It's always interesting to note editorialized headlines. People "fleeing" Instagram, or "scrambling", or covid "surging", etc. When they could just provide the data about how many users are leaving and let the data speak for itself.
Whether there's even any data behind this or if it's an article based on a handful of random tweets who knows, as it's a pay-walled article.
On a meta level, what percentage of HN readers are paying for a subscription to the washington post? Why are links to paid articles even allowed when the vast majority of readers won't even have access to read it?
And Instagram had 800 million in 2017 (the last year with data on Wikipedia). Other searches and sites have numbers well over a billion users. I don't see how Cara is at all a threat or significant at the moment.
>On a meta level, what percentage of HN readers are paying for a subscription to the washington post? Why are links to paid articles even allowed when the vast majority of readers won't even have access to read it?
HN rules explicitly say we aren't allowed to complain about that
I subscribe to WaPo, but the latest editor changes are making me leery. I'm keeping an eye on the slant of their political coverage, and I'm ready to unsubscribe.
I am on IG, I follow about a hundred people, a bunch of friends I know IRL and a bunch of artists i like, I had 3 artists bounce to Cara in the last week in my timeline alone.
At first I thought it's a scam but after the second artist who was clearly not the type I gave it a look.
I mean, based on the comments on the average post, at least half of the commenters never actually click the link, but instead imagine what it might have been based on the headline, so it seems fair that headlines would have a unique importance on this here orange website.
It might be an interesting link aggregator site that would only let users comment on an article after being able to confirm they clicked on it. Possibly with some kind of timer to theoretically allow for time to read the article based on its length.
all forms of happiness or disappointment ultimately boil down to a decision that's made somewhere in your brain: this is bad. this is good. it is entirely within your means to decide for yourself, no matter your circumstances or the vagueries of happenstance.
Zen meditation is less gaslighting yourself into "false good" and more "the wise old man bowed his head solemnly and spoke, 'There is actually no difference between good and bad things. You imbecile. You fucking moron.'"
it's the same in Stoicism, once you realise that you get to choose for yourself, it doesn't matter what external circumstances are, you can still be pleased with your choices within what's open to you.
your comment also reminds me of discourses by Epictetus. he sounds like an irritated teacher modelling what patience looks like.
really? the one I went to felt more like having a parent for an hour a week, and what she did was help me identify the emotions I was experiencing and asking the right questions to help me come up with strategies to deal with those emotions.
she was not the one telling me to learn to be happy with less - I demanded that for myself.
I found the process entirely ineffective. They apparently have done moral code (de jour) where they’ve decided not to ever tell you anything or give advice and instead to just listen. Ann’s her the patient talk. This may work for certain personality types but sole people find talking Anna’s talking about themselves exhausting and NOT therapeutic.
if you want specific advice, you need to find the right sort of coach for that advice. for example, cognitive behavioural therapy is a skill that you can get taught, and you won't have to talk about yourself much, the tying in with your own experience can be done by you in your head. I haven't taken such a course myself, but that's my understanding of it after my ex- went through it.
if your problems are financial, you need an accountant. if you struggle to speak to a large crowd, you need a public speaking coach or training group.
a therapist is the one you talk to when you have stuff weighting you down and when you try to talk to others they give advice that starts with "you just need to ...", like you haven't tried it. if reading my post has riled you up, paying somebody to listen to you complaining about it might work quite well.
I dunno if that's entirely the case. State of the art LLM reasoning is very impressive but it clearly can't replace actual people doing complex reasoning jobs yet. Even stuff like telephone helplines.
On the other hand we're starting to see some crazily impressive progress on the physical robotics front.
I think it's definitely still an open question whether we'll see e.g. a humanoid robot that can load a dishwasher, or an LLM that can reliably .. I dunno, edit a video, first.
We do not have complete coverage in the kitchen. It's a bit of stretch to call a dishwashing machine a robot. Your air-conditioner is a cooling robot by the same logic. And no laundry or bathroom-cleaning bots anywhere.
My experience was quite the opposite: I couldn't find a setting to toggle and had to search for a guide. When I found one like https://beebom.com/opt-out-meta-ai-training-instagram-facebo... the process was not only shockingly obfuscated but I could not even submit the form because it demanded that I provide a prompt that made their AI disclose my personal information.
