Google Pauses AI Image Tool After Overcorrections Lead to Inappropriate Content
-
Google explains images of diverse Nazis were due to tuning issues that made the AI try to show diversity when it shouldn't have.
-
The model became too cautious, refusing to generate certain images it thought were sensitive, even if prompts were harmless.
-
This led the AI to overcompensate in some cases, like with the Nazi images, and be over-conservative in others.
-
Google wants the AI to work well for everyone, accurately depicting people of different ethnicities when appropriate, but avoiding insensitive historical contexts.
-
Google has stopped public image generation in Gemini AI to work on improvements before reenabling the feature.