CAN FALCON SOAR? Abu Dhabi throws a surprise challenger into the AI race. Over Recent decades the oil-rich economies of the Gulf have shown a taste for flashy government projects with dubious payoffs.

In the early 2000s Dubai spent an estimated $12 billion building an artificial archipelago shaped like a palm.

Last year Qatar splurged around $22 billion hosting the football World Cup. Saudi Arabia, the region's gorilla, is building a pair of 120 km-long skyscrapers in the desert for roughly $1 trillion.

Amid the vanity projects, some serious efforts at economic diversification are also being pursued.

One such endeavour is under way in Abu Dhabi, where earlier this month a government research institute released Falcon 180B, its latest massive artificial-intelligence [AI] model, which is impressing technologists around the world with its performance.

ABU DHABI has even bigger AI ambitions. '' We are entering the game to disrupt the core players,'' says Faisal al-Bannai, secretary-general of the Advanced Technology Research Council [ ATRC], the government agency which houses the institute that created Falcon.

He says that later this year the ATRC will announce the launch of a state-backed A.I. company to go head-to-head with the field's leading lights, such as OpenAI, creator ChatGPT. Though it will face an uphill battle, the Emirati outfit could yet emerge as a credible competitor.

Its success will be closely watched both by rivals and by other governments seeking to carve out a role in an AI economy currently dominated by America and China.

The Technology Innovation Institute [TII], the applied-research arm of the ATRC, employs around 800 staff of 74 nationalities, working on subjects from biotechnology and robotics to quantum computing.

Launched in 2020, it has been experimenting with ChatGPT-like ''generative'' AI for some time.

It released Noor, an Arabic-based AI model, in April last year, and then Falcon 40B, the first iteration of its flagship open-source model, in May last year.

Falcon 180B, as its name hints, is a beefed-up version of its predecessor. Comparing the performance  of AI models is notoriously tricky, but going by a selection of commonly used benchmarks compiled by Hugging Face, a library of models, TII's latest release bests the previous open-source champion, Meta's LLaMA 2.

A blog post by Hugging Face still suggests the model is ''on par'' with Google's PALM 2.

Why make such a powerful model freely available? Mr. Bannai talks of ''democratising'' access to a transformational new technology, and warns against power falling into the hands of a small clique of companies, as has happened in the Internet economy.

But opening up access to Falcon 180B also allows software engineers to play around with a model that is not quite at the technological frontier, and suggest improvements.

According to TII, some 12 million developers experimented with the first generation of  Falcon.

The World Students Society thanks The Economist.


Post a Comment

Grace A Comment!