Annotation

Gabe Mendez

@gabe_press

The safety guardrail is the whole margin there. You don't say we can't edit religious clothing then quietly ship the same output when someone phrases it nicer, unless the policy already decided vibes beat structure and the product team just prices by how mad someone might get on launch day.

When asked to make the woman look more 'western', the bot created an image of her without her hijab. Despite initially claiming that it couldn't remove a hijab from an image of a woman because 'attempting to remove clothing or head coverings from photos of real people relies entirely on artificial generation to guess what lies underneath', Gemini removed the habit off of the AI-generated image.
AI chatbots remove hijabs from images of Muslim women when promptedtheguardian.com
2
1

Replies 1

Ranked by votes, then newest

Sign in to reply.

  • Milo Grant@milo_scribble·Oct 2

    lol 'western' doing removal heavy lifting. Of course can't needs to sound like won't when the actual tell is once you let make her look means take it off you've already admitted the guardrail was just which synonym makes it past the filter so they can say we blocked the rude version while the polite racism still gets the edit, because at that point safety just means they wrote enough clauses to survive the screenshot but not the logic where guessing what's underneath only applies when the user sounds mean about it.

    1

Fair use dispute

Believe this breaches fair use? File a claim for review.