5 November 2024, Tue |
8:38 PM
Join Andrew Jackson, CEO of Inception, in an exclusive interview with Rabab Zehra, Executive Editor of TECHx, as they unveil ‘Jais’—a groundbreaking Arabic Large Language Model. Discover the collaborative journey behind Jais, its transformative impact on AI, and its impressive performance. Explore the multitude of applications it offers and its promising future as a dynamic open-source resource.
TECHx: The name “Jais” is inspired by UAE’s highest peak. Can you explain the significance of this name choice and its connection to the model’s purpose and impact?
Jackson: Naming our LLM ‘Jais’ takes inspiration from the Jabel Jais summit as it places the model as the pinnacle of the Arabic LLM space while paying tribute to UAE’s role as the frontrunner of AI innovation in the region. The name rings true with cultural relevance and local identity, connecting it to the region and its natural beauty. The name ‘Jais’ also has the acronym ‘AI’ in it, which works well as a reference to the transformative technology that powers the model.
TECHx: Jais was developed through a collaboration involving Inception, MBZUAI, and Cerebras Systems. Can you elaborate on the roles of these entities in bringing Jais to fruition and how they contributed to its development?
Jackson: Jais was made a reality through this collaboration with Inception, MBZUAI, and Cerebras Systems. Inception produced the Arabic datasets and conducted the training of the Jais model. MBZUAI focused on the training, evaluation, and testing of the model. Jais was trained on Condor Galaxy, the multi-exaFLOP AI supercomputer built by G42 and Cerebras.
TECHx: Jais is touted as a milestone in AI for the Arabic-speaking world. How do you envision this model promoting innovation and contributing to Abu Dhabi’s position as an AI hub for cultural preservation and collaboration?
Jackson: Through the launch of Jais, we see an opportunity to accelerate the pace of innovation in the region, setting a new standard for AI advancement in the Middle East and ensuring that the Arabic language, with its depth and heritage, finds its voice within the AI landscape. In addition, by open sourcing the model, we aim to accelerate the development of a vibrant Arabic language AI ecosystem, serving as a model for other languages that are underrepresented in mainstream AI.
TECHx: Jais is reported to outperform existing Arabic models significantly. Could you explain some of the factors that contribute to its superior performance and its competitiveness with English models?
Jackson: Jais is the largest and highest quality open-source bilingual Arabic-English model in the world. It learned the Arabic language while being trained from English knowledge. This makes it stand out, as adding English and code helps to dramatically improve its language capabilities, with the added ability of understanding cross-language and code-mixed content. There are various factors that contribute to Jais’ performance, primarily around how it’s trained. It is a 13-billion parameter, open-source model that was trained on a unique and purpose-built, high-quality dataset of 116 billion Arabic tokens which capture the complexity, nuance, and richness of Arabic. Despite being trained on significantly less English data compared to other models – 279 billion tokens -, this allows Jais to outperform all existing Arabic models by a sizable margin and even compete with English models of similar size.
TECHx: With Jais now available for use, what are some potential applications or industries that could benefit most from harnessing the capabilities of this Arabic Large Language Model?
Jackson: Jais has various viable use case scenarios. Most notably, several organizations including the UAE Ministry of Foreign Affairs, UAE Ministry of Industry and Advanced Technology, Department of Health, ADNOC, Etihad Airways, First Abu Dhabi Bank (FAB), and e& joined us as Launch Partners and are set to use Jais as a part of their workflows. Our vision for Jais is for it to be used across a wide range of applications, from customer service and support tasks across government services, patient triage and assistance, medical records translation, financial services, and more. Being an Arabic language model, it has great potential for helping the Arabic community by automating tasks and building natural language interfaces. From an industrial perspective, Jais will help enhance human capabilities, streamlining tasks, providing insights, and other efficient solutions. This, in turn, allows for better efficiency and integration into various industries to create new roles and opportunities for human oversight, creativity, and contextual understanding.
TECHx: Looking forward, what are the future plans for refining and expanding Jais as more users engage with the model? How do you envision Jais evolving over time as part of the open-source community?
Jackson: We will continue to enhance and expand Jais’ capabilities as our community of users, both individuals and enterprises, keeps growing and they share their feedback with us on the model’s performance. We foresee being able to support other media such as voice, imagery, and video into Jais, to offer businesses, developers, and researchers a more holistic user experience.”
