UAE’s G42 develops AI Large Language Model in Arabic

The project is the culmination of a convergence of expertise, as a dedicated team of researchers and engineers from Inception, Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) and Silicon Valley chipmaker Cerebras Systems collaborated to bring Jais to fruition. The resulting large linguistic model boasts a configuration of 13 billion parameters, derived from a dataset of Arabic and English text elements. “(Code) gives the model a big leg up in terms of reasoning abilities, because it spells out the (logical) steps,” explained Mohamed bin Zayed University of Artificial Intelligence professor Timothy Baldwin to Reuters.

Jais, named after the highest mountain in the United Arab Emirates, underwent deep learning on a customized dataset comprising 116 billion Arabic tokens. This data aggregate was designed to encompass the complexities and subtleties inherent in the Arabic language. It also includes a vast repository of 279 billion English words, designed to enhance the model’s performance in all languages.

 

No limits to UAE’s tech ambitions

The Jais model has been refined through training on Cerebras’ supercomputer, the Condor Galaxy. But the UAE’s AI ambitions are far from fulfilled. The G42’s efforts in AI innovation are highlighted by the acquisition of three such supercomputer units from Cerebras, with the first unit to be deployed this year and subsequent units to be delivered in 2024. In addition, in August of the previous year, G42 established a $10 billion fund, in partnership with the Abu Dhabi Growth Fund (ADG), which focuses explicitly on technology investments in emerging markets. In addition, G42 has aligned itself with various prominent Abu Dhabi entities, including the Abu Dhabi National Oil Company, the Mubadala investment fund and Etihad Airways, to capitalize on the potential of this technology.