UAE G42 reveals open source Arabic language AI model.
A Silicon Valley-based chip manufacturer, engineers, and researchers worked together to create advanced Arabic language software for generative AI applications.
Arabic language AI model
The newly developed language model, named Jais, incorporates 13 billion parameters derived from a large dataset combining Arabic and English, including a segment of computer code. This endeavor, undertaken by a team including academics and engineers, was prompted in part by the scarcity of bilingual large language models.
Supercomputers from Silicon Valley's Cerebras Systems played a pivotal role in crafting this language model. Cerebras designs chips the size of dinner plates that rival Nvidia's powerful AI hardware. As Nvidia's chips face supply shortages, companies worldwide are seeking alternative solutions.
The Jais model
Dubbed after the highest peak in the United Arab Emirates, Jais is a collaborative venture involving Cerebras, Mohamed bin Zayed University of Artificial Intelligence, and Inception, a subsidiary of Abu Dhabi's tech conglomerate G42, focused on AI.
Given the limited availability of Arabic data for training a model the scale of Jais, the inclusion of computer code within the English data assisted in enhancing the model's reasoning capabilities. Timothy Baldwin, a professor at Mohamed bin Zayed University of Artificial Intelligence, explained that the code "provides a significant advantage in terms of reasoning abilities, as it outlines the logical steps."
Jais will be accessible through an open-source license. As The Jais model was trained by the group on Cerebras' supercomputer known as the Condor Galaxy. In the current year, Cerebras revealed the sale of three such units to G42, with the first anticipated to arrive within this year, and the remaining units scheduled for delivery in 2024.