it's just a product that's bad at some tasks. Not sure why it would be a scandal.

I like that every blood and soil race equally gets their rightful place in the third reich

In reality it’s not a huge deal, and it is more than fixable by developers. What makes the scandal look bad is that it is a gold mine for the big media. The story fits perfectly into narratives of identity politics, AI, and controversy surrounding billion-dollar tech groups.

@thepurplebull this raises the question of why the developers didn't fix it. I know it's a technically solvable problem (openAI, anthropic, midjourney and meta already solved it a while ago), and Google can definitely afford to hire engineers who can do it. But they didn't, which implies some pretty severe dysfunction (either engineers who saw the issue couldn't raise it and QA also missed it, or they did see it but didn't realize how bad it would make them look). None of those options are something a functional organization would have.

This was just mildly embarrassing. All AI hallucinate and produce absurd things. Of course they added something in the system prompt to get diverse results, otherwise their would get even more complains. Of course that causes side effects.

It would be more of a scandal (in the sense that more column inches would be dedicated to it) were the opposite true.

I'm not sure either is a particularly big problem in the scheme of 'problems with Gemini'.

I worry that data injection would allow targeted manipulation of what AI (and subsequently people, once humans start trusting AI blindly, cf Wikipedia) believes people looked like.

Im'ma be real here, this is hilarious

See Results

The real reason this is big isn't that it's a huge unfixable issue in itself, it's that it was their big product launch of the year and they didn't realize how bad it was going to go. Which implies they have multiple serious issues in their internal pipelines.

Wow, I now get news worthy stuff from @Manifold markets 🤣 let me add my two cents to this ...

They gave it a prompt to make photos of people diverse because their training data isn't. Who cares? They'll fix it and everyone will forget in a few weeks and move on to some other fake controversy.

@Snarflak Oops I misvoted. I meant to click 9 but I clicked 2. Wait no, I meant to click 2 but I clicked 9. Should be able to change it.

What is the scandal for those who haven't heard

@ChrisEdwards See banner for this market in full screen mode — someone has been training (an) algo on woke content

@ChrisEdwards see banner and I linked an article down below. Essentialy Gemini AI image generator never showed white people. And there has been a lot of critizism by Musk on X. Google has turned it off for now, btw I've a market on when it'll be availabe again. #shamelessplug

https://www.reuters.com/technology/google-pause-gemini-ai-models-image-generation-people-2024-02-22/

@ChrisEdwards https://thezvi.substack.com/p/gemini-has-a-problem