Bwahahahahahahahahahahahahaha!!! Oh, you weren’t kidding about corporate America being ethical? Sorry, bro.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
OpenAI Calls on U.S. Government to Let It Freely Use Copyrighted Material for AI Training
- Thread starter MacRumors
- Start date
- Sort by reaction score
It's almost unbelievable that OpenAI would claim that what's holding back AI from being more accurate and "knowledgeable" is the restrictions on free use of copyrighted content. What's holding back AI is its over-reliance on the LLM component. It's going to have to achieve real artificial general intelligence before it's actually AI, or at least developers need to improve how LLMs work. That requires more research, developers, etc., not simply increasing the industry's access to still more material to feed the ever-hungry maw of LLMs.
Last edited:
The logic is simple: If a human can use any copyrighted material to learn, an AI ('digital human') should be able to do the same. If it's posted on the web for public view, it should be available to AI under the same rules that apply to humans.
What do you think?
What do you think?
Then learn how an LLM works. Realize that it at the heart it is just a very sophisticated word generating system that works on the statistical occurrences of words in the corpus with which it is trained. It isn’t a database that stores the work. It does not verbatim spit out copies of copyrighted material.
If you are going to ban the training of an artificial neural network with copyrighted text, then you should do the same for the biological neural network in your brain. Your brain contains a neural networks trained on all the text you’ve ever read. The primary difference is that the ANN has a better memory. That is, its synapses (parameters) stay fixed once it has been trained. The read-modify-write mechanism of human memory is lossy and error prone.
Last edited:
Worse than that - it’ll be an interpretation of an interpretation of a copy.
Oh no! The plagiarism machine builders are upset they aren’t allowed to plagiarise!
Laws are only for the poor. If you’re wealthy enough you can just about do what you want.
Laws are only for the poor. If you’re wealthy enough you can just about do what you want.
You already do when you use knowledge you have gained from copyrighted textbooks and videos that have been stored in the neural networks of you brain. This is the same thing that artificial neural networks do. Computers simply have better memory systems than you.
Last edited:
Newsflash. People are animals. Read a copyrighted introduction to biology textbook. You’ll be amazed at what you find.
Why do you think the owner of one of those sources is currently combing through basically all the factual information the U.S. federal gov't has?
On its face, it certainly seems a fair proposal to require a company to buy a copy of any book they use in their training corpus. It is somewhat consistent with the way humans train their biological neural networks, excepting those pesky public libraries that allow people free access to copyrighted material.
As someone who has copyright material in the library of congress; go pound sand, OpenAI.
I can't wait until it's just X, Insta and FB bots talking to each other online, and actual humans unplug and go outside and talk with each other again.
Change your name to GhanwanAI. That’s it. That’s all you have to do.
The Chinese will do what they want to do for AI training, regardless of any US laws.
You say that as if the creators of the works (or their estates) aren't paid. People pay taxes, that are then used to buy books that go into libraries. Read one some time and you'll find that out.
The "LLMs are just like people reading books" is a complete fallacy. Repeating it over and over won't make it true.
If there wasn't a difference, these companies wouldn't be asking for a way around copyright laws.
I agree with the abolishment of copyright; people have shown to be more than willing to pay for the experience of going to museums, concerts, the movies - and the convenience of having immediate access to music and film. And they actually enjoy paying what they deem fair to keep artists going. Patents is not the same thing since they give businesses the ability to develop products openly - but the system should still be overhauled since it's currently being massively abused by patent trolls.
If they access the content in the same manner as everyone else; through public libraries, purchasing books, music, graphic material etc. it's literally no different from humans learning by consuming art, science etc. The networks are trained in how the published work is put together, averaging all the input and creating new content based on that. Exactly like carbon based neural networks do.
They’re not doing that though. They’re already training against copyrighted/published material and seeking legal cover after the fact.
They’re training on the output of workers, artists, academics, and scientific papers in an attempt to create tools that are “good enough” to devalue labor ever further.
So I can legally monetize copyrighted /trademark content for myself? Thanks for the tip.
It's a false analogy to try to equate individual human learning using copyrighted material, to AI/LLM learning using the same material.
At the simplest level, if a human consumes copyrighted material, often they're paying for it, but the AI companies seem to want to replace "often" with "seldom if ever". Makes a big difference in how well people are reimbursed for creating intellectual property.
It's true that once a human learns from copyrighted material, they can tell others what they've learned, either one-to-one or on a larger scale, by becoming a teacher, lecturer, authoring their own works, making YouTube videos, etc. All of that is normal and fits within the frameworks we're familiar with, and keeps in place the methods we've used for many years to reimburse people for their work.
But AI/LLM distribution of knowledge is very different. When a publicly-available LLM like ChatGPT, Gemini, Grok, etc. learns from copyrighted material, it then mass-distributes what it's learned to any number of people who access it, unbound by the individual human limitations on distribution of knowledge that we've operated with until recently. While that might sound wonderful in a sci-fi sort of way, it nearly completely upends the frameworks we've long had in place to ensure that the creators of that material are compensated for their work.
A lot of people (including myself) like the idea of having artificial entities we can sometimes call on to answer questions, do some things for us, etc. (as long as they get things right), but not if it means that the people who create the works that these entities learn on, are no longer compensated, or poorly compensated. Something else needs to be worked out than this.
At the simplest level, if a human consumes copyrighted material, often they're paying for it, but the AI companies seem to want to replace "often" with "seldom if ever". Makes a big difference in how well people are reimbursed for creating intellectual property.
It's true that once a human learns from copyrighted material, they can tell others what they've learned, either one-to-one or on a larger scale, by becoming a teacher, lecturer, authoring their own works, making YouTube videos, etc. All of that is normal and fits within the frameworks we're familiar with, and keeps in place the methods we've used for many years to reimburse people for their work.
But AI/LLM distribution of knowledge is very different. When a publicly-available LLM like ChatGPT, Gemini, Grok, etc. learns from copyrighted material, it then mass-distributes what it's learned to any number of people who access it, unbound by the individual human limitations on distribution of knowledge that we've operated with until recently. While that might sound wonderful in a sci-fi sort of way, it nearly completely upends the frameworks we've long had in place to ensure that the creators of that material are compensated for their work.
A lot of people (including myself) like the idea of having artificial entities we can sometimes call on to answer questions, do some things for us, etc. (as long as they get things right), but not if it means that the people who create the works that these entities learn on, are no longer compensated, or poorly compensated. Something else needs to be worked out than this.
Last edited:
But the proposal by OpenAI doesn't seem to include purchasing books, music, graphic materials, etc. from their creators, at least not in the same ways or at the same monetary levels as humans, using the systems that have been in place for many years to properly compensate the creators of those works. And the mass distribution of information that AI enables, unbound by copyright, is one of the things that makes AI literally very different from how humans distribute what they've learned.
And of course they’ll be publishing their copyrighted source code for others to learn from, right?