Unsurprisingly, artists are more neurotic than the average person. Online artists also seem to spend a lot of time lying to each other about how AI works.
While I agree it’s painful, I’ve seen examples of ai reproducing exact bastardized versions of famous instagram digital artists. You can clearly see where it learned from it and it’s extremely obvious, not in the sense of one who could say “oh yes I was inspired by this” but a literal 1:1 copy that has the drunken haze of melted ai nonsense over top. In this respect I was like damn … that may piss me off.
Yeah, you can do anything with AI so you can train it to reproduce a specific image if you want to. Normally this would be considered overfitting in a foundation model, so there's some protection in that they don't want to be doing it. But with fine-tuning it can happen intentionally.
But people can also copy you by hand if they want to.
As the saying goes, you can't put pandora back in the bag. AI art is here to stay, and it's here to dominate. The problem is the economic system within which art is created and assigned value, any attempts to solve the "problem" of AI art without addressing the underlying economic incentives are destined to fail in the same way that we have failed in preventing the use of sweatshops by clothing suppliers.
Even if we have strong, even overbearing legislation put into place to protect artists, we're just going to end up with the biggest most profitable offenders buying/using illegally trained models through middle men, and feigning ignorance when/if they are found out.
Though most of the "has this model been trained on X" conundrum is likely to be irrelevant on a years (months?) timescale as art styles are far less unique than artists would like to think. See tencent's PhotoMaker or other modified CLIP approaches. Even a model not trained on a particular face can be conditioned to generate said face using a oneshot approach because the vector representing that face can be constructed even if the face itself doesn't exist in the training data. I'm certain the same is true for artistic styles.
"AI" tech? yes we can, scrape the entire web, fire up the transformers (aka roll that VC money)! "prevent stealing from creators" tech? nah, too difficult, we don't know how, human problem (the neighborhood queer artist can't pay nearly as much).
Fingerprinting technologies. Generative tools can be mandated to embed watermarks. Original images can include fingerprinting and other counter-theft measures that tools must respect. Offenders who publish tools that don't can be prosecuted a la tornado cash. etc
It's not about art style. The model exists thanks to data, period. No data, no vectors. And almost all of that data is stolen.
Tornado Cash was a service that people used, Stable Diffusion is a 2gb blob of tensors. The latter is not something you can control, the MPAA/RIAA have been trying to control the proliferation of pirated media for 30 years, it has not worked and their case is WAY more clear cut under our current system of laws.
Additionally, there's no way you will be able to force people to insert fingerprinting into the model or surrounding tooling. There is no way to force fingerprinting into the source material, glaze etc are memes that are easily defeated and in fact can help stabilize outputs when training on glazed images is used as a negative embedding. The idea that any of this stuff is a viable method of combating gen AI is laughable. You can run this stuff on an 8 year old GPU with a blob the size of a couple netflix episodes. While outputs straight from SD may be identifiable, SD + manual intervention can easily create outputs that are not traceable. There is no workable solution that doesn't address the underlying economic incentives.
It doesn't matter that the model is trained on data that you believe is "stolen", people who want the model will have access to it, you won't be able to prove the outputs are AI, and even if you could, it won't be cost effective to prosecute people for creating in this way.
The problem here is our economic system and it's reliance on scarcity and markets to assign value to things that are fundamentally not scarce (AI models, digital goods) and it's failure to correctly assign value to things that have enormous positive externalities (artists creating culture).
This is not how that works my brother. The AI is already on my computer and the computers of millions of other people, it doesn't need to talk to the internet and uses an entirely open source stack. The minimum system requirements are a gaming computer from the last decade.
Nah mate you'll need to continuiusly acquire new data to stay updated, otherwise tomorrow a genius artist comes up and if your model can't imitate her style, what a sad day?
If you have the resources to scrape the entire web and sanitize the data and fight the anti-AI arms race and retrain the model all the time and you do it alone just for yourself at home for the lulz then I guess knock yourself out, no one will mind. But a gaming computer from last decade alone probably won't cut it
> entirely open source stack
I heard of some tornado cash and wow using OSS for crime totally made it not a crime!
