6 ms·

Open source licenses are no longer strong enough. The copyright laundering use case was not been anticipated when the licenses were written. All licenses need

Open source licenses are no longer strong enough. The copyright laundering use case was not been anticipated when the licenses were written. All licenses need a clause like the following: This software is for humans. AI training is prohibited and carries a default penalty of $1 trillion.

diggan 2y ago
Huh? FOSS licenses work exactly like designed! I'm literally using MIT because I don't give a fuck what people do with the code I publish, limiting it to "humans" or restricting the usage makes it very not FOSS. Sure, if you want to try to prevent AI training by licensing, do that, but it's no longer FOSS, so please don't call it that.
- fsflover 2y ago
  > I'm literally using MIT because I don't give a fuck what people do with the code I publish MIT license requires this: > The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. Do the AI companies follow this requirement?
  - diggan 2y ago
    > Do the AI companies follow this requirement? I haven't seen any LLMs being able to reproduce full copies or even "substantial portions" of any existing software, unless we're talking "famous" functions like those from Quake and such. You have any examples of that happening? I might have missed it
regularfry 2y ago
They are "no longer strong enough" to the extent that GitHub, backed by Microsoft, has had to indemnify their GitHub Copilot customers against copyright claims, and to provide a feature to explicitly prevent open source code from being regurgitated into your codebase. That's not "no longer strong enough". That's a very strong system applying leverage to a powerful actor.
pabs3 2y ago
Such a license would not meet the Open Source Iniative's Open Source Definition: https://opensource.org/osd/ https://opensource.org/osd/
- dithered_djinn 2y ago
  That might indeed apply to open source software. If we instead adopt the view of free software (https://www.gnu.org/philosophy/open-source-misses-the-point.en.html https://www.gnu.org/philosophy/open-source-misses-the-point....), the fact that OpenAI and other large corporations train their large-language models behind closed doors - with no disclosure of their training corpus - effectively represents the biggest attack on GPL-licensed code to date. No evidence suggests that OpenAI and others exclude GPL-licensed repositories from their training sets. And nothing prevents the incorporation of GPL-licensed code into proprietary codebases. Note that a few papers have documented the regurgitation of literal text snippets by large language models (one example: https://arxiv.org/pdf/2409.12367v2 https://arxiv.org/pdf/2409.12367v2). To me, this seems like the LLM-version of using coin-mixing to obscure the trail of Bitcoin transactions in the blockchain. The current situation also reminds me of how the generalization of the SaaS model led to the creation of the Affero GPL license (https://www.gnu.org/licenses/why-affero-gpl.html https://www.gnu.org/licenses/why-affero-gpl.html). LLM's enable the circumvention of the spirit of free software licenses, as well as of the legal mechanisms to enforce them.
  - pabs3 2y ago
    I absolutely agree with you that the current big LLMs enable an attack on all FOSS licenses and especially copyleft ones. That doesn't mean that one couldn't create LLM code generators in a respectful way. Do license analysis on the input code and then train separate models on the different license buckets, with the outputs from each model considered derivative works of the input corpus. Also I don't think a restriction on the FSF's freedom 2 "The freedom to study how the program works" based on what tools you use and how you use them fits with FSF philosophy, nor do I think it is appropriate. You should be able to run whatever analysis tools you have available to study the program. Being able to ingest a program into a local LLM model and then ask questions about the codebase before you understand it yourself is valuable. Or aren't a programmer and or aren't familiar with the language used, then a local LLM could help you make the changes needed to add a new feature. In that situation LLMs can enable practical software freedom, for those who can't afford to pay/convince a programmer to make the changes they want. https://www.gnu.org/philosophy/free-sw.html https://www.gnu.org/philosophy/free-sw.html In addition, OpenAI clearly do not respect copyrights and licenses in general, so would ignore any anti-AI clauses, which would make them ineffective and thus pointless. So, I think we should tackle the LLM problem through the law, and not through licenses. That is already happening with various caselaw in software, writing, artwork etc. It isn't possible or practical to change the existing body of Free Software to use new anti-AI clauses anyway.https://juliareda.eu/2021/07/github-copilot-is-not-infringing-your-copyright/ https://juliareda.eu/2021/07/github-copilot-is-not-infringin... BTW, LLMs could also in theory be used to licensewash proprietary software, see "Does free software benefit from ML models being derived works of training data?" by Matthew Garret: https://mjg59.dreamwidth.org/57615.html https://mjg59.dreamwidth.org/57615.html

diggan 2y ago

Huh? FOSS licenses work exactly like designed! I'm literally using MIT because I don't give a fuck what people do with the code I publish, limiting it to "humans" or restricting the usage makes it very not FOSS. Sure, if you want to try to prevent AI training by licensing, do that, but it's no longer FOSS, so please don't call it that.

