Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Some notes on the Stable Diffusion safety filter

vickiboykis.com

Post reply on HN
Print view View original

81–88 of 88 posts

Re: Some notes on the Stable Diffusion safety filter

#81

Post by hatenberg »

calebkaiser wrote: post #40
jimbob45 wrote: post #14

Earlier quoted context omitted.

A child entering this prompt is probably expecting one thing, but the internet is filled with pictures of another nature. There are possibly more adult 'my little pony' images than screenshots of the show on the internet. So everyone has to have gimpy AI just because parents can't be expected to take responsibility for what their child does and does not see? Why the fuck is a child being allowed to play with somethin…

I would recommend looking more closely at the article. Stability.ai, the company who developed and released the model being discussed, have not added a safety filter to the model. As the article points out, the filter is specifically implemented by HuggingFace's Diffusers library, which is a popular library for working with diffusion models (but again, to be clear, not the only option for using Stable Diffusion). The…

But they did. The line of code is literally in the CompVis release and you can disable it.

Re: Some notes on the Stable Diffusion safety filter

#82

Post by washadjeffmad »

maxbond wrote: post #65
cuddlyogre wrote: post #56

I can understand giving a user the option to filter out something they might not want to see. But the idea that the technology itself should be limited based on the subjective tastes and whims of the day makes my stomach churn. It's not too disconnected from altering a child's brain so that he is incapable of understanding concepts his parents don't like.

Nope, it's not like that in any way at all. AI aren't children, they're a technological artefact a group of people assembled, with no more moral dimension than a toaster. The people who would be the targets of Stable Diffusion-generated porn, depictions they did not consent to, they're actual people who's privacy would be harmed. Technology is an artefact of the society that produced it, and is shaped by it's values.…

> AI aren't children

That doesn't mean they couldn't or shouldn't be regulated as if they were. At some point, intelligence will be protected, and the implications of creating and training it may be governed in similar ways as humans currently allow.

Just as I see a clear and total delineation between commercial and non-commercial entities, defining non-adult and adult (or 'able to consent' and 'unable to consent') may be litmus tests for whether certain laws will apply under a non-human-centric world view.

Re: Some notes on the Stable Diffusion safety filter

#83

Post by dvngnt_ »

Gigachad wrote: post #50
iceburgcrm wrote: post #32

Earlier quoted context omitted.

Only available on the latest iphone

Apple ended up not implementing that iirc. While Google Photos has had it the whole time. Googles is actually worse. Apple was only going to match against known CSAM images while google has ML to identify new images which resulted in one parent being arrested for a medical image of their own child.

was never arrested. he was cleared but account is nuked

Re: Some notes on the Stable Diffusion safety filter

#84

Post by maxbond »

geysersam wrote: post #79
maxbond wrote: post #65

Earlier quoted context omitted.

Nope, it's not like that in any way at all. AI aren't children, they're a technological artefact a group of people assembled, with no more moral dimension than a toaster. The people who would be the targets of Stable Diffusion-generated porn, depictions they did not consent to, they're actual people who's privacy would be harmed. Technology is an artefact of the society that produced it, and is shaped by it's values.…

> The people who would be the targets of Stable Diffusion-generated porn, depictions they did not consent to Is it just me finding this notion of someone generating nsfw content to show it to other people without their consent really far fetched? It's about a hundred times more likely they just want it for themselves. The internet has billions of terrible images. If someone wants to spam awful porn today, they can ju…

There are already communities of people creating & sharing this imagery.

The problem is not that people would spam porn. The problem is that stalkers and creeps can & will use this technology to violate people's privacy. Celebrities and influencers complain bitterly in interviews about foot wiki, and try to control what pictures of them are posted to stop people from collecting pictures of their feet. How are they meant to feel about someone being able to push a button and produce a fictional sex tape of theirs? If every photo they post has the potential to be turned into porn, what impact do you imagine that having on people?

Furthermore, how are the people making these AIs meant to feel about that? If they want to limit their technology to not produce these images - we're many to understand that as "child abuse"? If I run a porn site and I take down revenge porn when it's reported to me, am I engaging in abuse?

Re: Some notes on the Stable Diffusion safety filter

#85

Post by maxbond »

washadjeffmad wrote: post #82
maxbond wrote: post #65

Earlier quoted context omitted.