Bruh, the idea that tornado cash and SD are somehow equivalent is laughable. The point is that the open source stack means that there's no point where you can force any of these fingerprinting things in. It's all already out in the open. It's just like HDCP, if I can see the video on my computer I can pirate it, the same goes for glaze and related approaches, fundamentally you are not going to be able to prevent someone from training on an image if a human can correctly interpret the image as the artist intended. Sure you may cause some extra steps, but those steps will quickly be automated away. The vast majority of the "theft" from artists in the form of LoRAs and finetunes is done by individuals and trivially propagated all over the internet.
I can train a LoRA on a particular artist in about 6 hours using my old ass computer and a base model that's almost 2 years old plus 20 images scraped from wherever and achieve great results. Most of what gives the model the generic capability to create images doesn't need to be retrained, finetuning on styles is very cost effective. Though many of the advances in the capability of SD have nothing to do with training the model at all but rather advances in instrumentation and conditioning. Art is like 90% building on generic work that others have done before you and like 10% your own contribution if that. Everything is a remix. That is the the core truth that allows style transfer approaches to work even if the specific style is not represented in the training data, in most cases there's a blend of conditioning vectors that gets you pretty close.
Maybe the future of art is "making art that's hard for the computer to recreate" which would be dope, but the stuff it's already good at? That's over and done with, that cat is not going back in the bag. And to be clear, the set of stuff it's good at is larger than the set of things that are in the training data because of advanced conditioning techniques like CoAdapters/IPAdapter/ControlNet/modified CLIP conditioning etc.
And yes, millions of people, SD 1.5 had 4.5mil downloads in the last month alone.
Nobody goes to jail for piracy, nobody is going to go to jail for this. Especially since proving piracy is infinitely easier than proving something was AI generated.
While I wasn't clear, you're also willfully misunderstanding me. The point is that none of the adversarial tools work, and I have strong doubts that any of them will ever work as a fundamental facet of how the data is processed by these models.
Though you may be surprised that you can make a style LoRA that preforms well off ~8 images and you can condition a model off a single image. Though if you're still doubting me I invite you to pose a challenge for my 8 year old GPU. ;)
No novelist wrote a novel without first reading other people's work, no painter painted a picture without studying other painters first. And then sold their work for money. All knowledge is derivative. AI just happens not be a person, that's all. But it isn't any different. As much as I hate AI, I don't see any problem with training models based on other people's work.
73 comments
[ 3.1 ms ] story [ 109 ms ] threadWhether there's even any data behind this or if it's an article based on a handful of random tweets who knows, as it's a pay-walled article.
On a meta level, what percentage of HN readers are paying for a subscription to the washington post? Why are links to paid articles even allowed when the vast majority of readers won't even have access to read it?
HN rules explicitly say we aren't allowed to complain about that
Admins seems to be quite open and helpful from what I saw.
I agree with their complaint.
At first I thought it's a scam but after the second artist who was clearly not the type I gave it a look.
Press 'M' to start a Crusade.
your comment also reminds me of discourses by Epictetus. he sounds like an irritated teacher modelling what patience looks like.
she was not the one telling me to learn to be happy with less - I demanded that for myself.
if your problems are financial, you need an accountant. if you struggle to speak to a large crowd, you need a public speaking coach or training group.
a therapist is the one you talk to when you have stuff weighting you down and when you try to talk to others they give advice that starts with "you just need to ...", like you haven't tried it. if reading my post has riled you up, paying somebody to listen to you complaining about it might work quite well.
https://x.com/AuthorJMac/status/1773679197631701238
On the other hand we're starting to see some crazily impressive progress on the physical robotics front.
I think it's definitely still an open question whether we'll see e.g. a humanoid robot that can load a dishwasher, or an LLM that can reliably .. I dunno, edit a video, first.
We have that, it's called a dish washer. We also have a robot for vacuuming, it's called a robovac, and so on.
We have robots for most mundane repetitive chores, they're just not humanoid like in SciFi, but are purposely designed for each task.
I like my little green squares
I can recommend gitea (easiest to setup) or gitolite + cgit (minimal)
But people can also copy you by hand if they want to.
For any other output, even if it does not reproduce 1:1, the only reason it works at all is thanks to acquired training data and most of it is stolen.
