https://arab.news/vjva8
- Advertising agency TBWA\RAAD last year partnered with Core42, parent company of Arabic large-language model Jais
- Developing Jais’ potential in media and marketing industries will not replace humans, says advertising executive
DUBAI: Advertising agency TBWA\RAAD partnered with Core42 last year to harness the potential of Arabic large-language model Jais in the creative sector, a move which could also see greater online content created in the language spoken by millions worldwide.
Named after the UAE’s highest peak, Jais was developed by Core42 in collaboration with the Mohamed bin Zayed University of Artificial Intelligence and Cerebras Systems.
Jais is bilingual, but “it is purpose-built for Arabic,” said Ihsan Anabtawi, executive vice-president and chief commercial officer at Core42.
Although there are over 400 million Arabic speakers around the world, only 1 percent of online content is in Arabic, so “there’s definitely a need to serve the Arabic language but also do it in a bilingual way in order to expand the reach,” Anabtawi told Arab News.
Jais’ capabilities are broad with potential for multiple applications across industries, but it currently specializes in content generation, summarization and translation, he explained.
The first version of the model was trained on 13 billion parameters with the most recent version, introduced in November, being trained on 30 billion.
There was an “element of cultural preservation and amplification” with regard to Jais’ Arabic capabilities, but the company also wanted to broaden its reach to a diverse audience that accurately reflects the demographics of Jais’ birthplace, the UAE, said Anabtawi.
Arabic has multiple nuances with several dialects across the Middle East and North Africa region. Purposefully building Jais for Arabic allowed the company to have “a differentiated approach in terms of quality and content” while its bilingual nature enables the model to have varied applications for both people and businesses.
Anabtawi believes Jais is equally fluent in both English and Arabic. The latest model showed a 40 to 60 percent increase in Arabic and a 233 percent increase in English when it comes to longer and more detailed answers. In addition, there was a 53 percent improvement in summarization in Arabic and 85 percent in English.
For TBWA, “innovation has been a big part of our narrative for the last several years, and we had been looking at generative AI as part of our creative toolset for quite some time,” said Noah Khan, regional president of Digital & Innovation, CEE, Middle East and Africa at TBWA.
He told Arab News that the partnership with Core42 was “incredibly exciting” for two reasons.
First, the agency was proud of its roots and believes in partnerships, particularly with homegrown companies, and second, it gives TBWA an opportunity to celebrate the Arabic language, which has been a priority for the agency and its chairman Ramzi Raad since its establishment.
The agency has worked with other generative AI platforms in the past but the work “is currently confined to internal use only due to legal and copyright concerns,” Khan said.
He describes the relationship between the two companies as a “two-way partnership.”
Jais will allow the agency to “superpower our creative capabilities, especially with Arabic” as well as enable it to “use these models to train the next generation of Arabic specialists,” Khan explained.
The latter means that the agency will play a role in training and developing Jais’ capabilities to shape it into a more useful tool for the media and creative industries.
He said: “The focus is to start growing this as part of our capabilities as we evolve and transform ourselves; we see this as a means to augmenting ourselves to be bigger, better, and faster.”
Khan and Anabtawi both liken Jais and other generative AI model technologies to the mobile phone or computer where humans learned to work with technology.
While there has been exponential growth and interest in large-language models, spurred on by ChatGPT, companies are still exploring how they can implement these technologies in a responsible way, Anabtawi said.
Individuals can use these models in their current form in multiple ways, but when “applying it to the business context, there’s a lot more that can be done,” he added.
The challenge is multifaceted as companies leverage AI to solve problems across industries, but it also raises the question of the future of the workforce.
Anabtawi said: “How do you prepare the next generation of workers? What jobs does the market need in the future and how do we reskill and upskill?”
Khan said: “There have always been concerns that the technology is going to replace the people, but what normally happens is the technology gives people the ability to do more.
“The speed at which you’re able to adapt and use that technology to the best of your abilities then gets you to stand out from everybody else.”
Anabtawi echoed the sentiment adding that it was not technology that replaced human talent, it was another human who knew how to use that technology.
At TBWA, for example, AI is not going to replace Arabic copywriters, said Khan.
He added: “There are tasks that we could offset to a generative machine that then frees up the time of individuals to focus on other areas that they normally (not) be able to do … so we see this as a way of freeing creativity.”