![](https://api.time.com/wp-content/uploads/2024/09/TIME100-AI-2024-LEADERS_Faisal-Al-Bannai_RM-NEW.jpg)
In 2023, the Advanced Technology Research Council (ATRC), an arm of the Abu Dhabi government, spent millions of dollars building a series of large language models. ATRC secretary general Faisal Al Bannai then decided to release them online for free, reasoning that if they were as good as the team’s internal testing showed, it would bolster the UAE’s credibility and attract talent. It worked. The models, named Falcon after the UAE’s national bird, were by some metrics the best open source models at the time of their release, beating offerings by Meta and Google. “I mean, two years ago, how many people would refer to UAE on the map of AI? Not many,” Al Bannai told TIME in February. Falcon changed that.
[time-brightcove not-tgx=”true”]Al Bannai says the focus is now on multimodality and getting better performance with less. To that end, the ATRC released two smaller models earlier this year: one with vision capabilities, and another based on a novel architecture. Al Bannai says he’s working on a larger, multimodal model that will compete with the likes of OpenAI’s GPT-4o.
Read More: The UAE Is on a Mission to Become an AI Power
It’s important that countries have a compelling open source alternative to the “proprietary AI from the large players,” he says. At home, a version of Falcon is being rolled out in the UAE’s healthcare system. Meanwhile, several states, including Serbia, Uzbekistan, and the Brazilian state of São Paulo, have bought into Al Bannai’s open source vision, inking deals with the UAE to use its models.
Al Bannai got started in the technology industry importing smartphones to the Middle East in the early 2000s. At that time, he says, it was “too early for us as a country to capture” the internet boom. But having spent the past two decades digitizing its government, Al Bannai believes the UAE is well-positioned to leverage AI. Part of that advantage, says Al Bannai, is the country’s willingness to use private data, such as records held by hospitals and state-backed industries, to train future proprietary models for government or enterprise clients. “We can all scrape data from public sources. The real jewel is in the private data.”