fsflover 2y ago
> I'm literally using MIT because I don't give a fuck what people do with the code I publish MIT license requires this: > The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. Do the AI companies follow this requirement?
- diggan 2y ago
  > Do the AI companies follow this requirement? I haven't seen any LLMs being able to reproduce full copies or even "substantial portions" of any existing software, unless we're talking "famous" functions like those from Quake and such. You have any examples of that happening? I might have missed it

regularfry 2y ago

They are "no longer strong enough" to the extent that GitHub, backed by Microsoft, has had to indemnify their GitHub Copilot customers against copyright claims, and to provide a feature to explicitly prevent open source code from being regurgitated into your codebase. That's not "no longer strong enough". That's a very strong system applying leverage to a powerful actor.

pabs3 2y ago

Such a license would not meet the Open Source Iniative's Open Source Definition: https://opensource.org/osd/ https://opensource.org/osd/

dithered_djinn 2y ago
That might indeed apply to open source software. If we instead adopt the view of free software (https://www.gnu.org/philosophy/open-source-misses-the-point.en.html https://www.gnu.org/philosophy/open-source-misses-the-point....), the fact that OpenAI and other large corporations train their large-language models behind closed doors - with no disclosure of their training corpus - effectively represents the biggest attack on GPL-licensed code to date. No evidence suggests that OpenAI and others exclude GPL-licensed repositories from their training sets. And nothing prevents the incorporation of GPL-licensed code into proprietary codebases. Note that a few papers have documented the regurgitation of literal text snippets by large language models (one example: https://arxiv.org/pdf/2409.12367v2 https://arxiv.org/pdf/2409.12367v2). To me, this seems like the LLM-version of using coin-mixing to obscure the trail of Bitcoin transactions in the blockchain. The current situation also reminds me of how the generalization of the SaaS model led to the creation of the Affero GPL license (https://www.gnu.org/licenses/why-affero-gpl.html https://www.gnu.org/licenses/why-affero-gpl.html). LLM's enable the circumvention of the spirit of free software licenses, as well as of the legal mechanisms to enforce them.
- pabs3 2y ago
  I absolutely agree with you that the current big LLMs enable an attack on all FOSS licenses and especially copyleft ones. That doesn't mean that one couldn't create LLM code generators in a respectful way. Do license analysis on the input code and then train separate models on the different license buckets, with the outputs from each model considered derivative works of the input corpus. Also I don't think a restriction on the FSF's freedom 2 "The freedom to study how the program works" based on what tools you use and how you use them fits with FSF philosophy, nor do I think it is appropriate. You should be able to run whatever analysis tools you have available to study the program. Being able to ingest a program into a local LLM model and then ask questions about the codebase before you understand it yourself is valuable. Or aren't a programmer and or aren't familiar with the language used, then a local LLM could help you make the changes needed to add a new feature. In that situation LLMs can enable practical software freedom, for those who can't afford to pay/convince a programmer to make the changes they want. https://www.gnu.org/philosophy/free-sw.html https://www.gnu.org/philosophy/free-sw.html In addition, OpenAI clearly do not respect copyrights and licenses in general, so would ignore any anti-AI clauses, which would make them ineffective and thus pointless. So, I think we should tackle the LLM problem through the law, and not through licenses. That is already happening with various caselaw in software, writing, artwork etc. It isn't possible or practical to change the existing body of Free Software to use new anti-AI clauses anyway.https://juliareda.eu/2021/07/github-copilot-is-not-infringing-your-copyright/ https://juliareda.eu/2021/07/github-copilot-is-not-infringin... BTW, LLMs could also in theory be used to licensewash proprietary software, see "Does free software benefit from ML models being derived works of training data?" by Matthew Garret: https://mjg59.dreamwidth.org/57615.html https://mjg59.dreamwidth.org/57615.html