Ask HN: Would open-source devs accept financial compensation for AI training?
2 comments
There is already legal precedent in the US that LLM training is a version of reading/studying akin to what people do, is not a form of wholesale duplication, and involves a great deal of transformation at significant effort and expense.
So the copyright ship has sailed. Current lawsuits are about how copies were obtained or how materials produced by an LLM are themselves infringing. But training? That's all open and legal already.
So the copyright ship has sailed. Current lawsuits are about how copies were obtained or how materials produced by an LLM are themselves infringing. But training? That's all open and legal already.
Thanks! would be interesting to know how this is in other countries.
I don't think an implementation has to be about copyright infringement by the training itself, though. Like you say, it can also be about materials produced. And my question is really if people would like it, not how feasible it is to implement.
I don't think an implementation has to be about copyright infringement by the training itself, though. Like you say, it can also be about materials produced. And my question is really if people would like it, not how feasible it is to implement.
I guess we can be sure that LLMs are massively trained with open-source code. And while IANAL, I'm convinced that the training is not covered by any open-source license I know (except those that mimic public domain maybe). The FSF also raised concerns about this lately because at the very least copyright attribution would go missing. And copyleft licenses are completely ignored.
There's a precedent for the use of unlicensed, copyright-protected work: private copying levies aka. blank media taxes. Not as old as copyright itself but as old as the audio cassette ;) The idea was that musicians would lose income because of private copies of their works, and should be compensated. So for every blank cassette sold a little amount of money goes to "copyright collectives" that then spread it among their members. Today such levies exist in many copyright-friendly countries, for digital media and devices.
I don't know how feasible it is to establish something similar for open-source code. However, I know things are being seriously discussed right now for other creators, e.g. journalists, musicians... But more importantly, I wonder: Is this something other open-source devs would want? Personally, I'd prefer getting paid with more contributions not money (as a firmware dev, I'd prefer hardware documentation :D ). But if somebody else found an unexpected (IMO sometimes cool) way to monetise our work, creators should get paid. Not just open-source devs, all creators. But right now it seems everybody else is lobbying and open-source developers are left behind.
More detailed questions (don't try to answer them all):
* Would you accept financial compensation for AI training?
* Would you accept financial compensation for AI training if you have to join a copyright collective?
* Would you accept financial compensation for AI training if you have to join a copyright collective, and AI training with open-source code would be legalized explicitly?
* What kind of services / products should be taxed? Code generating service? Code generating agent? Device capable of running a suitable LLM? Any device with a GPU? Any device with an NPU? (Device with) AI OS that creates its apps on demand?