Abu Dhabi launches Jais: advanced AI language model for Arabic
Abu Dhabi has launched Jais, a new artificial intelligence (AI) language model specifically developed for the Arabic language. Jais, an open-source bilingual Arabic-English model, was created by Inception, a unit of Abu Dhabi-based AI company G42, in collaboration with the Mohammed bin Zayed University of Artificial Intelligence and Silicon Valley-based Cerebras Systems. The developers claim that Jais is more accurate than other existing large language models (LLMs) for Arabic. It is available for download on the machine learning platform Hugging Face.
Jais aims to bring the Arabic language into the mainstream of AI technology and encourage the scientific and computing communities to focus more on non-English LLMs. According to Andrew Jackson, CEO of Inception, Jais is expected to be highly useful in generative use cases such as generating responses to questions, creating documents, translations, emails, and providing advice and recommendations.
What sets Jais apart is its ability to capture the linguistic nuances of different Arabic dialects and understand language, context, and cultural references, making it more accurate and contextually relevant compared to other models. This advanced capability is crucial for applications in government use as well as industries such as finance, energy, climate, and healthcare.
Jais was trained using a vast amount of data on the Condor Galaxy, referred to as the “world’s largest AI supercomputer.” During the training process, it utilized 116 billion Arabic tokens and 279 billion English tokens. The model is continuously expanding as more Arabic content is collected to generate new instruction sets.
Several public and private organizations in the UAE, including the Ministry of Foreign Affairs, the Ministry of Industry and Advanced Technology, the Department of Health – Abu Dhabi, ADNOC, Etihad Airways, FAB, have partnered with Jais as launch partners.
The news text emphasizes the importance of increasing the online presence of Arabic language content, as currently, only about 1% of Arabic content is available online despite Arabic being one of the most widespread languages globally. Jais aims to address this gap by collecting more Arabic data from offline sources.
The availability of large language models like Jais plays a significant role in enhancing the artificial intelligence capabilities of companies, particularly in the field of generative AI. Speed performance is crucial to developers as it enables them to introduce new models quickly, iterate on different models, and bring them to the market or into production more efficiently.
In conclusion, the launch of Jais signifies a significant advancement in integrating Arabic into the AI mainstream. With its improved accuracy and advanced understanding of Arabic language nuances, Jais opens up possibilities for various industries and government applications. The availability of Jais as an open-source model will contribute to the development of AI capabilities and the increase in Arabic language content online.
