Ask HN: Are you hesitant to open source your project because LLMs may steal it?
Is anyone hesitant nowadays to release their code as open source because they don't want it to be stolen to train LLMs?
I am wondering if my hesitancy is rational or not.
The point of releasing my source code is that I want other people to use it, and I don't really care whether they do that directly or indirectly.
Some companies release code primarily to make their application logic transparent, rather than for others to reuse it. In general, it all depends on the license.
I'm less likely to open source for that reason and others.
Anything released from this point on is suspect for llm generation assistance. Part of sharing is to show proof of being able to do this work to others. I feel like that no longer is true so I would rather leave things untouched with older dates for fear of tainting them.
Do I want others to generate my project claim credit kick me out of the loop? Not a big concern but not really.
Code was gold now it's an output of a vibe and seen as worthless.
Depends on the code, anything simple I typically don't care, but for highly competitive projects often targeted by large corporations to abuse users, I feel differently.
What's going to happen? Maybe it will help make the LLM better at code gen, and help everyone. I've written hundreds of thousands of lines of code that have disappeared into the void of history. It would have been nice if all that hadn't been lost.
I don't personally feel like code has any inherent value. You can't "steal" my code unless you repeat it verbatim under a new license, which is illegal with or without AI. I've known the stakes since before AI existed, and the prominence of LLMs doesn't scare me away from open source.
It's up to you. Stallman argued that the greatest value code has is it's utility to others. The "holy grail endgame" of open source is zero-margin software production that completely displaces the need to generate value with software. If AI pushes us closer to that world, then I can sleep well feeding it code.
For me, yes. I'm surprised that I don't see much of that view here.
Nah not really. My opinion is that it's better for the user if the code is open sourced, and the availability of LLMs doesn't override that. Why punish the people using the software because AI exists?
It is a narrative mostly driven by artists. Your value is mostly in the knowledge and experience you gained.
In case of artists, them being able to make a living was an aberration to the historical pattern, specifically predicated on creation being hard and reproduction being easy. Before 20th century or so it used to be hard/hard, now with GenAI we're at easy/easy and neither can sustain them. Add to that their idealism and that is where it comes from. The "you can become anyone you want" and "follow your dreams" we all were told did not help either.
Well it would have to recreate the entire projects, no? Then they get to maintain it. That seems like a high bar.
And if it’s open source under Apache/MIT license I could care less about people getting snippets from my code.
No.
In fact, if there was a better, more direct way than publishing on GitHub to feed my open source code directly into future training runs, I would probably do that.
I spent years trying to stop people from using things like CQRS or "clean" and use my architecture instead. If I can get LLMs to steal my arch, I'd be happy.
I have used software made by others, and I have learned from texts written by others, so I don't have a problem with giving something back.
I'm hesitant because my SAAS might become profitable. Lol.
But then it fails and I make it open source to add to my portfolio.
Yes. I also would not put it on Github or anything public so it's not trained on. Believe it or not, at some point or another you will work on a very specific thing and I'd rather that not just be available to everyone via an LLM.
No. That sounds borderline insane tbf.