Even if we have strong, even overbearing legislation put into place to protect artists, we're just going to end up with the biggest most profitable offenders buying/using illegally trained models through middle men, and feigning ignorance when/if they are found out.
Though most of the "has this model been trained on X" conundrum is likely to be irrelevant on a years (months?) timescale as art styles are far less unique than artists would like to think. See tencent's PhotoMaker or other modified CLIP approaches. Even a model not trained on a particular face can be conditioned to generate said face using a oneshot approach because the vector representing that face can be constructed even if the face itself doesn't exist in the training data. I'm certain the same is true for artistic styles.
"AI" tech? yes we can, scrape the entire web, fire up the transformers (aka roll that VC money)! "prevent stealing from creators" tech? nah, too difficult, we don't know how, human problem (the neighborhood queer artist can't pay nearly as much).
Fingerprinting technologies. Generative tools can be mandated to embed watermarks. Original images can include fingerprinting and other counter-theft measures that tools must respect. Offenders who publish tools that don't can be prosecuted a la tornado cash. etc
It's not about art style. The model exists thanks to data, period. No data, no vectors. And almost all of that data is stolen.
Additionally, there's no way you will be able to force people to insert fingerprinting into the model or surrounding tooling. There is no way to force fingerprinting into the source material, glaze etc are memes that are easily defeated and in fact can help stabilize outputs when training on glazed images is used as a negative embedding. The idea that any of this stuff is a viable method of combating gen AI is laughable. You can run this stuff on an 8 year old GPU with a blob the size of a couple netflix episodes. While outputs straight from SD may be identifiable, SD + manual intervention can easily create outputs that are not traceable. There is no workable solution that doesn't address the underlying economic incentives.
It doesn't matter that the model is trained on data that you believe is "stolen", people who want the model will have access to it, you won't be able to prove the outputs are AI, and even if you could, it won't be cost effective to prosecute people for creating in this way.
The problem here is our economic system and it's reliance on scarcity and markets to assign value to things that are fundamentally not scarce (AI models, digital goods) and it's failure to correctly assign value to things that have enormous positive externalities (artists creating culture).
You'd be surprised. Pass a law requiring ClosedAI to start doing it tomorrow and what will they do?
If you have the resources to scrape the entire web and sanitize the data and fight the anti-AI arms race and retrain the model all the time and you do it alone just for yourself at home for the lulz then I guess knock yourself out, no one will mind. But a gaming computer from last decade alone probably won't cut it
> entirely open source stack
I heard of some tornado cash and wow using OSS for crime totally made it not a crime!
> millions of other people
Yeah sure my phone also has ML
I can train a LoRA on a particular artist in about 6 hours using my old ass computer and a base model that's almost 2 years old plus 20 images scraped from wherever and achieve great results. Most of what gives the model the generic capability to create images doesn't need to be retrained, finetuning on styles is very cost effective. Though many of the advances in the capability of SD have nothing to do with training the model at all but rather advances in instrumentation and conditioning. Art is like 90% building on generic work that others have done before you and like 10% your own contribution if that. Everything is a remix. That is the the core truth that allows style transfer approaches to work even if the specific style is not represented in the training data, in most cases there's a blend of conditioning vectors that gets you pretty close.
Maybe the future of art is "making art that's hard for the computer to recreate" which would be dope, but the stuff it's already good at? That's over and done with, that cat is not going back in the bag. And to be clear, the set of stuff it's good at is larger than the set of things that are in the training data because of advanced conditioning techniques like CoAdapters/IPAdapter/ControlNet/modified CLIP conditioning etc.
And yes, millions of people, SD 1.5 had 4.5mil downloads in the last month alone.
Promising jail is that point.
Remember, open source is besides the point, whether what you do is against the law is what it's all about;)
> training on an image
Talking about training on a single image like it's enough means you probably don't get the point of ML m8
While I wasn't clear, you're also willfully misunderstanding me. The point is that none of the adversarial tools work, and I have strong doubts that any of them will ever work as a fundamental facet of how the data is processed by these models.
Though you may be surprised that you can make a style LoRA that preforms well off ~8 images and you can condition a model off a single image. Though if you're still doubting me I invite you to pose a challenge for my 8 year old GPU. ;)
https://www.theverge.com/2012/12/20/3790560/instagram-new-te...