Alarming discovery: Meta's AI sticker tool generates disturbing and controversial images
Meta's AI sticker tool raises alarm due to users creating explicit and controversial stickers. Content control issues emerge, prompting a limited rollout by Meta to address concerns.
artificial intelligence
Highlights
- Users create controversial and explicit stickers with Meta's new AI tool
- Some users find ways to bypass content restrictions, highlighting concerns about content control
- Meta responds by cautiously rolling out the feature to address misuse before a wider release
In a startling revelation, early user tests of Meta's groundbreaking AI-generated sticker tool have uncovered some rather dubious and amusing creations. Users on X (Formerly Twitter), including @Pioldes, who gained access to this tool on Facebook Messenger, have produced a range of inappropriate sticker images that have raised eyebrows.
Unintended consequences
Among these creations are sticker images depicting child soldiers, gun-toting Nintendo characters, Mickey Mouse in compromising situations, and even nude illustrations of Canada's Prime Minister, Justin Trudeau.
found out that facebook messenger has ai generated stickers now and I don't think anyone involved has thought anything through pic.twitter.com/co987cRhyu
— podesbiens.bsky.social (@Pioldes) October 3, 2023
The tool has also been found to be rather liberal in its application of certain elements, such as adding breasts to virtually anything, as seen in busty depictions of Sonic the Hedgehog and Karl Marx. One particularly peculiar sticker shows a woman breastfeeding Pikachu.
I'm sure this is fine pic.twitter.com/XbzT7RgtlQ
— podesbiens.bsky.social (@Pioldes) October 4, 2023
The challenge of content control
Meta's AI-generated chat stickers, powered by the Llama 2 large language model, were recently unveiled at the Connect event. They promise users the ability to create high-quality stickers within seconds using text-based prompts.
These stickers are gradually rolling out to select English language users across various Meta platforms, including Facebook Stories, Instagram Stories, DMs, Messenger, and WhatsApp. However, there have been concerns regarding content control, as certain blocked words are still slipping through the cracks.
Users have reported that inappropriate content can still be generated through typos or descriptions of restricted words, with some prompts generating problematic images without additional descriptors.
Meta's approach
The emergence of such issues is not uncommon when launching AI tools, prompting Meta to pursue a limited rollout of the AI-generated sticker feature. This strategy allows the company to address and rectify instances of misuse before making the feature available to a broader audience.
In short, Meta's AI-generated sticker tool, while innovative, has raised concerns over its content control measures and the unintended creations it can produce.
As Meta works to refine this technology, the journey towards responsible AI usage continues.
COMMENTS 0