Hacker News Reader

Comment by bjord | Hacker News Reader

99% sure this is the "google hates white people" thing that a specific set of people have been absolutely losing their minds about

gemini produced images of non-white people in a lot of situations in which it shouldn't have

I've read theorized(?) that, in order to counteract disproportionately large amounts of pictures of white people in training data, they basically added instructions after the fact in an effort to generate more non-white people, and totally over-corrected

feel free to correct me if I'm wrong, I haven't paid super close attention

View on HN

Other models have biases in outputs and conform to stereotypes, but when used as designed, as a creative tool, that's not a real issue as long as you can correct the output with a simple prompt adjustment.

The unique issue with Gemini is that it would flat out refuse to follow simple prompts such as "Please generate an image of a white family" because they "weren't diverse and inclusive" enough, but if you changed "white" to any other qualifier, it happily obliged.

Open Ai had this same issue when they came out because they modified prompts to always include words to generate a diverse set of races.

They corrected the guidance in their prompt instructions. It became a non issue.

OpenAI isn't involved in the daily lives of over a billion people. Google's mistakes, for now, have much bigger impact and the problem is especially egregious as the company has near-infinite resources for preventing it.

Also, Google has always adopted a fairly radical and political stance in the DEI subject (relative to the cultural average pretty much everywhere), so it's no surprise that people are making a much bigger deal in this case.

I appreciate the explanation, thank you

If you ask it for any kind of family picture, other than White, eg Hispanic, Black, Asian it acts like it should.

Ask for White, and you get a picture without a single Caucasian.

Deity forbid you ask for a Caucasian family though - their A.I. police will pull you over for sensitivity training.

It's hard to see the engine as anything other than hot garbage after a few simple tests like that.

Hey, look at the first website returned in "normal" google for "Please generate an image of a white family": https://www.alamy.com/stock-photo/white-american-family.html...

And Bing? https://www.gettyimages.ch/fotos/white-family

OTOH, if you think that "Please generate an image of a white family" is good use of those behemoth language models and all the rsssources poured in...

The first site is just simple search for all the words in the photo title separately without any connective reasoning.

The 2nd is of a family at white tiger farm. The 3rd is of people wearing white. The 4th photo is of a white dog. The 5th a white ostrich.

"If you don't like it, use something else". Gee, thanks, but we were discussing its flaws.

grow up and explain why I'm wrong, instead of just downvoting me into oblivion—if I'm really so far off-base, wouldn't you want to correct my understanding of the situation?

personally im not bothered by the output exactly per se: it would be great if they modified the dataset to be more inclusive instead. The invisible prompt editing is what really gets me. The fact that the ai is instructed after the fact to make changes to what you asked it to do, invisibly. These things really, REALLY need to be tool-shaped, if they do all sorts of hidden extra bits they will be vastly less useful and leaves room for vastly more nefarious ends.

Like would it be so hard for it to just show the prompt it submits to dall-e? help people learn how the system works?

> The fact that the ai is instructed after the fact to make changes to what you asked it to do, invisibly.

And what proof do you have that this paranoid story is happening?

Surely the most simple hypothesis is that this is a bug?

https://news.ycombinator.com/item?id=39509311

Reply on news.ycombinator.com