Meta and Source Initiative (OSI) do not match. The company led by Mark Zuckerberg claims that it is the ultimate in artificial intelligence (AI) models “open source””, but the organization has been in charge of defining the scope of this term since its creation in the late 1990s.
In February of the previous year; OSI published an article in which it should be noted that “LLaMa II is not open source“. The organization argued that the license terms of the Meta model have a series of restrictive features that simply make it non-compliant with the Open Source standard. At the instance of Meta, the OSI returned to battle.
Meta accuses OSI of “contaminating” the word Open Source
Stefano Maffulli, director of OSI, said in an interview with the Financial Times that Meta is “confusing” users and “Contaminate” the word Open Source continuing to use your models. This situation, according to Maffulli, can undermine what is truly an open source model, impeding the development of AI.
Call 3 license termsThe latest model of the Llama de Meta family, with the essence of “non-exclusive and limited”. If anyone can use this model to power their AI projects, users aren’t completely free to use it. In particular, there are two main limitations.
If you want to create a great and successful product, you can pay. Meta explains that if a product from Llama has more than 700 million monthly active users, the developer in question will have to request an additional license. Meta here can deny permission, or with additional requirements, for example, offer a solution.
You cannot use Llama III in missions. The use of the model is also subject to a Use of the policy is welcome which limits the use of the model. For example call III they cannot promote spam, develop firearms or operate critical infrastructure, transport technology or heavy machinery.
Meta continues to say that its models are open
OSI strives to ensure that the term Open Source is used according to a defined definition beyond Meta. Maffulli said companies like Google and Microsoft stopped using them for models that weren’t completely open, but their work with Meta didn’t bring results.
And there’s reason to believe that the Menlo Park-based multinational and nonprofit organization doesn’t agree. Those led by Mark Zuckerberg believe that the current definition of open source does not include the complexity of AI models, and do not hesitate to point out that the industry needs a new definition.
The Llama family of models, however, opens the door to a variety of capabilities that are closed to models such as OpenAI’s GPT-4. For example, you can access different versions of the model, with weights and evaluation codes, to set up your services and run them from your local infrastructure or from the cloud.
Images | Xataka with DALL·E 3 | Screenshot of the goal page
In Xataka | Amazon, Google and Microsoft are betting the future of AI on nuclear energy. It’s complicated
