923
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 10 Jan 2024
923 points (96.5% liked)
Technology
59570 readers
3354 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
Using publically available data to train isn't stealing.
Daily reminder that the ones pushing this narrative are literally corporation like OpenAI. If you can't use copyright materials freely to train on, it brings up the cost in such a way that only a handful of companies can afford the data.
They want to kill the open-source scene and are manipulating you to do so. Don't build their moat for them.
And using publicly available data to train gets you a shitty chatbot...
Hell, even using copyrighted data to train isn't that great.
Like, what do you even think they're doing here for your conspiracy?
You think OpenAI is saying they should pay for the data? They're trying to use it for free.
Was this a meta joke and you had a chatbot write your comment?
if someone said this to me I'd cry
The point that was being made was that public available data includes a whole lot amount of copyrighted data to begin with and its pretty much impossible to filter it out. Grand example, the Eiffel tower in Paris is not copyright protected, but the lights on it are so you can only using pictures of the Eiffel tower during the day, if the picture itself isn't copyright protected by the original photographer. Copyright law has all these complex caveat and exception that make it impossible to tell in glance whether or not it is protected.
This in turn means, if AI cannot legally train on copyrighted materials it finds online without paying huge sums of money then effectively only mega corporation who can pay copyright fines as cost of business will be able to afford training decent AI.
The only other option to produce any ai of such type is a very narrow curated set of known materials with a public use license but that is not going to get you anything competent on its own.
EDIT: In case it isn't clear i am clarifying what i understood from Grimy@lemmy.world comment, not adding to it.
So then we as a society aren't ready to untangle the mess of our infancy in the digital age. ChatGPT isn't something we must have at all costs, it's something we should have when we can deploy it while still respecting the rights of people who have made the content being used to train it.
I would go even further and say that we should have it until we can be sure it will respect others' rights. All kind of rights, not only Copyright. Unlike Bing at the beginning, with all it's bullying and menaces, or Chatgpt regurgitating private information gathered from God knows where.
The problem with waiting is the arms race with other governments. I feel it's similar to fossil fuels, but all governments need to take the risk of being disadvantaged. Damned prisoner's dilemma.
I didn't want any of this shit. IDGAF if we don't have AI. I'm still not sure the internet actually improved anything, let alone what the benefits of AI are supposed to be.
It doesn’t matter what you want. What matters is if corporations can extract $ from you, gain an efficiency, or cut their workforce using it.
That’s what the drive for AI is all about.
No doubt.
A perfectly valid stance to take.
Machine learning techniques are often thought of as fancy function approximation tools (i.e. for regression and classification problems). They are tools that receive a set of values and spit out some discrete or possibly continuous prediction value.
One use case is that there are a lot of really hard+important problems within CS that we can't solve efficiently exactly (lookup TSP, SOP, SAT and so on) but that we can solve using heuristics or approximations in reasonable time. Often the accuracy of the heuristic even determines the efficiency of our solution.
Additionally, sometimes we want predictions for other reasons. For example, software that relies on user preference, that predicts home values, that predicts the safety of an engineering plan, that predicts the likelihood that a person has cancer, that predicts the likelihood that an object in a video frame is a human etc.
These tools have legitamite and important use cases it's just that a lot of the hype now is centered around the dumbest possible uses and a bunch of idiots trying to make money regardless of any associated ethical concerns or consequences.
You don't have to use it. You can even disconnect from the internet completely.
Whats the benefit of stopping me from using it?
It's not like all this data was randomly dumped at the AIs. For data sets to serve as good training materials they need contextual information so that the AI can discern patterns and replicate them when prompted.
We see this when you can literally prompt AIs with whose style you want it to emulate. Meaning that the data it was fed had such information.
Midjourney is facing extra backlash from artists after a spreadsheet was leaked containing a list of artist styles their AI was trained on. Meaning they can keep track of it and they trained the AI with those artists' works deliberately. They simply pretend this is impossible to figure out so that they might not be liable to seek permission and compensate the artists whose works were used.
That's insane logic...
Like you're essentially saying I can copy/paste any article without a paywall to my own blog and sell adspace on it...
And your still saying OpenAI is trying to make AI companies pay?
Like, do you think AI runs off free cloud services? The hardware is insanely expensive.
And OpenAI is trying to argue the opposite, that AI companies shouldn't have to pay to use copyrighted works.
