Abu Dhabi makes its Falcon 40B AI model open source

The government’s Advanced Technology study Council (ATRC) said on Thursday that the emirate of Abu Dhabi is making a large-scale artificial intelligence model called “Falcon 40B” available as open source for study and business use.

VentureOne, the business investment arm of ATRC, said it would also back good ideas that come from using the plan. Falcon 40B is a foundational large language model (LLM) that was made by the Technology Innovation Institute (TII), a study centre within ATRC. It has 40 billion parameters and was trained on one trillion tokens.

Applications like OpenAI’s bot ChatGPT are run by technologies called “generative AI models.” ATRC said, “TII is making the model’s weights available as part of a fuller open-source package.”

“While most LLMs have only given exclusive licences to non-commercial users, TII has made a big step forward by making the Falcon 40B LLM available to both researchers and commercial users.”

Abu Dhabi is the capital of the United Arab Emirates, which is made up of seven smaller countries called emirates.

In the past few years, Abu Dhabi’s government has done a lot to grow its technology industry. For example, it started the G42 AI and cloud computing company and the EDGE security technology group.

ATRC Secretary General Faisal Al Bannai told Reuters, “We wanted to give back to the community and speed up the use of AI.”

Also, Microsoft improves ChatGPT and Bing in a wide-ranging launch of AI products.

Bannai is also the head of EDGE. A TII director, Ebtesam Almazrouei, said that they want to support the use of generative AI not just in chatbots, but also in engineering, healthcare, suitability, and code.

In the past few months, companies all over the world have been racing to bring AI goods to market. At the same time, worries have grown about how the technology could lead to privacy violations, scams, and campaigns to spread false information.

When asked about privacy issues with the Falcon model, Bannai said, “If we let these platforms train on their own terms, we don’t have access to the data that goes into these platforms.”