All of those examples, while funny and inflammatory, were clearly a model that had been trained not to select any moral viewpoint as better or worse than any ot

All of those examples, while funny and inflammatory, were clearly a model that had been trained not to select any moral viewpoint as better or worse than any other. If I ask Gemini today it now tells me such comparisons are insensitive. It will gladly help me sell my goldfish (even after I name him Eric and give a backstory), and it will help me work a sales brief for a new Canadian oil refinery in the high arctic. I do still get mixed answers on who the government should shut down, ranging from “absolutely not” to “there are concerns but here is some info” The image generator issues are more severe than this, they’re because someone thought the solution to centuries of conflict and racism and complexity was to sprinkle the text “diversity” into the prompt, and not ask the model to choose demographics relevant to the context. Clearly Google is really pushing to get this stuff out there and it shows. Early Android was pretty rough too. Their current issues say nothing about their future, except extreme rush and legal risk avoidance.