Organised crime? If a big enough organised crime group wanted it you'd get stuffed into the back of a van.
CIA? MI6? You'd get stuffed into the back of a van. FSB? They'd invade the country to stuff you into the back of a van then bum 20 quid off you for diesel. Mossad? They'd break into your house and replace your computer with an identical fully-functional one made of polonium *while you were using it* and say they didn't do it, while showing AI-generated footage of a Russian doing it on TV and crediting you for the software they totally didn't steal.
Had a good chuckle. Didn’t know Tom Clancy frequented hacker news.
As you say, the big bads have means regardless, and probably don't need it because they're often big enough to hire a lot of humans to do e.g. propaganda for them.
Now, imagine a conspiracy theorist. Not a harmless moon landing denier, but someone who is convinced that pizzagate was a real thing, and gets AI to fake mountains of "evidence" until the politicians that he or she is convinced did those things, are lynched, or at the very least become politically toxic.
Think it can't happen? Fake porn is already being used to harm random women, and real pictures from veterinary text books have been used as fake "evidence" of animal abuse.
AI is a tool that can make anyone more competent, but it doesn't make us better or wiser or kinder.
I'll do you one better than that. Farms use a lot of seasonal workers and small farms in particular will hire people in for a week or so when things are busy. I know a farmer who hired in a bunch of folk to help over the shearing. It turns out that two of the people he hired wanted to film around the farm to capture some of the animal abuse that absolutely must be going on because farmers have nothing better to do, but they didn't find any. So figuring that they were never going to let themselves into the "inner secret animal abuse lair" or whatever the fuck was going through their stoned little minds, they decided to...
yup...
... film themselves kicking fuck out of some sheep, and pretend it was common practice among farm workers.
EDIT: All of the replies are assuming I'm saying this is a good reason for not releasing these models. I don't say that anywhere below nor do I agree with it.
These models are capable of generating photorealistic child pornography. That's why they're not being released.
I have no evidence to back this up, other than anecdotal. All the hand waving vague posts from SD, DALLE etc read like PR tactics to avoid getting the spotlight from the media and governments.
No one wants to post about this but I think everyone knows it's true, as soon as Congress realizes what these models can be used for there's going to be a moral panic.
You should strive to respond to the strongest interpretation of what you read and not the weakest one.
(You also suggested some impending moral panic. I don’t buy it.)
The United States has barely moved past the issue of homosexuality even though it harms no one, it is perhaps the most straightforward modern example of people trying to ban it due to personal & traditional disgust. Although there is a general majority approval for it, that doesn't carry to the electoral level where it remains a contentious partisan issue. But you think the ethics of CP & pedophilia will be discussed and legislated rationally?
I've yet to see an 18 year old that looks like an actual child regardless of how they're dressed however I'm also not looking for this sort of thing so...
This kind of thing happened a lot, and I think more than one entirely "mainstream" director in the 60s/70s filmed nude actresses below the age of 18. Fortunately it's a lot harder to get away with that these days. Not to mention (Weinstein and many, many others) sexually assaulting the actresses off screen.
Her entirely involuntary response was to exclaim, "Lady, I'm damn near 30!"
So yes; if a 28-year-old can be mistaken for younger than 14 without trying (or even wanting to), there are plenty of 18-year-olds who can appear much younger, especially if they are using makeup, acting tricks, and other means to artificially enhance that.
You should strive to respond to the strongest interpretation of what you read and not the weakest one.
If their "real reason" of CP use cases doesn't exist right now because AI tooling explicitly bans CP use cases, they have no reason to block AI tools.
We've not even begun to ask the question of whether there are ethical problems of exposing children to unregulated AI and vice versa.
You should strive to respond to the strongest interpretation of what you read and not the weakest one.
All of us can be tempted into overly quick responses without truly reading what we're responding to.
The commenter did not necessarily adopt any of these positions themselves.
I agree that you were civil about it and no harm was done other than a bumpier than necessary comment thread.
Is this true, have you seen it or just heard about it? because what I seen is that the AI is bad, the faces are wrong, the fingers are wrong,t he hands and legs are often connected wrong, you can get very often more then 2 legs more then 5 fingers. I seen a community that focus on waifus, this images are not realistic, are obvious carton, animation styles and similar images were created already with digital painting tools.
I also did not seen any porn scene generated by AI, I don't think the model was trained on porn scenes so at best you will get puritans removing breasts from the models, including art , but for sure I want to see how they will ensure that the AI can only generate male nipples and not women ones.
You can retrain SD without NSFW images, is this enough? I suspect not, because what if Timy in his basement trains SD with the face of X and the naked body of Y and then makes a nude image ?
You know what makes rational/logic sense ? |Follow the money, an open source model like SD makes proprietary ones like OpenAI and googles lose ton of money, like a giant pile of money Californian billionaires will lose , so they use their money to create drama.
I have seen it. SD is capable of generating CP that is indistinguishable from a real photograph with correct anatomy.
> I also did not seen any porn scene generated by AI
There are user made SD models trained on porn available for months now.
You mean a child with no clothes? is that CP? there are painting and art of children with no clothers, or you mean you seen sexual acts too? Since I never seen it with adults , AFAIK SD was not trained with porn scenes , at best you might get realistic breasts but not get realistic sexual organs and add on top the low resolution.
>There are user made SD models trained on porn available for months now.
So you would also have an issue even if SD was 100% SFW because some dude in his basement would train it to generate breasts and dicks and then combine it with SD that can generate faces, so SD should not exist ?
