upvote
All things are built on knowledge that came before it. In what way is AI built on stolen labor?
reply
Has anyone who contributed to producing the wealth of human knowledge, art, code, and literature making up the training data been compensated in any way?

I'm certainly still waiting for my check.

reply
Well since the early 18th Century (in a limited way) copyright rights have been given to creators - the overwhelming majority of countries are signatories to international treaties (for the last 100 years) that give creators control of created artistic works (which has been deemed to include program code).

Exploitation of those rights is down to individuals, only since Google came on the scene has any company been able to get away with broad ranging infringement of these rights without censure.

As for your cheque, you have the right to sell your work. No-one is obliged to buy it, of course. You also, under copyright, have the right to make a license to exchange use of your works for binding the users to conditions of release of their derivative/associated works ([F]OSS licensing).

You still have the right to sell it, and if ML models have been trained on those works you have the right to seek compensation.

reply
That already didn't happen, regardless of AI. The thing about freely sharing knowledge is that people take that knowledge and use it to build further knowledge that helps you.

You might say that the fact LLMs are being provided as services by big techs invalidate that. But some people taking the provided knowledge for themselves has also always happened. More importantly, though, is that big tech provided LLMs aren't the only way to access the technology. Open weights and local models exist.

reply
Casting this as “knowledge” betrays a very fundamental lack of knowledge about how LLMs work, or a desire to mislead for whatever reason.

They are not databases of facts. Even if they were, that would still constitute a very serious violation. They aren’t, and they’re not particularly good at serving that function either.

No, LLMs are capable of reproducing things (text, art, other media) based on their training data. Consider being an artist who produced something original that somehow made it into that training set - your chances of actually making a living are now vastly smaller.

The result is the same as if there was no copyright at all: Big companies will just exploit whoever, no art would be produced.

This is a huge issue, and there needs to be a reckoning.

reply
Well, it was literally stolen in some cases, for starters.
reply
How do you steal knowledge? Knowledge shouldn't be locked up.
reply
I hope everything you make is public domain, then. Otherwise that would be hypocritical.
reply
One thing is that most if not all models more likely than not were trained on leaked Microsoft and other proprietary code sets.

For projects like WINE or ReactOS that absolutely have to enforce hard firewalls between people with "insider knowledge" to avoid contamination and resulting legal issues, this alone makes the use of AI for coding highly problematic.

reply
If you believe that, then why aren't LLMs free to use for every person on Earth?

If you want anybody to keep making anything, including programmers, artists, and writers, you're going to have to pay them somehow.

Anthropic and OpenAI would not have a product at all without the contributions of thousands of non-consenting humans. Yet they somehow have the gall to charge customers a premium.

reply
Anthropic and OpenAI aren't the only ways to use LLMs. There are plenty of LLMs where you can download the weights and run them locally. There are even others that you can train yourself (see the LLM360 family).
reply
You're paying for compute?
reply
They refuse to release the models, the training method and, training data. They don't even want you to ask it too many questions lest you teach another ai. Very "open".
reply
I'm all for it, free AI allotment and universal income is the future.
reply
It’s not a future that’s currently in the cards, no.
reply
[dead]
reply