A "prompt testing exercise" in India "involving Meta AI, Gemini, Co-pilot and Adobe Firefly" showed "Meta AI's text-to-image generation feature is being weaponised to create harmful AI images targeting Muslims in India by reinforcing negative stereotypes."

I'm reminded that we don't have the "real" LLMs. All the companies have LLMs that can do things that the public LLMs we have access to can't, because of the "guardrail adding" effort. So any time you say, "LLMs can't do X", well, maybe they can, just not the public ones you're using.

Exclusive: Meta AI's text-to-image feature weaponised in India to generate harmful imagery

#solidstatelife #ai #genai #llms #computervision