Core42 unveils advanced Arabic language AI model

Wam

• Jais30B programme launched
• Boon for 400m speakers
• Fine-tuning for specific sectors

A new version of a deep learning algorithm capable of translating or generating Arabic text was unveiled in the UAE this week.

Jais30B, which features 30 billion parameters, was introduced by Core42, a G42 company based in Abu Dhabi.

“With 160 percent longer answers in Arabic and a 233 percent increase in English, the platform outperforms its predecessor Jais 13B,” the company said in a statement.

• AI education takes centre stage in the UAE
• Careful what you wish for… AI and the future of agriculture
• G42 and OpenAI launch regional partnership

It offers a linguistic and culture-focused generative AI experience for over 400 million Arabic speakers worldwide.

“The initiative opens up possibilities for organisations, researchers and businesses operating in Arabic speaking regions,” Sid Bhatia, regional vice president and general manager META at US-based AI company Dataiku, told AGBI.

Large language models (LLMs) such as Jais30B can be fine-tuned for specific applications, providing organisations with “a compelling advantage across various functions such as sales, marketing, customer service and compliance,” Strategy& Middle East said in a report.

Jais resulted from the collaboration between Inception (now Core42), Mohamed Bin Zayed University of Artificial Intelligence and the US-based Cerebras Systems, trained on the Condor Galaxy 1 supercomputer. 

With the vast amount of data and linguistic nuances specific to Arabic, “leveraging a dedicated model can significantly enhance the accuracy and relevance of AI-driven insights”, Bhatia said.

The developers have also enhanced its process and policies to reduce biases and the production of hateful or harmful content by the model.

According to PwC Middle East, “the UAE is at the forefront of the GCC’s enthusiasm in generative AI and the development of LLM in the Emirates has gained global attention”.

The Gulf state previously developed an open-source LLM called Falcon using more than 300 Nvidia chips at the state-owned Technology Innovation Institute in Abu Dhabi, and has established itself as a leader in the industry.