Jais Language Model, Collaboration Of Silicon Valley Engineers And Chip Companies For Arabic AI Development

JAKARTA - The Silicon Valley-based group of engineers, researchers and chip companies are working together to release advanced Arabic-speaking software that can drive a generating AI app.

The new big language model called Jais contains 13 billion parameters made of large amounts of data combining Arabic and English, some of which come from computer codes. The group, which involved academics and engineers, started the project in part because they said that there were still few large bilingual language models.

This new language model was created with the help of a supercomputer produced by Silicon Valley-based Cerebras Systems, which designs a plate-sized chip that competes with Nvidia's powerful AI hardware. The Nvidia chip is scarce, which has prompted companies around the world to look for alternatives.

Named at the highest peak in the United Arab Emirates, Jais is the result of a collaboration between Cerebras, Mohamed's Artificial Intelligence University bin Zayed, and a subsidiary of Abu Dhabi-based technology conglomerate G42, which focuses on AI.

Because there is not enough Arabic data to train a model as big as Jais, computer code in English data helps train model capabilities to think, according to Mohamed bin Zayed University of Artificial Intelligence professor Timothy Baldwin.

"(Kode) provides a model of great advantage in terms of thinking ability, therefore it explains (logical) measures," Baldwin told Reuters.Jais will be available through an open source license.

The group trains the Jais model on a Cerebras supercomputer called the Condor Galaxy. This year, Cerebras announced it had sold three such units to G42, with the first unit scheduled to arrive this year and remaining units to be delivered in 2024.