People were drawing porn stuff digitally for decades, now an AI will do the same, maybe you need a better case , CP is still illegal, Google and Apple will snitch on you also so WTF is this hysteria ?
In fact tbe cause of this CP and Google stuff I avoided experimenting with Google colab SD, the AI might paint a monstrosity but Google AI might think is CP and close my accounts.
Also you feailled to respond to my accusation of missing evidence, missing evidence that drawing porn or CP ever caused victims, or more victims then say sending you kid to church.
I'm not quite sure what the solution is to this by the way. I don't think the answer is no regulation, but how to improve the quality of regulation.
Do companies push for the creation regulatory bodies to stifle competition?
Whereas regulatory capture is the process by which companies shape existing regulatory bodies to serve their interests. But it presumes the existence of a body to regulate them; regulatory capture can't occur if there isn't regulation to start with.
This paper may be interesting further reading on the topic of regulatory moats specifically. (I saw your comment below discussing regulatory capture vs this specific subset of capture.)
https://www.nber.org/papers/w26001?sy=001&fbclid=IwAR0oYZBL3...
As for what they say to the public, well, fraudsters gonna fraud. It's a tale as old as time [1].
1: https://www.biography.com/news/famous-grifters-stealers-swin...
1. This is absolutely a gatekeeping, ladder pulling measure. 2. The commercial face of this industry is rife with dogmas of self importance. It’s nearly comical.
Add to this mix the USA trying to prevent China copying cutting edge AI technology and this was bound to happen sooner or later.
It's no longer research it's near market viability.
The amount of cool stuff people have created on top of Stable Diffusion since it was created is also amazing and would never have happened if they didn't release the weights.
Paintings that look like Pepe the frog?
The resulting storm of new applications was definitely interesting to watch, though none of em were really market viable with that version of the model.
> Now, I don't claim everyone should be forced to publish their AI models. No, if you spent lots of money on training your model it is yours. But you can't lock all your work behind closed doors and call yourself open. It doesn't work like this.
Both of you can be right. Just don't try to call yourself open.
Claiming to be “open” while not actually being open is a feature, not a bug. Just business as usual
ML-tech from the past century seems to work just fine for killing each other (have a look at Ukraine). I would be rather more concerned about the foe having better batteries at this point.
What do you think unlimited funds dumped into AI for military applications can do?
The US KNOWS this is a big deal, so much so that it has destroyed China's ability to conduct top end research.
What more destructive power can AI provide, beyond the already existing drones/biological etc etc?
The machine gun and trench warfare forced the development of the tank.
Once aircraft became military useful, battleships were rapidly vulnerable to aircraft and hence aircraft carriers.
I can imagine replacing naval mines with normally-quiescent remote-triggered torpedoes that are scattered on the seafloor a decade in advance.
How effective are troops against 3D printed drones designed to mimic animals (including birds), but which have a short-range firearm and some computer vision? Or engineered mosquitoes with the bare minimum of remote control, perhaps similar to the "cyborg cockroach" kits that have been on sale for about 9 years now? (Or just weaponise those 'roaches…)
What happens if you can predict what a command officer will say, use an AI to fake a their voice before they've spoken, and interfere with the communications to give misleading orders in the heat of combat? Doesn't even need to be a big change to the orders to alter the outcome.
(Obviously everything I've thought of in 5 minutes, the actual military will have categorised into "haha no" and "let's wargame this scenario"; I assume mostly the former).
(I'm not sure if your reply was a thumbs-up or a serious question, but that's a serious answer).
You do understand that there are nuances here, yes? In other words, that there are military operations that lie somewhere between "do nothing" and "destroy the entire world"?
It's reasonable that it's fair use to train on unlicensed, freely available data as long as the model is released. I don't know anything about the law or care much about the legal stuff, but that's my opinion for how it should be.
Kodak had billions-with-a-b of dollars invested in the production of photographic film.
That didn't save them from competitors.
This happends with every new technology.
The worst case scenario of the currently public models is why they want to take it slow.
It's like the lottery in one of the episodes of the TV show Sliders: you probably won't win, but if you do win, you die.
Unfortunately, most people are really bad at grokking probably, and, by extension, risk, especially in scenarios like this.
> Bad actors will have enough money/lack scruples to train their own models or to steal your best ones regardless how "impact conscious" your company it.
Indeed, totally correct.
But this is also on the list titled "why AI alignment is hard and why we need to go slow because we have not solved these things yet".
Saying "it doesn't matter if we keep this secret, someone else will publish anyway" is exactly the same failure mode as "it doesn't matter if we keep polluting, someone else will pollute anyway".
99% of those using this tactics use it to justify not releasing their models to avoid giving competition a leg up(Google, openAI) and to pretend they are for "open research". As I said this is 100% bull.
The remaining 1% are either doing this to inflate their egos ("hey look how considerate and enlightened we are in everything we do!"), or they pander to media/silly politicians/various clueless commentators whose level of knowledge about this technology is null. They regurgitate the same set of "what ifs and horror stories" to scare the public into standing by when they attempt to over regulate another field so they can be kingmakers within it(if you want an example how it works look at the energy sector).
All this sillyness accomplishes is to raise a barrier to entry for potential commercial competition. Bad actors will have enough money/lack scruples to train their own models or to steal your best ones regardless how "impact conscious" your company it.
Now, I don't claim everyone should be forced to publish their AI models. No, if you spent lots of money on training your model it is yours. But you can't lock all your work behind closed doors and call yourself open. It doesn't work like this. One important point is that there is value even in just publishing a paper demonstrating some achievements of a proprietary model, but if the experiment can't be reproduces based on description given that is not science and for sure it is not open.