Hacker News Reader

Comment by mattzito | Hacker News Reader

“Misrepresenting reality” is an interesting phrase, considering the nature of what we are discussing - artificially generated imagery.

It’s really hard to get these things right: if you don’t attempt to influence the model at all, the nature of the imagery that these systems are being trained on skews towards stereotype, because a lot of our imagery is biased and stereotypical. It seems perfectly reasonable to say that generated imagery should attempt to not lean into stereotypes and show a diverse set of people.

In this case it fails because it is not using broader historical and social context and it is not nuanced enough to be flexible about how it obtains the diversity- if you asked it to generate some WW2 American soldiers, you could rightfully include other ethnicities and genders than just white men, but it would have to be specific about their roles, uniforms, etc.

(Note: I work at Google, but not on this, and just my opinions)

View on HN

> It seems perfectly reasonable to say that generated imagery should attempt to not lean into stereotypes and show a diverse set of people.

When stereotypes clash with historical facts, facts should win.

Hallucinating diversity where there was none simply sweeps historical failures under the rug.

If it wants to take a situation where diversity is possible and highlight that diversity, fine. But that seems a tall order for LLMs these days, as it's getting into historical comprehension.

>Hallucinating diversity where there was none simply sweeps historical failures under the rug.

Failures and successes. You can't get this thing to generate any white people at all, no matter how explicitly or implicitly you ask.

> You can't get this thing to generate any white people at all, no matter how explicitly or implicitly you ask

You sure about that mate?

https://imgur.com/IV4yUos

Idk, but watch this live demo: https://youtube.com/watch?v=69vx8ozQv-s He couldn't do it.

There could have been multiple versions of Gemini active at any given time. Or, A/B testing, or somehow they faked it to help Google out. Or maybe they fixed it already, less than 24 hours after hitting the press. But the current fix is to not do images at all.

You could have literally done the test yourself as I did only a day ago but instead link me to some Youtuber who according to Wikipedia:

> In 2023, The New York Times described Pool's podcast as "extreme right-wing", and Pool himself as "right-wing" and a "provocateur".

Which is kinda funny because the majority of his content is reading articles from places like the New York Times.

They're just straight up lying about him.

I think the root problem is assuming that these generated images are representations of anything.

Nobody should.

They’re literally semi-random graphic artifacts that we humans give 100% of the meaning to.

So you're saying whatever the model doesn't have to be tethered to reality at all? I wonder if you think the same for chatgpt. Do you think it should just make up whatever it wants when asked a question like "why does it rain?". After all, you can say the words generated are also semi-random sequence of letters that humans give meaning too.

I think going to a statistics based generator with the intention to take what you see as an accurate representation of reality is a non starter.

The model isn’t trying to replicate reality, it’s trying to minimize some error metric.

Sure it may be inspired by reality, but should never be considered an authority on reality.

And yes, the words an LLM write have no meaning. We assign meaning to the output. There was no intention behind them.

The fact that some models can perfectly recall _some_ information that appears frequently in the training data is a happy accident. Remember, transformers were initially designed for translation tasks.

> Do you think it should just make up whatever it wants when asked a question like "why does it rain?"

Always doing that would be preferable to the status quo, where it does it just often enough to do damage while retaining a veneer of credibility.

> They’re literally semi-random graphic artifacts that we humans give 100% of the meaning to.

They're graphic artifacts generated semi-randomly from a training set of human-created material.

That's not quite the same thing, as otherwise the "adjustment" here wouldn't have been considered by Google in the first place.

The fact that the training data is human curated arguably further removes the generations from representing reality (as we see here with this whole little controversy)

I think, with respect to the point I was making, they are the same thing.

But then if it simply reflected reality there also be no problem, right, because it’s a synthetically generated output. Like if instead of people it output animals, or it took representative data from actual sources to the question. In either case it should be “ok” because it’s generated? They might as well output planet of the apes or starship trooper bugs…

With emphasis on the "semi-". They are very good at following prompts, and so overplaying the "random" part is dishonest. When you ask it for something, and it follows your instructions except for injecting a bunch of biases for the things you haven't specified, it matters what those biases are.

Are they good at following prompts?

Unless I format my prompts very specifically, diffusion models are not good at following them. Even then I need to constantly tweak my prompts and negative prompts to zero in on what I want.

That process is novel and pretty fun, but it doesn’t imply the model is good at following my prompt.

LLMs are similar. Initially they seem good at following a prompt, but continue the conversation and they start showing recall issues, knowledge gaps, improper formatting, etc.

