Stay informed with free updates
Simply sign up for the Artificial Intelligence myFT Digest, delivered straight to your inbox.
Meta has been criticized for calling its artificial intelligence model “open source” by an organization that has led the way in open source technology in the software world for the past 25 years.
Stefano Maffri, head of the open source initiative, said the social media company’s use of the term open source to describe its Llama family of large-scale language models “confuses” users and calls it open source. “contaminate” the term. The organization coined the term open source in the late 1990s and has been considered the guardian of the concept ever since.
In an interview with the FT, he said this was “extremely harmful” at a time when institutions such as the European Commission were trying to support truly open source technologies that are outside the control of specific companies.
Mehta said Llama, which has been downloaded more than 400 million times, is among a wave of supposedly “open source” AI models that have sprung up to compete with major proprietary systems such as OpenAI’s GPT-4 and Google’s Gemini. It has become the most popular.
However, most of these, including Llama, are not fully open, preventing the kind of AI experimentation and adaptation that open source has brought to the software world.
According to the head of OSI, the meta is creating confusion about which models are truly open, and the long-term prospects for a form of AI that is manipulated and controlled by users rather than tightly controlled by a small number of tech companies. There is a danger that this will hinder further development.
Mahri said Google and Microsoft have stopped using the term open source for models that are not fully open, but discussions with Meta did not yield a similar result.
Meta said its company is “committed to open sourcing AI” and that Llama “is the foundation for global AI innovation.”
Additionally, “Existing open source definitions of software do not encompass the complexity of today’s rapidly evolving AI models. We are committed to continuing to work with industry on new definitions to provide
Many proponents of fully open source technology still credit Meta’s more restrictive Llama model with breaking the shackles of some of the largest US AI companies and opening up the generative AI market to more competition. There is.
Dario Gil, head of research at IBM, said Meta’s model is a “breath of fresh air” for developers, offering an alternative to what he calls the “black box” models of major AI companies. He said that
However, transparency has its limits. Meta allows developers to download Llama models for free, but the only technical details exposed are the weights, or “biases” that affect how the model responds to certain prompts. .
Additionally, the license under which Llama was released does not comply with the OSI-recognized definition of open source, as it does not allow free use of the technology by Meta’s biggest rivals.
Recommended
Other technology groups, such as French AI company Mistral, choose to call such models “open weight” rather than open source.
“Open weight[models]are great…but they’re not enough for development,” said Ali Farhadi, director of the Allen Institute for AI, which released a completely open source AI model called Olmo. I am.
Developers using models such as Llama can see how the model was developed and build on it to create their own new products, just as they would with open source software. It is not possible, he added.
Compliance with OSI’s open source AI definition, which is expected to be officially published next week, will require model developers to be more transparent. Along with the model weights, you must also disclose the training algorithms and other software used to develop the model.
OSI also called on AI companies to make publicly available the data on which their models were trained, although it acknowledged that privacy and other legal considerations may prevent this.
Mahri said institutions such as the European Commission are seeking to give special recognition to open source in their regulations to encourage its use in widely used technology standards.
If companies like Meta succeed in turning it into a “common term” that they can define for their own benefit, “the EC and other bodies will be able to promote true openness of revenue-generating patents.” “This will allow them to be incorporated into existing standards,” he warned.