Nope, it's not like that in any way at all. AI aren't children, they're a technological artefact a group of people assembled, with no more moral dimension than a toaster. The people who would be the targets of Stable Diffusion-generated porn, depictions they did not consent to, they're actual people who's privacy would be harmed. Technology is an artefact of the society that produced it, and is shaped by it's values.…

> AI aren't children That doesn't mean they couldn't or shouldn't be regulated as if they were. At some point, intelligence will be protected, and the implications of creating and training it may be governed in similar ways as humans currently allow. Just as I see a clear and total delineation between commercial and non-commercial entities, defining non-adult and adult (or 'able to consent' and 'unable to consent') m…

No, it does mean that. There are actual humans who's rights need to be protected, and there exists no AGI. In a world where AGI exists we'll probably want to create some legislation around that. But it'd be a mistake not to acknowledge that isn't the world we're living in and to make laws based on science fiction.

Realized problems take precedence over hypotheticals.

Re: Some notes on the Stable Diffusion safety filter

#86

Post by washadjeffmad »

maxbond wrote: post #85
washadjeffmad wrote: post #82

Earlier quoted context omitted.

> AI aren't children That doesn't mean they couldn't or shouldn't be regulated as if they were. At some point, intelligence will be protected, and the implications of creating and training it may be governed in similar ways as humans currently allow. Just as I see a clear and total delineation between commercial and non-commercial entities, defining non-adult and adult (or 'able to consent' and 'unable to consent') m…

No, it does mean that. There are actual humans who's rights need to be protected, and there exists no AGI. In a world where AGI exists we'll probably want to create some legislation around that. But it'd be a mistake not to acknowledge that isn't the world we're living in and to make laws based on science fiction. Realized problems take precedence over hypotheticals.

Do you think the world we lives in constitutes intelligences that are non-human? Do you believe that all humans possess the same intelligence?

It's not just humans we need to protect, is it?

Re: Some notes on the Stable Diffusion safety filter

#87

Post by maxbond »

washadjeffmad wrote: post #86
maxbond wrote: post #85

Earlier quoted context omitted.

No, it does mean that. There are actual humans who's rights need to be protected, and there exists no AGI. In a world where AGI exists we'll probably want to create some legislation around that. But it'd be a mistake not to acknowledge that isn't the world we're living in and to make laws based on science fiction. Realized problems take precedence over hypotheticals.

Do you think the world we lives in constitutes intelligences that are non-human? Do you believe that all humans possess the same intelligence? It's not just humans we need to protect, is it?

> Do you think the world we lives in constitutes intelligences that are non-human?

I've seen enough evidence to believe animals have a subjective and individual experience, and I believe most or all of them are intelligent, sure. That doesn't seem like what you meant though, I don't see any evidence that AI has a subjective experience, and I don't understand the relevance to the topic at hand, so I'm not sure if I'm engaging with this thought experiment correctly.

> Do you believe that all humans possess the same intelligence?

I don't believe relative intelligence is a measurable quantity, or even coherent as a concept (it evaporates if you really start to scrutinize it), so my answer to this question is undefined.

> It's not just humans we need to protect, is it?

Until such a time as AI have a subjective experience, it is not possible to harm them, morally speaking. Any more than it would be possible to harm a toaster - you could damage a toaster in the physical sense, but a toaster cannot suffer. If we are weighing harms against humans and AI as it exists today, we should come down on the side of humans every time.

Which isn't to say that intelligence is a bar you have to pass to deserve protection, for instance I support protecting rivers and even granting them water rights but I don't claim that rivers are intelligent. But the claim I was responding to was specifically that building limitations into an AI was harmful to the AI and analogous to child abuse.

Re: Some notes on the Stable Diffusion safety filter

#88

Post by knaik94 »

Gigachad wrote: post #78
knaik94 wrote: post #71

Earlier quoted context omitted.

I am fine with photos that are uploaded to the cloud being scanned. I do not want my own device spending energy and scanning images even before they completely leave my network or device. Google scanning my Google drive files is fine with me. Apple is much worse.

The apple one was only going to scan photos stored on iCloud. It scans them on your phone but if you don’t use iCloud it doesn’t scan anything. It’s a neat trick that means the Apple servers can know they aren’t storing illegal content without ever having to look at it. If it’s anything like the regular scanning iPhones do, it’s done overnight while plugged in.

That makes the phone the scanner and by definition it's before iCloud. Apple makes it very hard to use an iPhone/iPad without keeping iCloud enabled. I don't want my phone spying on me, that sets a dangerous precedent. I don't care how many times it is scanned by Apple after I store it unencrypted on iCloud servers, but scanning without my permission on my device before it leave my device is a violation of my privacy. Apple will look at it if your phone detects something and it will inform authorities without letting you know. With other companies, once it is detected on their server, they forward your information to government authorities. Apple instead wanted to have an inhouse team to filter false positives before notifying authorities. Apple is worse in every single way.
Post reply on HN