Credit By: Pro Pakistani
An AI consortium with solid ties to Abu Dhabi’s leadership has unveiled a cutting-edge Arabic AI tool, signaling the United Arab Emirates’ ambition to lead the Gulf region’s generative AI movement.
The Jais Model: A Collaborative Breakthrough
The Jais model, born out of a partnership between the UAE’s technology holding company, G42, Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), and California-based Cerebras, stands as an open-source, bilingual innovation engineered to serve over 400 million Arabic speakers worldwide.
Elevating UAE’s AI Capabilities
Simultaneously, this release coincides with substantial investment by the UAE and Saudi Arabia in Nvidia chips, vital components for AI software, as the world races to secure resources for AI advancement.
From Falcon to Cerebras Collaboration
Previously, the UAE had introduced the open-source Falcon model, powered by over 300 Nvidia chips. This year, Cerebras made a monumental stride by signing a $100 million contract to provide G42 with nine supercomputers, marking a landmark deal and potentially challenging Nvidia’s supremacy.
Addressing Language Disparities
Andrew Jackson of G42’s Inception questioned the disparity in language focus, where most large language models (LLMs) predominantly cater to English, despite Arabic’s status as one of the world’s major languages. He questioned why the Arabic-speaking community should have a different dedicated LLM.
Jais’s Superiority in Arabic Accuracy
Developers of Jais assert its superiority in Arabic accuracy compared to Falcon and other open-source models like LLaMA. Furthermore, Jais is designed to possess a more profound understanding of the region’s culture and context, setting it apart from the primarily US-centric models, according to MBZUAI’s acting provost, Professor Timothy Baldwin.
Cultural Sensitivity and Rigorous Testing
Baldwin emphasized the measures taken to ensure Jais respects cultural and religious sensitivities. Extensive testing has been conducted to eliminate harmful, sensitive, offensive, or inappropriate content that contradicts the values of the organizations involved in its development.
The Inspirational Namesake: Jais
Named after the UAE’s highest peak, Jais underwent 21 days of intensive training on a segment of Cerebras’s Condor Galaxy 1 AI supercomputer, expertly managed by a dedicated team in Abu Dhabi. G42’s partnership extends to other influential Abu Dhabi entities, including Abu Dhabi National Oil Company, Mubadala, and Etihad Airways, as they embrace this transformative technology.
Navigating Data Challenges
The model’s training journey was challenging, mainly because of the online scarcity of high-quality Arabic language data compared to English. Jais effectively addresses this issue by incorporating modern standard Arabic, widely understood across the Middle East, and the rich tapestry of regional spoken dialects sourced from various media, social media, and coding platforms.
Jais’s Multifaceted Excellence
Baldwin concludes that Jais undeniably shines in Arabic and holds a competitive edge, if not a slight advantage, in English across a spectrum of tasks compared to existing models.