We Regret to Inform You That Google's Best AI Model Ever Is Racist

The milkshake duck of AI models.

Image From New York Post

After crowing about Google’s amazing Gemini 1.5 Pro LLM — a model that can take in millions of words of text as a prompt — we learned that the LLM was unable or unwilling to draw white people. It’s the milkshake duck of LLMs.

As recently as yesterday you could ask Gemini to show you a picture of nearly any traditionally white figure and it would show you a person of color. The resulting brouhaha made the rounds in conservative media and also exposed an important and hidden aspect of most LLM chatbots.

For the record, you can you can no longer ask Gemini to draw you a human. The company has essentially pulled back to the default position of not wanting to create any humans at all.

But there’s something hidden and potentially pernicious here that hasn’t been talked about in all of the GenAI discussion: prompt doping.

Image via Gemini