Could Falcon become the Linux of AI?
Could Falcon become the Linux of AI?
Abu Dhabi R&D institute announces open-source AI foundation
Abu Dhabi's Advanced Technology Research Council forms a new open-source foundation, initially to be driven by the Technology Innovation Institute (TII), to build a global ecosystem for the Falcon series of large language models - an apparent bid to create a global open-source LLM standard, in a similar way that the Linux Foundation has championed the Linux open-source operating system.
This week, at the World Governments Summit in Dubai, His Excellency Faisal Al Bannai, Secretary General of the Abu Dhabi government's Advanced Technology Research Council (ATRC) announced the Falcon Foundation, together with $300 million in funding to advance the development of open-source AI models. The ATRC's applied research arm, Technology Innovation Institute (TII), began talking about plans to cultivate a global open-source ecosystem for its large language models (LLMs) when it waived royalties for its Falcon AI model last summer. The formation of the foundation is intended to provide the means to do this.
The announcement of the Falcon Foundation follows the ATRC's November launch of AI71, which will develop and sell enterprise solutions on the TII's Falcon large language models. Whilst AI71 can play a key role in ensuring that robust, secure AI solutions are brought to market on Falcon models, the foundation will focus on promoting and supporting development on Falcon by global developers, plus providing the governance and continuity that enterprise customers need from a software platform. There's also the potential for the Falcon Foundation to support other open-source AI initiatives, where it deems they will help build confidence in open-source AI, in general.
TII's release of powerful Falcon LLMs under an open-source licence last year seemed to strike the right note with many AI developers and potential enterprise users.
The huge wave of hype accompanying the launch and growth of OpenAI's ChatGPT, put Generative AI and large language models high on the agenda of software developers, businesses and governments around the world. It has also proved to be a massive learning experience, with business and government leaders having to get to grips with how to create strategy and policy for generative artificial intelligence.
A top priority is data sovereignty - basically where an organisations' data resides and who controls it - therefore, where data is processed by AI models has received more scrutiny. Large commercial and government organisations want their data and the outputs from AI to remain within their control, whereas OpenAI's ChatGPT and Google's Gemini are black box systems, that require data to be processed on their servers. Hence, the need for Sovereign AI and for AI models to be within an organisation's control, they must be able to run them on systems of their own choosing. This is fueling the demand for open-source AI models that can be run within an organisation's network, and provide the ability for users to 'look under the hood'.
However, the competition to become the world's de facto open-source large language model is far from a one horse race. Facebook holding company Meta released its own highly successful open-source LLM - Llama - in February last year, while a new competitor, Mistral, was formed and released its own Llama-beating model within a few months, capping the year with a $2 billion company valuation.
In the past, establishing technology standards has often been more about community building, than about the technology itself. For example, the battles between Betamax and VHS video cassette standards, GSM and CDMA mobile communications protocols, or Windows and competing desktop operating systems, were all decided by the strength of the ecosystems that were built around them. Aside from great technology, future standards need go-to-market strategy and, it must be said, deep pockets.
Falcon Foundation differentiators
TII's Falcon LLM series, and the foundation established to back it, arguably have distinct advantages when it comes to building out a future open-source AI ecosystem.
Firstly, the funding for technology development will always be critically important. The Falcon series of large language models all performed well in comparison testing when launched last year, but top Hugging Face rankings come and go. To remain among the top performing AI models, there must be continuous development, requiring significant long term investment.
The UAE, on a federal-level, and Abu Dhabi, on an emirate-level, have identified AI as a top priority and Abu Dhabi is investing hundreds of millions of dollars in AI locally and globally. It is because the emirate sees investment in AI development as strategic to its future success, that funding has been allocated for the Falcon Foundation.
Secondly, the Abu Dhabi government, which owns both the ATRC and TII, and has provided the funding for the new Falcon Foundation, has identified the clear need for Sovereign AI itself, making it a big potential customer for open-source AI. The government's policy to accelerate AI adoption, whilst ensuring data sovereignty, make it the ideal customer for Falcon LLMs and the solutions built on them.
Government departments have among the highest demands for security, safety, reliability and scalability when considering technology. If the new foundation can ensure that Falcon meets local government needs, there's every reason to believe that it will tick the same boxes when being considered by other government and enterprise customers.
Thirdly, the Falcon Foundation is better placed to grow a government ecosystem than potential competitors driven by R&D or commercial stakeholders. The global government sector is an enormous opportunity for open-source AI software, but only if that software meets the often stringent requirements of government customers.
Being backed by a government AI leader such as the Abu Dhabi government, the foundation will have the opportunity to help fast-track the consideration process as Falcon models are introduced for government usage worldwide.
To emphasise this last point, Falcon AI memoranda of understanding were signed during the World Governments Summit with the government of Serbia. and with São Paulo state, which accounts for about 30% of Brazil's GDP.
The devil is in the detail
So, the stage is set for a powerful play to build a global open-source AI ecosystem that could be very valuable to some of the world's largest technology buyers, and create a new open-source AI standard in the process. However, there is much detail that must also be addressed to create a global open-source ecosystem.
To gain market traction, Falcon will need leading technology, a marketplace of solutions built on that technology, the backing of key technology players and widespread adoption of Falcon products by government and business customers.
Of course, open-source's free-to-use promise is a good start. Millions of developers worldwide have already downloaded the Falcon model and there are a growing number of commercial products that have been developed on it. The assurance of support and governance from the new Falcon Foundation, should give many others the confidence they need to develop on Falcon. In addition, we can expect the foundation to be interested in funding AI models and solutions developed on Falcon that can be made freely available via the Falcon ecosystem.
Meanwhile, whilst its government sponsors will certainly open doors for the foundation with governments around the world, the longevity of those relationships is going to depend on technology adoption. In order to have widespread adoption, Falcon must win support from key technology solution and service providers. Enterprise customers buy biggest from big enterprise solution sellers, so both buyer and seller need to be convinced of the benefits to building on Falcon's open-source platform.
The Falcon Foundation will also need to keep in step with global AI and data regulations, such as the new EU AI Act, in order to ensure there are no legal barriers to widespread adoption. So, there is a crucial legal element too, that applies not only to the compliance and future compliance of the core Falcon models themselves, but also to the developers and vendors creating solutions built on Falcon.
Realising the vision
The ATRC's vision for the Falcon Foundation is an ambitious one. Although building a global open-source alliance taking in academia, software development, industry and government stakeholders could take shape relatively quickly, given the drive and resources behind the foundation.
On the other hand, building a working ecosystem that creates real value, has more moving parts. It requires close alignment with developers and vendors, and the ability to convince enterprise users of the future of the new technology. How quickly backers can create a foundation that is able to execute all of that, is likely to weigh heavily on Falcon's chances of becoming the de facto open-source standard.
Read more about Abu Dhabi’s AI R&D:
ATRC to launch a series of new tech ventures (Middle East AI News)
Disrupt or be disrupted (Middle East AI News)
Making R&D a national priority (Middle East AI News)
Abu Dhabi's latest R&D milestone (Middle East AI News)