You have zero idea what is going on, but you are really confident you do
I clarified the comment above which was misunderstood, whether it makes a moral/sane argument is subjective and i am not covering that.
I am not sure why you think there is a claim that openAI is trying to make companies pay, on the contrary the comment i was clarifying (so not my opinion/words) states that openAI is making an argument that anyone should be able to use copyrighted materials for free to train AI.
The costs of running an online service like chatgpt is wildly besides the argument presented. You can run your own open source large language models at home about as well as you can run Bethesda's Starfield on a same spec'd PC
Those Open source large language models are trained on the same collections of data including copyrighted data.
The logic being used here is:
The Ethical dilemma as i understand it is:
No. Many creatives fear that AI allows anyone to do what they do, lowering the skill premium they can charge. That doesn't depend on free training.
Some seem to feel that paying for training will delay AI deployment for some years, allowing the good times to continue (until they retire or die?)
But afterward, you have to ask who's paying for the extra cost when AI is a normal tool for creatives? Where does the money come from to pay the rent to property owners? Obviously the general public will pay a part through higher prices. But I think creatives may bear the brunt, because it's the tools of their trade that are more expensive and I don't think all of that cost can be passed on.
I don't think lowering the skill level is something we will need to worry about as over time this actually trickles up, A Creative professional trained with AI tools will almost always top a Amateur using the same tools.
The real issue is Style. If you are an Artist with a very recognizable specific style, and you make your money trough commissions you are basically screwed. Many Artists feature a personal style and while borrowing peoples style is common (disney-esque) it's usually not a problem because within a unique and diverse human mind it rarely results in unintentional latent copying.
That is very well put, I really wish I could have started with that.
Though I envision it as a loss for BigProfit Enthertainment since I see this as a real boon for the indie gaming, animation and eventually filmmaking industry.
It's definitely overall quite a messy situation.
...
Yes, you can download an executable of a chatbot lol.
That's different than running something remotely like even OpenAI.
The more it has to reference, the more the system scales up. Not just storage, but everything else.
Like, in your example of video games it would be more like stripping down a PS5 game of all the assets, then playing it on a NES at 1 frame per five minutes.
You're not only wildly overestimating chatbots ability, you're doing that while drastically underestimating the resources needed.
Edit:
I think you literally don't know what people are talking about..
Do you think people are talking about AI image generators?
No one else is...
I think you're confusing training it with running it. After it's trained, you can run it on much weaker hardware.
I am talking about generative AI, be it text or image both have a challenge with copyrighted material.
Are you refering to my joke?
I am far from overestimating capacity, Starfield runs mediocre on a modern gaming system compared to other games. The Vicuna 13b llm runs mediocre on the same system compared with gpt 3.5. To this date there is no local model that i would trust for professional use and chatgpt 3.5 doesnt hit that level either.
But it remains a very interesting, rapidly evolving technology that i hope receives as much future open source support as possible.
I presume you must believe the the following lemmy community and resources to be typed up by a group of children, either that or your just naive.
https://lemmy.world/c/fosai
https://www.fosai.xyz/
https://github.com/huggingface/transformers
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard
https://huggingface.co/microsoft/phi-2 & https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/
https://www.theguardian.com/technology/2023/may/05/google-engineer-open-source-technology-ai-openai-chatgpt
Hey man, that’s damn hurtful
I’m not sure if someone else has brought this up, but I could see OpenAI and other early adopters pushing for tighter controls of training data as a means to be the only players in town. You can’t build your own competing AI because you won’t have the same amount of data as us and we’ll corner the market.
If the data has to be paid for, openAI will gladly do it with a smile on their face. It guarantees them a monopoly and ownership of the economy.
Paying more but having no competition except google is a good deal for them.
Eh, the issue is lots of people wouldn't be willing to sell tho.
Like, you think an author wants the chatbot to read their collected works and use that? Regardless of if it's quoting full texts or "creating" text in their style.
No author is going to want that.
And if it's up to publishers, they likely won't either. Why take one small payday if that could potentially lead to loss of sales a few years down the row.
It's not like the people making the chatbits just need to buy a retail copy of the text to be in the legal clear.
The publisher's will absolutely sell imo. They just publish, the book will be worth the same with or without the help of AI to write it.
