In addition to the possible business threat, forcing OpenAI to identify its use of copyrighted data would expose the company to potential lawsuits. Generative AI systems like ChatGPT and DALL-E are trained using large amounts of data scraped from the web, much of it copyright protected. When companies disclose these data sources it leaves them open to legal challenges. OpenAI rival Stability AI, for example, is currently being sued by stock image maker Getty Images for using its copyrighted data to train its AI image generator.
Aaaaaand there it is. They don’t want to admit how much copyrighted materials they’ve been using.
“Ripped off”? That isn’t how LLMs work.
That’s, uh, exactly how they work? They need large amounts of training data, and that data isn’t being generated in house.
It’s being stolen, scraped from the internet.
If it was publicly available on the internet, then it wasn’t stolen. OpenAI hasn’t been hacking into restricted content that isn’t meant for public consumption. You’re allowed to download anything you see online (technically, if you’re seeing it, you’ve already downloaded it). And you’re allowed to study anything you see online. Even for personal use. Even for profit. Taking inspiration from something isn’t a crime. That’s allowed. If it wasn’t, the internet wouldn’t function at a fundamental level.
I don’t think you understand how copyright works. Something appearing on the internet doesn’t give you automatic full commercial rights to it.
An AI has just as much right to web scrape as you do. It’s not a violation of copyright to do so.
It’s not an AI webscraping. It’s a commercial company deciding to do a mass ingest.
It’s the same thing. Just because you have personal opinions on the matter, however valid they may be, doesn’t make it any less the exact same thing.
That’s like saying that McDonald’s Super Sized fries aren’t fries because they’re commercially large. No, it’s still fries, there’s just a lot of fries being processed in one serving. And yet, despite the arguments and outcries of many, still legal.
Exact same thing with LLMs.
If it’s the same thing, then why describe it as an AI scraping it’s not. It’s a company that has scraped a corpus of data from the internet and has used that to train an AI.
The problem is that intellectual property law is complex. Simply saying two things are the same thing is your personal opinion. Content on the internet is not by-and-large public domain. It comes with a license, which lets you use it for certain purposes and not others. Saying, for example an AI reading a book is just like a human reading a book’ (not something you said, I don’t think) betrays a certain naivety about the way IP works.
But Google and Bing do that too. They scrape all the internet that they can get to so that they can sell ads (with a few steps in between)
They respect the NoRobots metatag.
Who says OpenAI doesn’t respect it too?
if you read a copyrighted material without paying and then forgot most of it a month later with vague recollection of what you’ve read the fact is you still accessed and used the copyrighted material without paying.
Now let’s go a step further, you write something that is inspired by that copyrighted material and what you wrote become successful to some degree with eyes on it, but you refuse to admit that’s where you got the idea from because you only have a vague recollection.
FWIW that’s pretty much exactly how all creative works are created. Especially music.
Except that nobody has a superhuman ability to create endless amounts of content almost instantly based on said work.
People throw this “artists/writers use inspiration to create X” argument all the time and it just totally ignores the fact that we’re not talking about some person spending 10s/100s/1000s of hours of their time to copy someone’s working style.
It’s a piece of software churning it out in seconds.
Do generative AI models typically focus on ONE person’s style? Don’t they mix together influences from thousands of artists?
FWIW this is not an area I read up on, and so I don’t have a strong opinion one way or the other.
For the image generating ones like Midjourney you could ask for an artist’s style by putting their name in the prompt.
It probably works the same in OpenAI.
If I would create a very slow AI that takes 10 or 100 hours for each response, would that make it any better in your opinion? I do not think calculation speed of a software is a good basis for legislation.
If analyzing a piece of art and replicating parts of it without permission is illegal, then it should be illegal regardless of the tools used. However, that would make every single piece of art illegal, so it’s not an option. If we make only the digital tools illegal then the question still remains where to draw the line. How much inefficiency is required for a tool to still be considered legal?
Is Adobe Photoshop generative auto-fill legal?
Is translating with deepl.com or the modern Google Translate equivalent legal?
Are voice activated commands on your mobile phone legal (Cortana, Siri, Google)?
All of these tools were trained in similar ways. All of these take away jobs (read: make work/life more efficient).
It’s hard to draw a line and I don’t have any solution to offer.
Except the illegally obtaining the copyrighted material part, which is the main point. And definitely not on this scale.
Web scraping isn’t illegal, though.
I think there can be said to be a meaningful difference due to the sheer scale and speed at which AIs can do this though.
Ultimately, I think it’s less of a direct legal question and more a societal question of whether or not we think this is fair or not. I’d expect it to ultimately be resolved by legislative bodies, not the courts.
That’s still not how LLMs work. I can’t believe everybody who is upset with them doesn’t understand this.
The LLM has no idea what it’s reading. None. It’s just doing a word association game, but at a scale we can’t comprehend. It knows what arrangement of words go together, but it’s not reproducing anything with any actual intent. To get it to actually output anything that actually resembles a single piece of material it was trained against would require incredibly specific prompts to get there, and at that point it’s not really the LLM’s making anymore.
There’s plenty of reasons to be against AI. Such as the massive amounts of data scraping that happens to train models, the possible privacy invasions that come with that, academic cheating, etc. But to be mad at AI for copyright infringement only shows a lack of understanding of what these systems actually do.
The training process of LLMs is to copy the source material word for word. It’s instructed to plagiarize during the training process. The copyrighted material are possibly in one way or another embedded into the model itself.
In machine learning, there’s always this concern whether the model is actually learning patterns, or if it’s just memorizing the training data. Same applies to LLMs.
Can LLMs recite entire pieces of work? Who knows?
Does it count as copyright infringement if it does so? Possibly.
No it isn’t. That;s not how neural networks work, like at all
It’s learning patterns. It’s not memorising training data. Again, not how the system works at all
No. No they can’t.
That’d be one for the lawyers were it to ever come up, but it won’t
Here’s a basic description of how (a part of) LLMs work: https://huggingface.co/learn/nlp-course/chapter1/6
LLMs are generating texts word for word (or token by token if you’re pedantic). This is why ChatGPT is slowly generating the response word by word instead of giving you the entire response at once.
Same applies during the training phase. It gets a piece of text and the word it’s supposed to predict. Then it’s tuned to improve its chances to predict the right word based on the text it’s given.
Ideally it’s supposed to make predictions by learning the patterns of the language. This is not always the case. Sometimes it can just memorize the answer instead of learning why (just like how a child can memorize the multiplication table without understanding multiplication). This is formally known as overfitting, which is a machine learning 101 concept.
There are ways to mitigate overfitting, but there’s no silver bullet solution. Sometimes it cannot help to memorize the training data.
When GitHub Copilot was new people quickly figured out it could generate the fast inverse square root implementation from Quake. Word for word. Including the “what the fuck” comment. It had memorized it completely.
I’m not sure how much OpenAI has done to mitigate this issue. But it’s a thing that can happen. It’s not imaginary.
No, that is simply a completely misunderstanding of… well the entire concept
That’s not how copyright works either.
In that case feel free to train your LLM without using copyrighted materials. OK?