Google CEO says Gemini AI diversity errors are ‘completely unacceptable’::In an internal memo, Google CEO Sundar Pichai told employees that the historically inaccurate photos generated by the company’s Gemini AI were ‘completely unacceptable.’

all 22 comments

sorted by: hot top controversial new old
[–] 47 points 2 years ago (2 children)

if you find QA people do to QA and pay them to do QA this won’t happen.

but that’s not cost saving, is it?

  • source
  • hideshow 4 child comments
  • [–] 6 points 2 years ago (1 child)

    They had QA people, this was actually intended behavior

  • source
  • parent
  • hideshow 2 child comments
  • [–] 9 points 2 years ago* (1 child)

    no? alphabet lost $90bil in value over this, why on earth would an entity choose to make a design decision that breaks investor confidence?

    the initial choice to make the AI “diverse,” sure, that’s intentional. but the method by which it was done had unintentional consequences, and that is the problem which QA would have fixed.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 13 points 2 years ago (1 child)

    They got the result they wanted, they just didn't get the response they wanted from the public. There's no way to QA this

  • source
  • parent
  • hideshow 2 child comments
  • [–] 8 points 2 years ago* (last edited 2 years ago) (1 child)

    there’s no way to QA public response

    https://en.m.wikipedia.org/wiki/Focus_group

    there are literally dozens of ways marketers have come up with to do exactly what you are claiming is impossible 🙃

  • source
  • parent
  • hideshow 2 child comments
  • [–] 25 points 2 years ago (2 children)

    Well, Americans think Jesus was white with blue eyes. So it was just being human.

  • source
  • hideshow 4 child comments
  • [–] 3 points 2 years ago (2 children)
  • [–] 16 points 2 years ago

    I shrug at the whole thing...we spent a decade bitching about real world data being racist.

    When we put in fake data, we got fake results.

    Queue outrage.

  • source
  • [–] 11 points 2 years ago

    I'm not sure if the goal of txt2img models is supposed to be historically accurate photos....

  • source
  • [–] 8 points 2 years ago

    Faux outrage from white nationalists who are hopped up on replacement theory algorithms. Nothing to see here.

  • source
  • [–] 4 points 2 years ago (1 child)

    We want diversity... Wait no not like that!

  • source
  • hideshow 2 child comments
  • [–] 2 points 2 years ago (1 child)

    Yep. Black Nazis are not weirder than black vikings or black people in Arthurian Britain. Or European main characters in samurai-themed movies, God forbid.

    Anyway, an image generation tool should be orthogonal to this, it generates what's asked from it.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 1 point 2 years ago (1 child)

    I'm not disagreeing, just providing (I hope) interesting context.

    Black Nazis

    At the time, there were a few Afro-Germans, obviously hated by the nazis. A few were children of individual travelers, such as diplomats, like Hans-Jürgen Massaquoi. More were children of French Colonial soldiers. They were denigrated as "Rhineland bastards" by racists/nationalists even before the nazis rose. Obviously, they were prosecuted; subjected to forced sterilization. Perhaps surprisingly, they were never systematically murdered.

    There could not be such a thing as a black nazi. However, a few "mixed race" children (African and Asian) were conscripted into the Wehrmacht.

    One should also bear in mind that the term Aryan existed before the nazis and was used by white supremacists in the US even pre WW1. The nazis stopped using it in the last years of their rule.

    black vikings

    Black vikings are historically quite possible. In Western Europe and the English-speaking world, the vikings are known as raiders and conquerors. But they were also merchants and mercenaries. An archeological dig, near today's Stockholm, found a viking age Buddha statue from India. "Vikings" served as the Varangian Guard in today's Istanbul. People also travelled the other way, of course, such as notably Ahmad ibn Fadlan.

    There's nothing particularly implausible about a dark-skinned African travelling north to the Bosporus, meeting vikings, and travelling with them to their homeland to go on raids. It's not like we would expect to know if this happened once or twice during the viking age.

    Arthurian Britain

    I don't know much about this period. It was basically just after the Roman presence had ended. The Roman Empire included northern Africa, so thinking of black people in the Roman Legion makes perfect sense. After the decline/fall of the empire and the disruption of trade routes, the presence of dark skinned people in Britain becomes rather less plausible.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 2 points 2 years ago

    This is interesting, I agree, but, first,

    1 - I know that, but the images in question had SS symbols on helmets;

    2 - I know that too, and it's likely to have happened once or twice or a dozen times, but I think no genetic or archeological trace of something like this definitely happening has yet been found;

    3 - now here I think I've read about some genetic traces found even;

    ... and, second, it's more about something being normal as your first association.

    Say, when you think of a universally gifted person, Dolph Lundgren won't quite fit your stereotype.

  • source
  • parent
  • [–] [B] 1 point 2 years ago

    This is the best summary I could come up with:


    The historically inaccurate images and text generated by Google’s Gemini AI have “offended our users and shown bias,” CEO Sundar Pichai told employees in an internal memo obtained by The Verge.

    Last week, Google paused Gemini’s ability to generate images after it was widely discovered that the model generated racially diverse, Nazi-era German soldiers, US Founding Fathers who were non-white, and even inaccurately portrayed the races of Google’s own co-founders.

    While Google has since apologized for “missing the mark” and said it’s working to re-enable image generation in the coming weeks, Tuesday’s memo is the first time the CEO has widely addressed the controversy.

    In the memo, which was first reported by Semafor, Pichai says the company has “been working around the clock” to address “problematic text and image responses in the Gemini app.” He doesn’t say that Google has fixed the problem.

    “No Al is perfect, especially at this emerging stage of the industry’s development, but we know the bar is high for us and we will keep at it for however long it takes,” he writes.

    You can read Sundar Pichai’s full memo to Google employees below:


    The original article contains 189 words, the summary contains 189 words. Saved 0%. I'm a bot and I'm open source!

  • source
  • That’s what you consider unacceptable about it? Really? Just that one thing?

  • source