I guess there is a possibility that people start replacing bought books with personalized book llm outputs but that strikes me as unlikely.
deleted
It's current and it's the only open source project that's under direct threat? I am both a fan of open source and of generative AI, not sure what that changes in the validity of my arguments.
This isn't a gotcha but pure rhetoric, which is on par with you. Attack my arguments, or just ignore me the moment it becomes clear you can't insult yourself out of a debate like you did last time.
I'm not even sure what exactly you are implying but I am not impressed.
deleted
OpenAI is definitely not the one arguing that they have stole data to train their AIs, and Disney will be fine whether AI requires owning the rights to training materials or not. Small artists, the ones protesting the most against it, will not. They are already seeing jobs and commission opportunities declining due to it.
Being publicly available in some form is not a permission to use and reproduce those works however you feel like. Only the real owner have the right to decide. We on the internet have always been a bit blasé about it, sometimes deservedly, but as we get to a point we are driving away the very same artists that we enjoy and get inspired by, maybe we should be a bit more understanding about their position.
Thats basically my main point, Disney doesn't need the data, Getty either. AI isn't going away and the jobs will be lost no matter what.
Putting a price tag in the high millions for any kind of generative model only benefits the big players.
I feel for the artists. It was already a very competitive domain that didn't really pay well and it's now much worse but if they aren't a household name, they aren't getting a dime out of any new laws.
I'm not ready to give the economy to Microsoft, Google, Getty and Adobe so GRRM can get a fat payday.
If AI companies lose, small artists may have the recourse of seeking compensation for the use and imitation of their art too. Just feeling for them is not enough if they are going to be left to the wolves.
There isn't a scenario here in which big media companies lose so talking of it like it's taking a stand against them doesn't make much sense. What are we fighting for here? That we get to generate pictures of Goofy? The small AI user's win here seems like such a silly novelty that I can't see how it justifies just taking for granted that artists will have it much rougher than they already have.
The reality here is that even if AI gets the free pass, large media and tech companies are still primed to profit from them far more than any small user. They will be the one making AI-assisted movies and integrating chat AI into their systems. They don't lose in either situation.
There are ways to train AI without relying on unauthorized copyrighted data. Even if OpenAI loses, it wouldn't be the death of the technology. It may be more efficient and effective to train them with that data, but why is "efficiency" enough to justify this overreach?
And is it even wise to be so callous about it? Because it's not going to stop with artists. This technology has the potential to replace large swaths of service industries. If we don't think of the human costs now, it will be even harder to make a case for everyone else.
I fully believe AI will be able to replace 50% or more of desk jobs in the near future. It's definitely a complicated situation and you make good points.
First and foremost, I think it's imperative the barrier for entry for model training is as low as possible. Anything else basically gives a select few companies the ability to charge a huge subscription fee on all our goods and services.
The data needed is pretty heavy as well, it's not very pheasible to go off of donated or public domain data.
I also think any job loss is virtually guaranteed and trying to save them is misguided as well as not really benefiting most of those affected.
And yea, the big companies win either way but if it's easier to use this new tech, we might not lose as hard. Disney for instance doesn't have any competition but if a bunch of indie animation companies and groups start popping up, it levels the playing field a bit.
That depends on what your definition of "publicly available" is. If you're scraping New York Times articles and pulling art off Tumblr then yeah, it's exactly stealing in the same way scihub is. Only difference is, scihub isn't boiling the oceans in an attempt to make rich people even richer.
We have a mechanism for people to make their work publically visible while reserving certain rights for themselves.
Are you saying that creators cannot (or ought not be able to) reserve the right to ML training for themselves? What if they want to selectively permit that right to FOSS or non-profits?
That’s exactly what they’re saying. The AI proponents believe that copyright shouldn’t be respected and they should be able to ignore any licensing because “it’s hard to find data otherwise”
Essentially yes. There isn't a happy solution where FOSS gets the best images and remains competitive. The amount of data needed is outside what can be donated. Any open source work will be so low in quality as to be unusable.
It also won't be up to them. The platforms where the images are posted will be selling and brokering. No individual is getting a call unless they are a household name.
None of the artists are getting paid either way so yeah, I'm thinking of society in general first.
Yeah, by using the argument you just gave as an excuse to "launder" copyleft works in the training data into permissively-licensed output.
