The announcement of the launch of the G30 B model was made by Core 42, a part of the G42 Group based in the United Arab Emirates, introducing the latest and most efficient Arabic language model, which is considered an open-source linguistic system.
This new version is based on 30 billion indices and follows the release of the previous model last August, which contained 13 billion indices. This step reflects Core 42’s commitment to providing an intelligent and culturally rich linguistic experience for over 400 million Arabic speakers worldwide, with a significant focus on the cultural aspect.
The G30 model is a partnership between Insben Company, the AI Center of the G42 Group, Mohammed Bin Zayed University for Artificial Intelligence, and the American Cyberpass Systems Company. The model was trained using the Condor Galaxy 1, one of the world’s fastest supercomputers based on artificial intelligence, capable of up to 4 exaflops (a million trillion calculations per second), with 54 million cores and 64 nodes. This model was developed in partnership between the G42 Group and Cyberpass Systems.
The new G30 B model underwent intensive training on a dataset that includes over 126 billion Arabic symbols, 251 billion English symbols, and 50 billion special symbols. The model provides answers that are 160% longer for the Arabic language and more detailed, with a 233% increase for the English language, indicating significant improvements in language production. The enhancements to the model “G30 B” were tested and verified using a variety of methods, comparing models and human evaluations. Tests showed that the updated version of the model outperformed the previous model “G13 B” in response time by 96% for Arabic and 97% for English.
The development team made further improvements to the model’s operations and policies to protect it from bias and generate harmful and hateful content, a function made easier thanks to the open-source release.
G30 B has demonstrated its flexibility and capabilities in the Arabic language, showing promising results through its diverse applications in various fields such as communications, energy, education, healthcare, and innovative marketing solutions.
Regarding this, Dr. Andrew Jackson, Vice President and CEO of Artificial Intelligence at Core 42, stated, “Launching G30 B is a new achievement for Core 42 and represents a significant advancement in the field of Arabic language speakers, especially in their ability to harness generative artificial intelligence capabilities.
We are excited to closely collaborate with our partners and clients to explore new applications, continue developing models, and work hard to innovate high-quality linguistic models in various languages.”
Mr. Andrew Feldman, CEO and Co-founder of Cyberpass Systems, stated that the Core 42 team and Cyberpass Systems have managed to launch a new and enhanced model in less than eight weeks after the release of G13 B for the global Arabic-speaking community, with the new model significantly outperforming its predecessor.
The G30 B model relies on the supercomputer Condor Galaxy 1 and achieves a new record in binary language performance.
