UAE Launches Arabic AI Tool Jais, Powered by Supercomputer

As the United Arab Emirates (UAE) works to lead the Gulf’s generative AI movement, an AI group with ties to Abu Dhabi’s rulers has released a top-tier Arabic AI tool.

The Jais model was developed in partnership with California-based Cerebras, the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), and G42, the UAE’s technology holding company. It is open-source and bilingual and was built for the more than 400 million Arabic speakers who live around the world.

In a global rush to secure supply for AI expansion, this release coincides with the UAE and Saudi Arabia making significant purchases of Nvidia processors essential for AI software.

A different open-source model called Falcon was earlier created by the UAE employing more than 300 Nvidia chips. One of the largest deals of its sort with a prospective Nvidia rival occurred this year when Cerebras agreed to a $100 million contract to equip G42 with nine supercomputers.

Despite Arabic being one of the most widely spoken languages in the world, Andrew Jackson of G42’s Inception remarked that the majority of large language models (LLMs) concentrate on English.Why shouldn’t the Arabic-speaking community have access to an LLM?
he questioned.

Jackson asserted that although existing sophisticated LLMs such as OpenAI’s ChatGPT, Google’s PaLM, and Meta’s LLaMA can comprehend and produce Arabic text, the Arabic component in these models is greatly diluted.

Jais outperforms Falcon and other open-source models like LLaMA in Arabic accuracy, according to its developers. According to Professor Timothy Baldwin, interim provost of MBZUAI, Jais is also intended to have a more accurate understanding of the region’s culture and context than most US-centric models do.

Baldwin added that measures had been taken to ensure Jais maintained cultural and religious sensitivity.

It underwent rigorous vetting to eliminate any damaging, delicate, offending, or inappropriate content that did not reflect the ideals of the groups involved in its production.

Jais, named after the highest peak in the UAE, was trained on a section of Cerebras’ Condor Galaxy 1 AI supercomputer for 21 days. As launch partners for the technology, G42 worked together with other Abu Dhabi organizations like the Abu Dhabi National Oil Company, Mubadala, and Etihad Airways.

Due to the dearth of high-quality Arabic language online data compared to English, training the model proved difficult. Jais solves this by utilizing both contemporary standard Arabic, which is understood throughout the Middle East, and the region’s various spoken dialects, which are drawn from media, social media, and code.

Baldwin came to the conclusion that Jais is clearly superior to existing models in Arabic and competitively equivalent or even slightly better in English across multiple tasks.

Leave a Comment