Including even a single copyleft work in the training data ought to force every output of the system to be copyleft. Or if it doesn't, then the alternative is that the output shouldn't be legal to use at all.
100% agree, making all outputs copyleft is a great solution. We get to keep the economic and cultural boom that AI brings while keeping the big companies in check.
Scientific research papers are generally public too, in that you can always reach out to the researcher and they'll provide the papers for free, it's just the "corporate" journals that need their profit off of other peoples work...
True, Big Tech loves monopoly power. It's hard to see how there can be an AI monopoly without expanding intellectual property rights.
It would mean a nice windfall profit for intellectual property owners. I doubt they worry about open source or competition but only think as far as lobbying to be given free money. It's weird how many people here, who are probably not all rich, support giving extra money to owners, merely for owning things. That's how it goes when you grow up on Ayn Rand, I guess.
All of the AI fear mongering is fuelled by mega corps who fear that AI in some sort will eat into their profits and they can't make money off of it.
Image generation also had similar outcry because open source models smoked all the commercial ones.
Yeah, just wait until they see the ai design tools that allow anyone to casually describe the spare part or upgrade they want and it'll be designed and printed at home or local fab shop.
Lot of once fairly safe monopolies are going to start looking very shaky, and then things like natural language cookery toolarms disrupting even more...
We've only barely started to see what the tech we have now is able to do, yes a million shitty chat bots / img gen apps are cashing in on the hype but when we start seeing some killer apps emerge it's when people won't be able to ignore it any longer
This is the hardest thing to explain to people. Just convert it into a person with unlimited memory.
Open AI is sending said person to view every piece of human work, learns and makes connections, then make art or reports based on what you tell/ask this person.
Sci-Hub is doing the same thing but you can ask it for a specific book and they will write it down word for word for you, an exact copy.
Both morally should be free to do so. But we have laws that say the sci-hub human is illegally selling the work of others. Whereas the open ai human has to be given so many specific instructions to reproduce a human work that it’s practically like handing it a book and it handing the book back to you.
Too bad
If you can’t afford to pay the authors of the data required for your project to work, then that sucks for you, but doesn’t give you the right to take anything you want and violate copyright.
Making a data agnostic model and releasing the source is fine, but a released, trained model owes royalties to its training data.
The point is the entire concept of AI training off people's work to make profit for others is wrong without the permission of and compensation for the creator regardless if it's corporate or open source.
I think I've decided to not publish anything that I want to keep ownership of, just in case. There's an entire planet's worth of countries, which will all have their own sets of laws. It takes waay too long to polish something, only to just give it away for free haha. Someone else is free to do that work if it is that easy. No skin off my back.
I think it's similar to many other hand-made crafts/items. Most people will buy their clothes from stores, but there are definitely still people who make beautiful clothing from hand better than machines could.
Don't even get me started on stuff like knitting. It already costs the creator a crap ton of money just for the materials. It takes a crap ton of time to make those, too. Despite the costs, many people just expect those knitted pieces for practically free. The people who expect that pricing are also free to go with machine-produced crafts/items instead.
It comes down to what people want, and what they're willing to pay, imo. Some people will find value in something physically being put together by another human, and other people will find value in having more for less. Neither is "wrong" necessarily, so long as no one is literally ripped off. (With over 8 billion people, it's bound to happen at least once. I feel bad for whoever that is.)
That being said, we'll never be able to honestly say that the specific skills and techniques that are currenty required are the exact same. It would be like calling a photographer amazing at realism painting because their photo looks like real life. Photographers and painters both have their place, but they are not the exact same.
I think that's also part of what's frustrating so many artists. Coding AI is not the same as using the colour wheel, choosing materials, working fine motor control, etc. It's not learning about shadows, contrast, focal points, etc. I can definitely understand people not wanting those aspects to be brushed off, especially since it usually takes most of a lifetime to achieve. A music generator and a violin may both make great music, but they are not the same, and they require different technical skills.
I'll never buy AI art if I have any say in the matter. I'll support handmade stuff first, every time.
There is definitely more value in hand made art. Even the fanciest prints on canvas can't compare and I don't think AI art will be evoking the same feelings a john waterhouse exhibit does any time soon.
On the subject of publishing, I've chosen to embrace it personally. My view is that even the hidden stuff on our comp ends up in a Chinese or US databases anyways.
What data is public?