So much rent seeking. It's the equivalent of becoming famous and suddenly everyone wants to be your friend. There's not really a way around it, it's going to be en vogue for a while, but it should be called out, lest more people really think they're owed something.
Do you have any constructive refutations to the ideas presented in the article or just wanted to to use the term “rent seeking” as its in vogue right now ?
Society is built on copying. Nearly everything humanity does, all of our culture, all of our memes, our language, or alphabet, or behaviors are copied from other people.
Every single one of these ideas I've heard on pay for data on AI could be turned around and used on corporations, and if so, individuals in the US. I've not seen one yet that would not reverse the situation and turn in to a corporation buying up as much IP as possible and turning into, yes, rent seeking.
If some company makes a zillion dollars with AI then tax the living shit out of them and distribute the gains as UBI. If the future is going to AGI the idea that humans will have enough interesting to come up with to actually support themselves seems like a very unlikely premise anyway.
You seem to misunderstand the term rentseeking. If I incur a cost an expense running a service it is not rent seeking, that's just paying for a service. Now if that violates copyright that's a different law altogether.
Now if you paint something once and you want to have a monopoly from everyone event thinking about it for 100 years, we'll, that is rentseeking behavior.
Those posts probably got you some value even if it wasn’t direct compensation.
What value is there now if some AI is going to suck up all your posts and regurgitate them without giving you proper attribution or giving you attribution for something with the context changed just enough to make you look less knowledgeable?
I view it as a form of plagiarism. Search at least has mutual benefit. AI only benefits the owner of the AI.
Well then its a matter of how much of their content was original and bespoke vs pointing to existing documentation. Running a blog vs responding to people on irc might be a good example?
against the near-entirety of the internet, our individual contributions to the training is so infinitesimal that if OpenAI were to cut us out a check it would be for a couple of cents.
All of these articles are worthless, what will decide this isn't some discussion, its the U.S. supreme courts(among others) view on machine learning being fair use or not.
Not necessarily. There should be discussion on whether we want to allow companies to use content to train models, independent of fair use. I mean can’t platforms just say that you’re waiving your content rights when you post on their platform? I think this requires discussion because it’s going to require brand new policy because it’s brand new technology, especially on the legal timescale.
In a healthy democracy, public discourse precedes government rulings. You are suggesting that we, the impacted, should not bother to voice our concerns, should not bother to hone our understanding via dialogue with others, and should instead silently await judgement by deaf authorities? I cannot comprehend the mind of one so blind to the negative effects of autocratic rule that they would chastise their neighbors, "Hey! Don't bother even having an opinion! It's worthless to discuss things! The only discussion of value is happening outside your milieu!"
The suggestion always seems to go in one direction. I don't see people suggesting they should have to pay for the costs they generate in lots of public things and on others (taxes cover a small part of this), only the value they believe they generate.
They then often also complain about socializing costs and privatizing profits, as if they are not literally trying to do the same thing
I wonder if there might already be enough datasets laying around that even if we stopped making new ones, models can generate entire new datasets on their own, process/label them on their own, and generate new and improved models in a closed feedback loop. The only data that's really needed going forward is information on current events.
There is no mechanism to fact check generated data, it could be slowly making a fantasy world for itself. There's already no factuality to ChatGPT, it has no understanding of if the things it says are based in fact.
Just an example, you ask it when a certain variety of grape was introduced in Australia and it just picks a probable Australian town name. Different each time. It doesn't know how to be right in that circumstance.
You do? Salaried workers won't. Casual workers won't. Freelancers might be able to, unless they are hourly (yes you could service more clients, but there are only so many clients, and some might choose to service themselves now that it's so easy)
The Jarod Lanier article this post links to is pretty interesting, though I still don't agree with all of its conclusions. (I don't trust a system that doles out money from OpenAI based on how much my creations train their models any more than I'd trust a plain UBI.) I like this bit in particular (emphasis mine):
>A program like OpenAI’s GPT-4, which can write sentences to order, is something like a version of Wikipedia that includes much more data, mashed together using statistics. Programs that create images to order are something like a version of online image search, but with a system for combining the pictures. In both cases, it’s people who have written the text and furnished the images. The new programs mash up work done by human minds. What’s innovative is that the mashup process has become guided and constrained, so that the results are usable and often striking. This is a significant achievement and worth celebrating—but _it can be thought of as illuminating previously hidden concordances between human creations_, rather than as the invention of a new mind.
34 comments
[ 4.3 ms ] story [ 33.2 ms ] threadEvery single one of these ideas I've heard on pay for data on AI could be turned around and used on corporations, and if so, individuals in the US. I've not seen one yet that would not reverse the situation and turn in to a corporation buying up as much IP as possible and turning into, yes, rent seeking.
If some company makes a zillion dollars with AI then tax the living shit out of them and distribute the gains as UBI. If the future is going to AGI the idea that humans will have enough interesting to come up with to actually support themselves seems like a very unlikely premise anyway.
The scale and cadence of it seems to be different.
Also, once again Karl ideas of “ownership of the means of copying and training” seems like it will be relevant.
I have nothing cleaver to say; I agree with you. Just wanted to add some colors.
I find it sad that IP laws can be enforced ruthlessly in a case. And fully ignored in another.
Actually, musing; I have something to say:
We could force AI manufacturers to reveal what the data underlying the training; and use IP at that stage to effect some repartitions.
Raws BOM-style list of training data should not reveal their secret sauce? ( actually question )
The English alphabet is not the Phonecian alphabet. That is an undeniable fact.
Building on something doesn't make it a copy. You blur the relations between words. It is unwise.
Now if you paint something once and you want to have a monopoly from everyone event thinking about it for 100 years, we'll, that is rentseeking behavior.
Does most people's dormant project code on GitHub have any value before enriched and incorporated into an ML model?
The marginal cost of an idea is zero, and the world is turning into pure thought.
No.
Stop rent seeking.
What value is there now if some AI is going to suck up all your posts and regurgitate them without giving you proper attribution or giving you attribution for something with the context changed just enough to make you look less knowledgeable?
I view it as a form of plagiarism. Search at least has mutual benefit. AI only benefits the owner of the AI.
Positively medieval serf outlook.
They then often also complain about socializing costs and privatizing profits, as if they are not literally trying to do the same thing
Just an example, you ask it when a certain variety of grape was introduced in Australia and it just picks a probable Australian town name. Different each time. It doesn't know how to be right in that circumstance.
I'm in favor of universal basic income, but paying people for their "data" is not the path.
A rising tide lifts ALL boats in the end. All of them. No exceptions. Just let it happen.
>A program like OpenAI’s GPT-4, which can write sentences to order, is something like a version of Wikipedia that includes much more data, mashed together using statistics. Programs that create images to order are something like a version of online image search, but with a system for combining the pictures. In both cases, it’s people who have written the text and furnished the images. The new programs mash up work done by human minds. What’s innovative is that the mashup process has become guided and constrained, so that the results are usable and often striking. This is a significant achievement and worth celebrating—but _it can be thought of as illuminating previously hidden concordances between human creations_, rather than as the invention of a new mind.