Hacker News Reader

Comment by andy99 | Hacker News Reader

Curious to know if this extends to LLMs and if so how they would define open source. Specifically it would be nice to see repudiation of Meta's "Open" BS by a nation state.

View on HN

https://www.comparia.beta.gouv.fr/modeles compares models and Llama different licenses are not mislabeled as "open source".

Also, https://opensource.org/ai/endorsements shows code.gouv.fr in the list.

Cool, thanks!

  Cette licence permet d'utiliser, reproduire, modifier et distribuer librement le code avec attribution, mais impose des restrictions pour les opérations dépassant 700 millions d'utilisateurs mensuels.

Interesting they only mention the 700 million users thing and not the other restrictions on use. Personally I could regard the prohibition against basically Google and Microsoft using it to be a minor transgression, it's the larger list of unacceptable uses that's the big problem.

Agreed. If you feel the need to report inconsistencies, the source code is here: https://github.com/betagouv/ComparIA

It's actually targeting Apple since both Google and Microsoft have their own models.

I'm a little confused by the context of this since I don't speak French, but it seems like you're unfamiliar with the OpenELM family of models.

https://huggingface.co/apple/OpenELM

that looks like a dinosaur now

OSI's OSAID is complete and utter bullshit. Look to Debian if you want a real definition of open source AI.

https://opensource.org/ai https://salsa.debian.org/deeplearning-team/ml-policy

https://elevenfreedoms.org/

> The traditional Four Freedoms of free software are no longer enough. Software and the world it exists in have changed in the decades since the free software movement began. Free software faces new threats, and free AI software is especially in danger.

They're just those eight guidelines. Not particularly precise, with intent mattering more than any definition. This isn't a policy, just a goal.

I wouldn't call data "source", whether a book, a sound track, video, etc.

In my view of the world, the code to train, the software to run, that's open source joy.

Now... should the trained, and vectored data be free? Maybe so.

But I bet this UN thing doesn't cover that.

I didn't call the data the source and in the past have explicitly argued that training data is not necessary to exercise the freedoms normally associated with open source.

Llama models have usage restrictions that go against any mainstream definitions of open source.

The model is part of the data, agreed?

Anyhow, I wasn't trying to put words in your mouth, simply stating my thoughts. And I focused on data because I see OSS code everywhere, so presume there is no issue there.

Reply on news.ycombinator.com