UAE-developed LLM Jais 70B aims to democratise access to AI

Developer says Jais 70B delivers Arabic-English capabilities at an unprecedented size and scale

In addition to Jais 70B, Inception has also released an Arabic-centric LLM small enough to run on a laptop. Photo: PA

The latest release of a UAE-developed large language model, Jais 70B, will be able to deliver Arabic-English bilingual capabilities at an “unprecedented size and scale”, according to Inception, a company under the umbrella of G42, Abu Dhabi’s AI and cloud company.

Inception described Jais 70B as a 70 billion parameter model.

In the context of AI, a parameter is a variable that can help increase the accuracy of LLM predictions.

“It has increased the ability to handle complicated and nuanced tasks, as well as better capability to process complex data sets,” the company said.

In addition to Jais 70B, Inception announced that 20 different models ranging from 590 million parameters to 70 billion would be available.

This particular Jais release is based in part on the feedback from developers in the Arabic natural language processing community. As a result of that outreach, one of the Jais models available is the first Arabic-centric LLM small enough to run on a laptop, according to Inception.

The company also said that Jais 70B was trained using “the largest Arabic data set ever used to train an open-source foundational model”.

“Jais was created to preserve Arabic heritage, culture and language, and to democratise access to AI,” said Andrew Jackson, chief executive of Inception.

“Releasing Jais 70B and this new family model reinforces our commitment to delivering the highest quality AI foundation model for Arabic speaking nations.”

Inception, a G42 company, said its Jais 70B LLM is able to deliver Arabic-English capabilities at an unprecedented size and scale for the open-source community. Photo: Inception

Neha Sengupta, principal applied scientist at Inception reflected on the latest performance tests for Jais.

“For models up to 30 billion parameters, we successfully trained Jais from scratch consistently outperforming adapted models in the community,” she said, adding that Inception made the choice to build Jais 70B on Meta's Llama 2 open source LLM.

“Jais 70B retains and in specific cases, exceeds the high-quality English-language processing capabilities of Llama2, while vastly excelling on Arabic outputs versus the base model,” she added.

The name of the LLM, Jais, refers to the UAE's highest peak in Ras Al Khaimah.

Although Arabic is one of the most widespread languages in the world and is spoken by about 400 million people, during the initial growth of AI developments and large language models it was not at the forefront of AI offerings, with English proving to be most prevalent.

The complexity and diversified Arabic dialects, coupled with various language nuances, also posed a challenge for engineers and programmers trying to perfect machine learning technologies.

The tide has turned in the last year, however, with the introduction of various Arabic LLMs, including Jais, along with other advancements and efforts made to include Arabic.

Core42, developer of Jais Chat, says the AI chatbot is engineered with an Arabic-centric approach for efficient processing of Arabic text. Photo: Cody Combs / Ismaeel Naar

The UAE is not a stranger to Arabic AI and LLM efforts.

Back in May, Core42, a unit of G42, introduced a bilingual Arabic and English chatbot, Jais Chat.

The app, available to download free of charge on iPhone devices, is proficient in both Arabic and English. According to Core42, the app was engineered with an Arabic-centric model for efficient processing of Arabic text.

Much like OpenAI's ChatGPT or Microsoft's CoPilot, users can use a variety of prompts on Jais Chat to get answers, gather information and find solutions that might elude conventional search engines.

In 2023, a UAE developed bilingual LLM dedicated to climate research, Jais Climate, was unveiled as part of a partnership between Core42 and Mohamed Bin Zayed University of Artificial Intelligence.

That bilingual model, included 1.4 million climate-related instructions and was trained on the largest bilingual data set on climate and sustainability-related topics.

The UAE aims to become a global AI leader and continues to tap into new-age technologies in preparation for the economy and society of the future.

The Emirates has attracted interest for its potential to become a global AI hub, and has pivoted towards partnerships with technology majors from the West.

G42, meanwhile, has been at the forefront of the UAE's AI growth strategy.

In April, it received a $1.5 billion investment from Microsoft, which will help to further boost G42's global expansion plans and strengthen the UAE's position as a global technology hub.

In October, G42 unveiled a new entity, Core42, that merged three of its key units – G42 Cloud, its research and development arm Inception and ICT unit Injazat – to focus on delivering AI solutions and services on a national scale.

It also announced a partnership with OpenAI, focusing on using its models in verticals where G42 operates, including financial services, energy, health care and public services.