This isn’t a corner case it injects words like inclusive or diverse into the prompt right in front of you. “A German family in 1820” because “a diverse series of German families”

And it ignores male gendering. People were posting pictures of women when the prompt asked for a "king".

Though technically it would be ok if it was an Korean or Chinese one, because the word in those languages for "King" is not gendered.

Have fun with that AI.

And, in Turkish, we have two gendered (Slavic origin for either gender) and one genderless (Mongolian origin) words for "king".

They were clearly willing to spend time adjusting the knobs in order to create the situation we see now.

The famous 80/80 rule

The first 80% of a software project takes 80% of the time. The last 20% of a software project takes 80% of the time. And if you prove this false, you're a better engineer than me!

> And if you prove this false, you're a better engineer than me!

Probably cheating somehow!

That’s only 60% over budget. What takes up the last 40%? Agile project management scrum standup virtual meetings?

40% is taken by managers forwarding e-mails among themselves and generating unnecessary meetings.

Or, how Gemini would say...

40% is taken by A DIVERSE SET of managers forwarding e-mails among themselves and generating unnecessary meetings.

As a low paid, part-time intern I finished a project with another low paid, part-time intern. We were supposed to have some FT people on with us, but that never happened so we just kept working. We only had one weekly meeting and it was a challenging project. When it was done we had a lunch/meeting to celebrate and discuss how it went with the project people.

They commented that this was the first one that was both under budget and late and wondered how that could be. I volunteered that it was because "we are interns and don't get paid much and it was late because we spent a month waiting for the DBA to correct a mistake he made - we could have been done early."

There were shocked faces. Lunch was good and there were not many more questions.

None of these are "corner cases". The model was specifically RLHF'ed by Google's diversity initiatives to do this.

Do you think Google's diversity team expected it would generate black nazis?

> Do you think Google's diversity team expected it would generate black nazis?

Probably not, but that is precisely the point. They're stubbornly clinging to principles that are rooted in ideology and they're NOT really thinking about consequences to the marginalized and oppressed that their ideas will wreck, like insisting that if you're black your fate is X or if you're white your guilt is Y. To put it differently, they're perpetuating racism in the name of fighting it. And not just racism. They make assumptions of me as a gay man and of my woman colleage and tell everyone else at the company how to treat me.

Do you think no one internally thought to try this, but didn't see a problem with it because of their worldview?

> Do you think no one internally thought to try this

This is Google on one hand and the Internet on the other.

So probably not?

It's not difficult to notice that your images are excluding a specific race (which, ironically, most of the engineers building the thing are a part of).

I'd hazard a guess that the rate at which Google employees type "generate a white nazi" and the rate at which the general Internet does so differs.

It's clear there is a ban on generating white people, and only white people, when asked to do so directly. Which is clearly an intervention from the designers of this system. They clearly did this intentionally and live in such a padded echo chamber that they didn't see a problem with it. They thought they were "helping".

This is a debate between people who want AI to be reflective of reality vs. people who want AI to be reflective of their fantasies of how they wish the world was.

I feel like it's more of a debate about the extent of Google's adversarial testing.

"What should we do about black nazis?" is a pretty basic question.

If they'd thought about that at all, they wouldn't have walked this back so quickly, because they at least would have had a PR plan ready to go when this broke.

That they didn't indicates (a) their testing likely isn't adversarial enough & (b) they should likely fire their diversity team and hire one who does their job better.

Building it like this is one thing. If Google wants to, more power to them.

BUT... building it like this and having no plan of action for when people ask reasonable questions about why it was built this way? That's just not doing their job.

No. Let's ask those directly responsible and get an answer.

Won't happen.

They'll hide behind the corporate veil.

Nah, they just call you a racist

I don’t think they expected that exact thing framed in that exact way.

Do I think that the teams involved were institutionally incapable of considering that a plan to increase diversity in image outputs could have negative consequences? Yes, that seems pretty clear to me. The dangers of doing weird racecraft on the backend should have been obvious.

I suspect that Google's solution to this mess will be to retain said racecrafting except in negative contexts. That is, `swedish couples from the 1840s` will continue to produce hordes of DEI-compliant images, but `ku klux klansmen` or `nazi stormtroopers` will adhere to the highest standard of historical accuracy.