It’s not dishonest to say semi-random. It’s accurate. The detokenizing step of inference, for example, is taking a sample from a probability distribution which the model generates. Literally stochastic.

Why should facts win? It's art, and there are no rules in art. I could draw black george washington too.

[edit]

Statistical inference machines following human language prompts that include "please" and "thank you" have absolutely 0 ideas of what a fact is.

"A stick bug doesn't know what it's like to be a stick."

If there are no rules in art, then white George Washington should be acceptable.

But I would counter that there are certainly rules in art.

Both historical (expectations and real history) and factual (humans have a number of arms less than or equal to 2).

If you ask Gemini to give you an image of a person and it returns a Pollock drip work... most people aren't going to be pleased.

Art doesn't have to be tethered to reality, but I think it's reasonable to assume that a generic image generation ai should generate images according to reality. There's no rules in art, but people would be pretty baffled if every image that was generated by gemeni was in dr seuss's art style by default. If they called it "dr seuss ai" I don't think anyone would care. Likewise, if they explicitly labeled gemini as "diverse image generation" or whatever most of the backlash would evaporate.

If you try to draw white George Washington but the markers you use keep spitting out different colors from the ones you picked, you’d throw out the entire set and stop buying that brand of art supplies in the future.

Because white people exist and it refuses to draw them when asked explicitly. It doesn’t refuse for any other race.

>It seems perfectly reasonable to say that generated imagery should attempt to not lean into stereotypes and show a diverse set of people.

It might be "perfectly reasonable" to have that as an option, but not as a default. If I want an image of anything other than a human, you'd expect the sterotypes to be fulfilled. If I want a picture of a cellphone, I want an ambiguous black rectangle, even though wacky phones exist[1]

[1] https://static1.srcdn.com/wordpress/wp-content/uploads/2023/...

The stereotype of a human in general would not be white in any case.

And the stereotype the person asking would expect will heavily depend on where they're from.

Before you ask for stereotypes: Whose stereotypes? Across which population? And why does those stereotypes make sense?

I think Google fucked up thoroughly here, but they did so trying to correct for biases also gets things really wrong for a large part of the world.

And a stereotype of a phone doesn't have nearly the same historical context or ongoing harmful effects on the world as a racial stereotype.

Reality is statistics and as are the models.

If the data is lumpy in one area then I figure let the model represent the data and allow the human to determine the direction of skew in a transparent way.

The Nerfing based upon some internal activism that's hidden is frustrating because it'll call into question any result as suspect to bias towards unknown Morlocks at Google.

For some reason Google intentionally stopped historically accurate images from being generated. Whatever your position, provided you value Truth, these adjustments are abhorrent.

It's actually not hard to get these right and these are not stereotypes.

Try these exact prompts in Midjourney and you will get exactly what you would expect.

> It seems perfectly reasonable to say that generated imagery should attempt to not lean into stereotypes and show a diverse set of people

No, it's not reasonable. It goes against actual history, facts, and collected statistics. It's so ham-fisted and over the top, it reveals something about how ineptly and irresponsibly these decisions were made internally.

An unfair use of a stereotype would be placing someone of a certain ethnicity in a demeaning context (eg, if you asked for a picture of an Irish person and it rendered a drunken fool).

The Google wokeness committee bolted on something absurdly crude, seems like "when showing people, always include a black, an asian and an native american person" which rightfully results in a pushback from people who have brains.

How is "stereotype" different from "statistical reality"? How does Google get to decide that its training dataset -"the entire internet" - does not fit the statistical distribution over phenotypic features that its own racist ideological commitments require?

Really hard to get this right? We're not talking about a mistake here or there. We're talking about it literally refusing to generate pictures of white people in any context. It's very good at not doing that. It seemingly has some kind of supervisory system that forces it to never show white people.

Google has a history of pushing woke agendas with funny results. For example, there was a whole thing about searching for "happy white man" and "happy black man" a couple years ago. It would always inject black men somewhere in the results searching for white men, and the black man results would have interracial couples. Same kind of thing happened if you searched for women of a particular race.

The sad thing in all of this is, there is actively racism against white people in hiring at companies like this, and in Hollywood. That is far more serious, because it ruins lives. I hear interviews with writers from Hollywood saying they are explicitly blacklisted and refused work anywhere in Hollywood because they're straight white men. Certain big ESG-oriented investment firms are blowing other people's money to fund this crap regardless of profitability, and it needs to stop.

Reply on news.ycombinator.com