A group of engineers and researchers, along with a chip company based in Silicon Valley, has collaborated to launch an advanced program in Arabic that can run artificial intelligence applications. The new large language model, named "Jais," consists of 13 billion parameters created from a vast dataset combining both Arabic and English, part of which comes from computer code.
One of the motivations for the team, which included academics and engineers, was their observation that there are few large bilingual language models. The new language model was developed with the aid of supercomputers produced by "Cerebras Systems" in Silicon Valley.
"Jais" derives its name from the highest mountain in the United Arab Emirates and is the result of collaboration between Cerebras, the "Mohammed bin Zayed" University for Artificial Intelligence, and the "Inspection" company, part of the Abu Dhabi-based "G42" technology holding group, which focuses on artificial intelligence.
Timothy Baldwin, an AI professor at "Mohammed bin Zayed" University, explained that due to the lack of sufficient Arabic data to train a model the size of Jais, the inclusion of computer code within the English language data helped enhance the model's reasoning ability. He stated to "Reuters": "(The code) provides the model with a significant boost in terms of reasoning because it illustrates the logical steps."
"Jais" will be available through an open-source license.