Meta's AI-generated stickers spark controversy over inappropriate content
Meta's new AI-generated sticker tool for Facebook Messenger and Instagram has stirred up some controversy. Users have managed to create a range of inappropriate images, such as child soldiers, gun-wielding Nintendo characters, and explicit illustrations of public figures. The AI-generated sticker tool, powered by Meta's Llama 2 large language model, was announced at the recent Connect event. It is currently rolling out to "select English language users" across Facebook, Instagram, and WhatsApp.
These stickers can be deemed inappropriate
Users have been able to get around the filters
Despite Meta's attempts to block offensive content, users have found ways to get around the filters by using typos or descriptions of restricted words. Some prompts even generated problematic images. A variety of inappropriate sticker images were also created featuring copyrighted characters like Mickey Mouse and Winnie the Pooh engaging in activities such as smoking marijuana or holding firearms. Artist Pier-Olivier Desbiens shared some of these controversial sticker images on social media, quickly gaining thousands of views and comments.
Meta says its models are trained on safety guidelines
"We're training our models on safety and responsibility guidelines," reads Meta's blog. "Teaching the models guidelines means they are less likely to share responses that are potentially harmful or inappropriate for all ages on our apps." "In keeping with our commitment to responsible AI, we also stress test our products to improve safety performance and regularly collaborate with policymakers, experts in academia and civil society, and others in our industry to advance the responsible use of this technology," it added.
Ensuring responsible use of AI-generated content
As generative AI systems continue to evolve, it becomes important for companies like Meta to address the potential for inappropriate content creation and distribution. By refining filters, monitoring user feedback, and improving AI-generated features, tech companies can work toward ensuring more responsible use of AI-generated content and minimizing the risks associated with these powerful tools. That could be why Meta's AI sticker generation tool is having a limited rollout.