'Meta's Newly Released Large Language Model Llama-2 Is Not Open Source' 27
Earlier this week, Meta announced it has teamed up with Microsoft to launch Llama 2, its "open-source" large language model (LLM) that uses artificial intelligence to generate text, images, and code. In an opinion piece for The Register, long-time ZDNet contributor and technology analyst, Steven J. Vaughan-Nichols, writes: "Meta is simply open source washing an open but ultimately proprietary LLM." From the report: As Amanda Brock, CEO of OpenUK, said, it's "not an OSI approved license but a significant release of Open Technology ... This is a step to moving AI from the hands of the few to the many, democratizing technology and building trust in its use and future through transparency." And for many developers, that may be enough. [...] But the devil is in the details when it comes to open source. And there, Meta, with its Llama 2 Community License Agreement, falls on its face. As The Register noted earlier, the community agreement forbids the use of Llama 2 to train other language models; and if the technology is used in an app or service with more than 700 million monthly users, a special license is required from Meta. Stefano Maffulli, the OSI's executive director, explained: "While I'm happy that Meta is pushing the bar of available access to powerful AI systems, I'm concerned about the confusion by some who celebrate LLaMa 2 as being open source: if it were, it wouldn't have any restrictions on commercial use (points 5 and 6 of the Open Source Definition). As it is, the terms Meta has applied only allow some commercial use. The keyword is some."
Maffulli then dove in deeper. "Open source means that developers and users are able to decide for themselves how and where to use the technology without the need to engage with another party; they have sovereignty over the technology they use. When read superficially, Llama's license says, 'You can't use this if you're Amazon, Google, Microsoft, Bytedance, Alibaba, or your startup grows as big.' It may sound like a reasonable clause, but it also implicitly says, 'You need to ask us for permission to create a tool that may solve world hunger' or anything big like that." Stephen O'Grady, open source licensing expert and RedMonk co-founder, explained it like this: "Imagine if Linux was open source unless you worked at Facebook." Exactly. Maffulli concluded: "That's why open source has never put restrictions on the field of use: you can't know beforehand what can happen in the future, good or bad."
The OSI isn't the only open-source-savvy group that's minding the Llama 2 license. Karen Sadler, lawyer and executive director at the Software Freedom Conservancy, dug into the license's language and found that "the Additional Commercial Terms in section 2 of the license agreement, which is a limitation on the number of users, makes it non-free and not open source." To Sadler, "it looks like Meta is trying to push a license that has some trappings of an open source license but, in fact, has the opposite result. Additionally, the Acceptable Use Policy, which the license requires adherence to, lists prohibited behaviors that are very expansively written and could be very subjectively applied -- if you send out a mass email, could it be considered spam? If there's reasonably critical material published, would it be considered defamatory?" Last, but far from least, she "didn't notice any public drafting or comment process for this license, which is necessary for any serious effort to introduce a new license."
Maffulli then dove in deeper. "Open source means that developers and users are able to decide for themselves how and where to use the technology without the need to engage with another party; they have sovereignty over the technology they use. When read superficially, Llama's license says, 'You can't use this if you're Amazon, Google, Microsoft, Bytedance, Alibaba, or your startup grows as big.' It may sound like a reasonable clause, but it also implicitly says, 'You need to ask us for permission to create a tool that may solve world hunger' or anything big like that." Stephen O'Grady, open source licensing expert and RedMonk co-founder, explained it like this: "Imagine if Linux was open source unless you worked at Facebook." Exactly. Maffulli concluded: "That's why open source has never put restrictions on the field of use: you can't know beforehand what can happen in the future, good or bad."
The OSI isn't the only open-source-savvy group that's minding the Llama 2 license. Karen Sadler, lawyer and executive director at the Software Freedom Conservancy, dug into the license's language and found that "the Additional Commercial Terms in section 2 of the license agreement, which is a limitation on the number of users, makes it non-free and not open source." To Sadler, "it looks like Meta is trying to push a license that has some trappings of an open source license but, in fact, has the opposite result. Additionally, the Acceptable Use Policy, which the license requires adherence to, lists prohibited behaviors that are very expansively written and could be very subjectively applied -- if you send out a mass email, could it be considered spam? If there's reasonably critical material published, would it be considered defamatory?" Last, but far from least, she "didn't notice any public drafting or comment process for this license, which is necessary for any serious effort to introduce a new license."
Open (Score:5, Insightful)
Not open source but you get the weights for free, you can fine-tune them, distribute them, and can use them for commercial uses as long you have less than 700M active daily users users. So, ironically, Meta's model is a billion times more open than OpenAI's shit, which was founded with the mission statement of making open models. Confused yet?
