Stay up up to now with free updates
Simply log in Artificial intelligence myFT Digest – delivered straight to your inbox.
Meta has been criticized by the group, which has pioneered open source technology within the software world for 25 years, for calling its artificial intelligence models “open source.”
The social media company is “confusing” users and “polluting” the term “open source” by utilizing it to explain its Llama family of huge language models, said Stefano Maffulli, head of the Open Source Initiative. The organization coined the term “open source” within the late Nineteen Nineties and has been considered the guardian of the concept ever since.
In an interview with the Financial Times, Maffulli said this was “extremely damaging” at a time when organizations just like the European Commission were attempting to support true open source technologies which are outside the control of any given company.
Llama, which has been downloaded greater than 400 million times, in response to Meta, has change into the most well-liked amongst a wave of supposed open-source AI models difficult leading proprietary systems resembling OpenAI's GPT-4 and Google's Gemini.
However, most of those, including Llama, fall wanting full openness, stopping the type of experimentation and customization in AI that open source has delivered to the software world.
By creating confusion about which models are truly open, Meta risks hindering the long-term development of types of AI which are driven and controlled by users, reasonably than tightly controlled by a handful of technology corporations, in response to the OSI Boss.
Maffulli said Google and Microsoft have stopped using the term “open source” for models that will not be fully open, but discussions with Meta haven’t produced an analogous result.
Meta said it was “committed to open source AI” and that Llama “has been a cornerstone of AI innovation worldwide.”
It added: “Existing open source definitions for software don’t have in mind the complexity of today's rapidly evolving AI models.” We are committed to continuing to work with the industry on latest definitions to maintain everyone within the AI community protected and to serve responsibly.”
Many proponents of fully open source technologies still credit Meta's more limited Llama models with breaking the stranglehold of a handful of huge U.S. AI corporations and opening the generative AI market to more competition.
Meta's models were “a breath of fresh air” for developers, said Dario Gil, head of research at IBM, giving them an alternative choice to what he called the “black box” models of leading AI corporations.
However, transparency is proscribed. Meta offers developers the chance to download its Llama models without cost. However, the one technical details released are the weights or “biases” that affect how the model responds to specific prompts.
In addition, the license under which Llama was released doesn’t comply with the open source definitions recognized by the OSI, because it doesn’t allow Meta's foremost competitors to make use of the technology without cost.
Other technology corporations, resembling the French AI company Mistral, now consult with models like this as “open weight” as an alternative of open source.
“Open weight (models) are great. . . But it’s not enough to evolve,” said Ali Farhadi, head of the Allen Institute for AI, which has released a completely open-source AI model called Olmo.
Developers using models like Llama cannot see how they were developed or construct on them to create their very own latest products, as was the case with open source software, he added.
To meet the OSI definition of open source AI, set to be officially released next week, model developers have to be more transparent. In addition to the weights of their models, they must also disclose the training algorithms and other software used to develop them.
OSI also called on AI corporations to reveal the information on which their models were trained, but acknowledged that privacy and other legal considerations sometimes prevent this.
Maffulli said organizations resembling the European Commission have sought to offer open source special recognition of their regulations to encourage its use in widely used technology standards.
If corporations like Meta can turn this right into a “general term” that they will define to their very own profit, they are going to have the opportunity to “incorporate their revenue-generating patents into standards that the (Commission) and other bodies will look to adopt.” really openly,” he warned.
