Recently, Meta has faced fierce criticism for labeling its artificial intelligence model as "open source." This accusation comes from Stefano Maffulli, the head of the Open Source Initiative, who believes that Meta is confusing users and tarnishing the term "open source."
Maffulli points out that Meta's Llama large language model does not meet the true standards of open source, which is particularly harmful given the current efforts by institutions like the EU to support genuine open source technology.
Meta's Llama model has been downloaded over 400 million times, making it the most popular among the so-called "open source" AI models. Despite this, Meta's model is not fully open, limiting users' ability to experiment and adaptively develop.
Maffulli emphasizes that if companies like Meta use the term "open source" ambiguously, it could hinder the long-term development of user-driven AI.
Although Meta insists on its commitment to open source AI and claims that Llama is the cornerstone of global AI, its openness is still limited. Meta only allows developers to download the Llama model for free, but in reality, it only provides the model's weight information without disclosing detailed technical data on how these models are developed. Additionally, Llama's license does not meet the OSI's recognized open source standards because it restricts the free use of Meta's competitors.
Some other tech companies have also started to refer to such models as "open weight," not open source. Ali Farhadi, head of the Allen Institute for Artificial Intelligence, says that while open weight models are valuable, they still fall short in supporting developers for in-depth development. Developers cannot see the development process of these models and cannot create new products based on them, which starkly contrasts with the advantages of open source software.
Maff reveals that OSI plans to officially release the definition of open source AI next week, requiring model developers to provide more transparency, including the training algorithms and other development software information, in addition to the model's weights. OSI also calls on AI companies to release the data used to train the models, although this is sometimes impractical due to privacy and legal reasons.
Maffulli warns that if companies like Meta successfully turn "open source" into a "generic term," they can influence the industry's fair competition by including patents that benefit their revenue in the push for truly open technical standards by the EU and other institutions.
Key Points:
🌐 Meta faces criticism for calling the Llama model "open source," claiming it confuses user understanding.
📊 Despite its popularity, the Llama model does not meet true open source standards, restricting developer use.
📈 OSI will release a new open source AI definition, calling for more transparency and information disclosure.