The Open Source Initiative is making it clear what the term ‘Open’ AI means. This is designed to set the stage for what can and cannot be included under the category.
So far, the recent clarification is not sitting down well with tech giants like Meta whose models aren’t obeying the rules. They feature elements that are not covered by the likes of usual licenses such as data used to train models. Now, for AI systems to be deemed open source, they need to provide a list of paperwork to get considered.
This includes greater transparency about which data is used for training so others can comprehend and re-build it. Secondly, the entire code used to design and run AI. Third, the settings and weights used for training that assist AI in producing the right results.
This definition is in direct challenge to Meta whose Llama reportedly does not fit the criteria list. Despite getting marketed by Meta as the world’s biggest open-source model, it’s filled with restrictions.
Yes, you can download it for public use but it does have limits on commercial use for certain apps with more than 700M users. It also does not provide any access to training data, making it ineligible for consideration, thanks to the OSI standards.
As per the company’s spokesperson, the company does not agree with such a definition outlined by the OSI. It fails to understand the need to add more complexities to definitions of open-source models and fails to stick to old definitions.
They hope to continue their working relationship with the OSI so models are widely available to the masses and offered without added costs, no matter what definition is in place. It’s actually interesting as the OSI definition for open-source systems was the same for 25 years.
Right now, AI is trying to reshape the whole landscape and tech giants face a difficult decision. They can either accept or reject. But other than Meta, others are accepting the definition with open arms. This includes making training data accessible to all.
Meta disagrees and defends claims to limit access to training data, citing safety concerns.
Image: DIW-Aigen
Read next: Forbes Unveils Top Creators List of 2024 With Bigger Earnings And Greater Global Influence
So far, the recent clarification is not sitting down well with tech giants like Meta whose models aren’t obeying the rules. They feature elements that are not covered by the likes of usual licenses such as data used to train models. Now, for AI systems to be deemed open source, they need to provide a list of paperwork to get considered.
This includes greater transparency about which data is used for training so others can comprehend and re-build it. Secondly, the entire code used to design and run AI. Third, the settings and weights used for training that assist AI in producing the right results.
This definition is in direct challenge to Meta whose Llama reportedly does not fit the criteria list. Despite getting marketed by Meta as the world’s biggest open-source model, it’s filled with restrictions.
Yes, you can download it for public use but it does have limits on commercial use for certain apps with more than 700M users. It also does not provide any access to training data, making it ineligible for consideration, thanks to the OSI standards.
As per the company’s spokesperson, the company does not agree with such a definition outlined by the OSI. It fails to understand the need to add more complexities to definitions of open-source models and fails to stick to old definitions.
They hope to continue their working relationship with the OSI so models are widely available to the masses and offered without added costs, no matter what definition is in place. It’s actually interesting as the OSI definition for open-source systems was the same for 25 years.
Right now, AI is trying to reshape the whole landscape and tech giants face a difficult decision. They can either accept or reject. But other than Meta, others are accepting the definition with open arms. This includes making training data accessible to all.
Meta disagrees and defends claims to limit access to training data, citing safety concerns.
Read next: Forbes Unveils Top Creators List of 2024 With Bigger Earnings And Greater Global Influence
