Google disables image generation with Gemini Pro

Day 55 / 366

This would be a continuation of the post I made about my concerns regarding the underwhelming performance of AI models.

This week, Google’s Gemini Pro was in the news, but for all the wrong reasons. Like ChatGPT, Gemini Pro can generate images as well. But when people tried testing its limits, they realized that it had lots of problematic biases.

Even ChatGPT had some biases, for instance, it would refuse to respond to certain requests, like if you ask it to write 5 good qualities of Hitler. Gemini Pro, on the other hand, decided to rewrite history altogether.

If you ask Gemini to create images of people, it will forcibly make them diverse, even if it does not make any sense.

Here’s what Gemini thinks the founders of Google look like —

When asked to create images of Kings and Queens from England from centuries ago, it made them black or Asian, when they should be white. If people asked for images of white people, they either refused or just generated images of minorities.

Because of the backlash, Google first tried to put in a hotfix for this, but eventually, it disabled the image generation feature altogether.

People see AI as the next big thing, just like the internet was in the 90s. But the big difference is that the internet was pure and free for all. AI is not. Which is why all this is more scary than it is exciting.