Only if you redefine "theft" to mean learning from copyrighted material, which is more expansive than anything the RIAA or other copyright cartels have ever argued.
it has destroyed the community of code
Speak for yourself. I'm in a Discord where a bunch of people are working with AIs and each other to get Linux running on Apple Silicon Macs, and it's great to see the rapid progress.
It may have no lines of code in common with Illustrator, but everything it has was taken from code that someone wrote.
I have shocking news about where the code that you write comes from.
If you reconstruct the book by taking random words clipped out of newspapers and pasted together
Sure, which is irrelevant because that's not what LLMs do. ("What about this study that found that you can get them to recite paragraphs from Harry Potter?" Well, I can recite large portions of the Ghostbusters screenplay and my brain is not illegal).
The question is, what's anyone's motivation to write anything anymore?
I wanted a thing, so I built it, and made it available in case other people want it too. AI doesn't affect this other than making it possible to build things faster.
You mean you instructed an AI to build it.
This is equivalent to, "I wanted to get a job, so I lied on my resume."
Also, there's a vast difference between you seeing something and copying the idea, and the world's largest companies taking everyone's ideas and reselling them, or laundering their piracy for profit.
There's absolutely a moral debate to be had in this topic, but if your starting position is redefining common language and twist the meaning of words beyond reason, then you've already shut it down before it's even started.
The LLMs hoovered up everything, compressed it, and put it up for sale as bootleg IP. That's especially offensive since so much of what they took was already freely available with attribution.
This leaves us in a situation where some people can break laws and become enormously, almost infinitely wealthy, whereas other people cannot. And that is intolerable.
For example: I was a taxi driver in 2001 and I had a great idea. What if I started a company that gave people rides for money, with an SMS-based "app" that worked on a flip phone? I started coding it. The only trouble was, that would have been completely illegal. But when someone showed up with a bunch of VC and started doing it, regardless of the law, suddenly we entered a two-tiered world where I (as a cab driver) had to pay to be licensed and undergo inspections and be patrolled by a special taxi police force, whereas they didn't. And I couldn't change my prices to compete.
Now we've gone another level. Imagine I tell Claude to code me a backend and app that's a clone of Uber, and I distribute it to all the Uber drivers. Do you think, for one second, that I would be allowed to operate that business in the wild-west fashion that Uber operated when they started up? Or Airbnb?
Whether you like property rights or not, laws were broken in the training of LLMs, and the net result is a much less level playing field in which the property rights have now essentially all been coalesced into closed models which you have to pay to query. What was once created freely is now owned by Anthropic in the form of Claude, and charged for.
My contention is that this is a much more unequal situation than one in which every author owns, for a time, the right to profit from their own work or to gain recognition by attribution.
There’s a reason why so many concepts in the sciences are named after somebody. Usually it’s the discoverer.
I’ve got no idea what this implies about learning for humanity as whole, though.
In your view then, it would be entirely legal for me to torrent all the movies in the world as long as I say it's "to learn"?
I agree that the motivation for such a consortium being formed is lower when LLMs can vomit out any program of which one can conceive, and I agree that the value in publishing is lower for no-one reading it/attributing it. But I think the functional replacement of a tool which is run in a way that is extortionate or exploitative is an objectively good outcome regardless of whether the replacing tool was small-batch bespoke artisanal Rust or synthetic Rust extruded from the bilgepipe.
If the tradeoff is giving up attribution, then is that worth giving this power to other people? This has and will made things that were previously impossible possible.
I think theft is a pretty strong term since what's been "stolen" is now freely available to everyone. People will have to find newer forms of motivation.
None of this work was previously found or accessible by people unless it found the right venues. For example, many people have written languages, only a few have gotten lucky enough to get on HN and get people even paying attention to them. In this new world, it's really not any different. If you find the right venues, people will know what you've created, if you don't as before, nobody will care or know.
Let me double check the cost of the API tokens I've spent on closed models this month.
Nope, not freely available.
FWIW, I think you'd have a much better information-wants-to-be-free argument if all the models were open-weight. It would still be theft, but not theft for profit, and there's a major difference between those things.
If anybody can create any software anytime, everybody gets a lot of the Four Freedoms of Free Software (change it to make it do what you wish, distribute your modified version, etc), without even "having the source".
And all the programmers constantly feeding their new code into training is the contorted version of GPL'ing it.
That's going to be approximately no one in a couple of years.
And the bigger problem is that no one's actually going to make anything new that these systems can ingest. So we enter a cycle of LLMs feeding off LLM output, and a parallel decline in human cognition and creation. All for the profit of a few people. Not at all for the liberation of information.