Re: (Score:2)
Well other than the fact OpenAI does not seem to object to people using ChatGPT to train other models.(Including Llama derivitives like Alpaca). Not sure what OpenAIs TOS says, but they havent been stopping people doing it.
Re:Open (Score:4, Insightful)
Nah, that's business as usual over here in Europe. We don't have a list of freedoms but in general we're freer than the US who does have one.
Re: (Score:2)
https://www.citizensinformatio... [citizensinformation.ie].
Re: (Score:2)
This is true, but don't give Meta too much credit. Their weights are available because someone leaked them.
Re: (Score:2)
Not for LLaMA-2, and it's under a much more permissive license
Re: (Score:2)
Sure. And Facebook totally didn't decide they were all about open source when their first version ended up all over the internet. And that completely didn't affect their decision to also release the more tuned updated version.
Re: (Score:2)
I'm struggling to see your point. I never said it didn't affect their decision. What are you talking about?
Re: (Score:2)
I don't think you read my original response carefully. I said simply "their weights are available because someone leaked them." You responded "not for llama-2." Perhaps you understood that to mean someone leaked the llama 2 weights, but that's not what I said.
Facebook's decision to distribute the weights for their large language models was in response to someone doing it for them on bittorrent. Thus "don't give them too much credit."
Re: (Score:2)
But LLaMA-2 weights are totally different from LLaMA-1 weights, and way better. They could've easily been more restrictive with 2 and benefited, not less restrictive.
OpenLlama (Score:3)
Meanwhile, the OpenLlama project distributes their model under a true open source license.
It is still debatable whether they have the legal authority to claim copyright over their model, and thus give it an open source license. Training a model that can recite sections of Wikipedia articles verbatim when asked is a massive 'fair use' issue, as you need to be able to credit the contributors. Language models mix everything up and can't tell you where anything came from.
Are we just splitting hairs here? (Score:4, Interesting)
It's open-source for all individuals and small-to-medium organizations. And large organizations are free to examine the source to their hearts' content, just not to commercialize it.
The perfect is the enemy of the good, and that is certainly open enough for me.
Sadly, since it's Meta code, I can't get very excited. I can't think of a single time their team was successful at making anything remotely approaching revolutionary.
Re: (Score:2)
It's open-source for all individuals and small-to-medium organizations. And large organizations are free to examine the source to their hearts' content, just not to commercialize it.
To be honest... That is exactly what I think open source should be.
Re: (Score:2)
There's a lot of blah blah over what "open source" means. And those italics absolutely represent histrionic nerd hand wringing. The OSI kind of sort of says you shouldn't limit who can use it, although maybe you can limit what they use it for? GNU kind of sort of said you should limit what people should use it for, then went all in and said you definitely should.
Keep it simple. If you can read the source, then it's open source.
Now there's just the problem of people applying software terms to things that are
Open Source Won - Now it is time to betray it (Score:2)
Solved on Wikipedia (Score:4, Informative)
This problem has been solved on Wikipedia: https://en.wikipedia.org/wiki/Source-available_software [wikipedia.org].
Companies that persist in confusing the industry by not adopting this term are being deceptive.
Re: (Score:3)
I’d counter that “open source” was unclear to begin with, no clearer than the “free software” term from which it was trying to distance itself.
It was originally coined in 1998 to draw a distinction from the free software movement’s perceived and actual issues that were keeping it from widespread adoption. Unfortunately, just like “free” is a confusing term for most people (free as in beer or free as in freedom?), “open” has similar confusion (open
Commercial Use should be paid (Score:2)
Re: (Score:3)
This can be accomplished with dual licensing. Here [spencerjones.blog] is a more in-depth explanation.
Re: (Score:2)
This is completely within the control of the open source authors. They can choose fully open source, or they can choose to make it free for noncommercial use, or whatever they want. Go ahead, pursue your licensing objectives, it's your call!
Is there suddenly a requirement (Score:3)
that AI models be open source?
Facebook is, after all, a money-making enterprise. If they develop something, it's their call whether to make it open source, or not.
Re: (Score:3)
No, but Facebook advertised "open source" as the unique value their product had to offer. It turns out, to no one's surprise, that they were lying.
Llama 2? (Score:2)
Did they think that through before naming it? Because the slur, "yeah, their LLM is lame-ass, too" practically writes itself.
Llama? (Score:3)
They had to wait until Winamp* wasn't popular any more to release it.
* It really whips the llama's ass!