Meta insists that Llama, its artificial intelligence system, is open source. However, the definition published by the OSI (Open Source Initiative) does not align with what the company claims.
The OSI has published on its website the characteristics that an artificial intelligence model must have to be considered open source. Among them, the system must be usable for any type of purpose without needing to ask for permission, for example.
It must also be transparent and offer access to the data with which the model has been trained, in addition to making public the entire code used and the training weights. This definition by the OSI contrasts with how Meta promotes Llama, which has restrictions on its commercial use. We also cannot access the databases used for its creation.
Prior to this definition, the OSI already published a post in 2023 where it firmly stated that Llama 2 was not open source. There, they wrote that “Meta confuses ‘open source’ with ‘resources available to some users on certain occasions’.” In that same post, the OSI encouraged Meta to “correct their mistake.”
Now, a spokesperson for Meta, Faith Eischen, stated to The Verge that they maintain their disagreements with the OSI’s definition. According to Eischen, “there is no single definition of open-source AI, and defining it is a challenge because previous definitions do not come close to understanding the complexities of current AI models, which are rapidly advancing.”
Stefano Maffulli, executive director of the OSI, states that it took two years to adjust the definition and that it was done through a collaborative process with numerous experts in machine learning, philosophers, Creative Commons content creators, and much more.
