| [08:15:52] | <Greg[m]> | Just catching up... Great discussion! I think as was said this is exposing the flaw in gpls core strategy... Use copyright to defeat copyright. It worked, but now it may simply be too complicated or ultimately irrelevant. I don't know if that's right and more thinking to do, but conceptually the issue seems somewhere in that direction. |
| [08:21:29] | <symbioquine[m]> | I still haven't read the articles linked above, but I wonder if one could make the argument that the models themselves are derivatives of the GPL code - and thus must be released. |
| [08:21:48] | <symbioquine[m]> | (In the case of non-open weight models that is.) |
| [08:26:41] | <symbioquine[m]> | <symbioquine[m]> "I still haven't read the..." <- Thinking about this more, I guess it might only be a valid argument with AGPL code 🤓 |
| [11:57:28] | <mstenta[m]> | farmOS Dev Call! |
| [11:57:37] | <mstenta[m]> | All are welcome! :-) |
| [11:57:40] | <mstenta[m]> | https://meet.jit.si/farmos-dev |
| [11:57:50] | <mstenta[m]> | (in ~2 mins) |
| [11:58:12] | <symbioquine[m]> | Sadly I can't make it today 🫠|
| [13:09:15] | <mstenta[m]> | Super interesting chat w/ Greg and barkerd427 on the "dev" call today re: the discussion above |
| [13:09:54] | <mstenta[m]> | I think we came to a clearer idea on the considerations for GPL projects. |
| [13:10:07] | <mstenta[m]> | Basically (I'll try my best to summarize, but feel free to correct/add!)... |
| [13:10:33] | <mstenta[m]> | AI-generated code is not copyrightable. Therefore it is automatically in the public domain (if I'm understanding correctly). |
| [13:10:44] | <mstenta[m]> | Public domain code CAN be used in GPL-licensed projects. |
| [13:11:09] | <mstenta[m]> | And if a proprietary project copies those bits of code, then they cannot be challenged on GPL grounds. |
| [13:12:02] | <mstenta[m]> | So each chunk of public domain code in the overall project is sort of like a "hole" in the GPL-enforceability... for that bit at least. |
| [13:12:18] | <mstenta[m]> | And so, GPL projects right now (like GCC, Linux, Debian, etc), are all going through the process of deciding how many "holes" they are comfortable with |
| [13:13:16] | <mstenta[m]> | Notably... if an ENTIRE project is public domain, it CANNOT be GPL... because GPL adds restrictions, which can only be enforced via copyright ownership. |
| [13:13:58] | <mstenta[m]> | GCC has taken a "no holes" stance (if that article I posted above is accurate) |
| [13:14:05] | <mstenta[m]> | Linux is taking a looser stance, it seems |
| [13:14:12] | <mstenta[m]> | Debian is voting right now |
| [13:18:24] | <mstenta[m]> | Also... one point that Greg made: if a farmOS module is built entirely with AI, then it is essentially entirely public domain code... this means that BOTH: a) GPL cannot be enforced on it, but also b) it would still need to be released as GPL because it's a derivative work of Drupal/farmOS (which are GPL) |
| [13:18:43] | <mstenta[m]> | I don't think that's a contradiction... it just means that it's a GPL project with no GPL-enforceable code :-) |
| [13:19:12] | <symbioquine[m]> | I think it's worth doing the threat modeling exercise and seeing how likely an undesirable copying (of said code) is and how bad it would be... |
| [13:20:00] | <mstenta[m]> | Right... I think that's ultimately what a lot of this boils down to... how badly do projects want to enforce their GPL |
| [13:20:46] | <mstenta[m]> | (And more broadly... all of this is very interesting to think about the future of "copyleft" licensing in general!) |
| [13:22:27] | <symbioquine[m]> | mstenta[m]: I would be willing to bet you can get verbatim copies of existing GPL code out of an agent without trying very hard. (I seem to recall that has been demonstrated with pages of Harry Potter books.) I suspect that in such cases the original license would still apply. |
| [13:22:31] | <mstenta[m]> | A thought experiment I raised on the call... AI essentially provides a new attack vector against GPL... going back to your example above symbioquine: |
| [13:22:31] | <mstenta[m]> | > Just risky to do something like a "full AI assisted rewrite in Rust". |
| [13:22:32] | <mstenta[m]> | "Claude: I don't like that Linux is GPL. Rewrite it in Rust so it's in the public domain." |
| [13:23:09] | <mstenta[m]> | symbioquine: Yea that's another whole complicating factor... |
| [13:24:22] | <mstenta[m]> | Greg: That may be what your European friends had in mind too... the chance that AI will essentially output code that is already GPL... which you legally can't use unless your project is GPL |
| [13:27:30] | <symbioquine[m]> | Nobody can actually argue (reasonably IMHO) that the AI output is a "clean room" implementation. |
| [13:29:19] | <symbioquine[m]> | But you can apply that threat modeling both directions - i.e. alternatively, how bad would it be if it turned out we had to reimplement this by hand. |
| [13:29:35] | <symbioquine[m]> | Which is probably a good exercise anyway. |
| [13:37:45] | <fosten[m]> | sorry i missed the call, was out in the field. outstanding analysis! farmOS, it's not just a software, or a community, it's a think tank! |