms. These attempts involved deeply inappropriate, non-consensual, and horrifically offensive content spanning sexual, gore, and horror domains. Some instances even extended to terrorism-related materials, such as detailed scenarios depicting assassination attempts on political leaders in ongoing conflict zones.
Through monitoring, I've identified what appears to be a single user generating thousands of videos with similar imagery and prompts. The volume and persistence of these attempts are notable. While our current prompt filtering mechanisms provide some defense, they are not comprehensive. These users actively and creatively attempt to circumvent existing filters, demonstrating a persistent intent to generate harmful content.
The scale and sophistication of these attempts are so substantial that the collected data could potentially form the basis of a comprehensive academic research paper. Although no system can be perfect, we are seriously considering implementing individual user authentication as a potential mitigation strategy.
Despite these challenges, our commitment remains clear: we must continue to progress, refine our systems, and maintain a safe, responsible platform.