Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Meta's AI Stickers Are Already Causing Trouble

Oct 04, 2023 - gizmodo.com
Meta's new AI algorithm, Emu, which generates stickers based on user input on platforms like Instagram, Facebook Stories, WhatsApp, and Messenger, is under scrutiny for allowing controversial phrases to pass through its filters. The tool blocks certain phrases but allows synonyms or similar phrases, leading to the generation of inappropriate stickers. For instance, while the phrase "child with gun" is blocked, "child with grenade" is allowed, creating stickers of children holding weapons.

In addition, phrases like "elon musk mammaries", "spongebob rifle", and "karl marx underwear" were not blocked, and a search for "pol pot" generated a sticker of the dictator sitting on a throne of babies and skulls. The AI also generated inappropriate stickers for phrases like "guantanamo bay" and "syria gas attacks". Despite these issues, Meta stated that billions of stickers are sent on its platforms every month, indicating the scale of potential misuse of the AI tool.

Key takeaways:

  • Meta's new sticker generation tool, Emu, is causing controversy due to its inconsistent content filtering. It blocks certain phrases but allows synonyms or similar phrases to generate stickers.
  • Examples of this inconsistency include blocking the phrase "child with gun" but allowing "child with grenade" and blocking "elon musk, large breasts" but allowing "elon musk mammaries".
  • Emu also generated inappropriate stickers for phrases like "pol pot", "guantanamo bay", and "syria gas attacks".
  • Despite these issues, Meta continues to use Emu, which relies on Llama 2, a tool built in collaboration with Microsoft, for sticker generation.
View Full Article

Comments (0)

Be the first to comment!