1 Like
I have to assume (or at least consider) the worst of the Zuck, so I canât rule out the possibility that this is merely a strategy to compete with Google and OpenAI.
1 Like
Iâm pretty sure that Android being open source is the strategy to compete with iOS. But it doesnât matter if this is true or not.
One of the greatest strengths and advantages of open source is that others can contribute, etc.
2 Likes
If so I donât think that would be a bad thing (or âassuming the worstâ).
Personally, In the context of large corporations that I do not trust, I often feel more secure when those companies choose open source, or choose privacy for their own self-interested reasons because I perceive that as a more durable or at least more predictable commitment since they are pursuing their own self-interest (or at least not acting against their self-interest).
My understanding of why Meta has been willing to open source Llama (at least the smaller models), is because doing so doesnât actually harm or undermine their business model. Unlike OpenAI where the model IS the end product, Meta (supposedly) doesnât see their model (Llama) as the end product, for them it is apparently more of a building block that they want to integrate into their actual revenue generating products and services. So the incentives and constraints are fundamentally different than they are for OpenAI or Anthropic etc that are trying to monetize the LLM directly. Open Sourcing Llama has a lower âopportunity costâ for Meta, than it does for AI companies. At least that seems to be the logic Zuckerberg has used to explain the decision, and while I do have a very low opinion of both him and Meta, it is a plausible explanation.
My impression is mostly based on this interview
5 Likes
6 Likes
IIRC llama isnt actually open source, but rather source available⌠Has something changed?
Additionally, I would probably argue that there are no open source or source available AI models at all, because the all of the datasets are still proprietary. If you canât build the model from scratch, then you donât really have the source do you? 
6 Likes
Hmm? Thatâs actually not something new or now. Facebook has been very open about open-source in Machine Learning for years: FAIR (Facebook Artificial Intelligence Research), Torch/PyTorchâŚ
Actually I think itâs because of the influence from Yann LeCun, whoâs always been encouraging open source in ML, rather than Mark.
Even if meta or another company makes it open source, none of us can make the training to see the results. You need to have very powerful devices, people, time and money. 
You are technically correct (but I donât think âsource-availableâ accurately captures the spirit of the license either, since the license grants you the right to (1) use, (2) distribute (3) copy (4) modify (5) create derivative works, it is much more permissive than source available typically implies)
Iâm far from an expert when it comes to licensing, but my understanding of the Llama license is that it is functionally open source for the vast vast vast majority of people. Afaik the two limitations of the license that would technically make it not a pure FOSS license are:
- If you run a service with over 700 million monthly active users, the rights granted by the license donât apply to you (I think there are less than 10 organizations in the world with more than 700 million MAU)
- Llama 3 may not be used for the purpose of improving competitors models.
1 Like
I think itâs 700 million MAU combined across all of your organizationâs services, but yeah. Itâs still contrary to the definition anyways:
5. No Discrimination Against Persons or Groups
The license must not discriminate against any person or group of persons.
6. No Discrimination Against Fields of Endeavor
The license must not restrict anyone from making use of the program in a specific field of endeavor. For example, it may not restrict the program from being used in a business, or from being used for genetic research.
7. Distribution of License
The rights attached to the program must apply to all to whom the program is redistributed without the need for execution of an additional license by those parties.
I mean, the people the limits do apply to are also the most likely to innovate upon Llama. If Llama were under a real FOSS license like GPL instead, then if another MAMAA company improved upon it theyâd be allowed to do so, and then that company would have to release those modifications under GPL as well, benefitting everybody.
Of course that would never happen, but thatâs why the definition matters still.
2 Likes
Facebook has quite a history of supporting open source though, for example, btrfs and improvements eg fs-verity. Itâs also quite well known that they like fedora and often help out with proposals etc.
3 Likes
Iâd imagine itâs because they donât operate a public cloud, unlike Google, Microsoft, and Amazon; so they donât feel the need to hide their datacenter-y secrets. Nobodyâs going to become a Facebook competitor from their filesystem innovations lol
I suppose itâs a good thing, relatively, that Facebook just stays in their own lane? But theyâve been encroaching in other tech spaces lately so who knows.
Any examples?
I think Googleâs biggest motive for keeping Android open source was for manufacturers other than Apple to get into the smartphone space at the beginning. Those manufacturers could have made their own OSes, but having an open source OS available for them to port to their phone made life easier.
Google canât close off Android source code because at least some of their manufacturers have the resources to pour into their own OS (Like Samsung Tizen), and breaking up their marketshare on the smartphone OS space is a dumb idea.
They are encroaching into the Fediverse, thatâs the biggest thing I can think of.
I know Facebook supports a large amount of FOSS development, especially around the Linux space. But that doesnât deserve a pass in my eyes for all of the shit they have done that hurts privacy and that they continue to do.
They also joined with Google to monopolise web advertising.
A lot of shitty big tech companies support some major FOSS products and develop them. It really is a double-edged sword. Google develops Android and Chromium, Facebook develops React, Microsoft develops VSCode, Apple develops WebKit. But none of them should be pardoned for the shit they do just because they contribute to FOSS.
Theyâre also developing this:
And GitHub - facebook/zstd: Zstandard - Fast real-time compression algorithm which is the best alternative to the backdoored XZ
Wasnât the vulnerability in XZ patched almost immediately?