TMTPost -- Alibaba Group CEO Eddie Wu beleives artificial intelligence ( AI ) has much more potential than people expected and a profound transformation of artificial general intelligence ( AGI ) , AI with capabilities that rival those of a human, is just in its bud burst.

Credit：Alibaba Cloud

"The pace of AI development over the past 22 months has surpassed any historical period, yet we remain in the early stages of the AGI transformation,"   Wu said in a keynote at Apsara Conference 2024, an annual tech event organized by Alibaba Cloud Intelligence Group ( Alibaba Cloud ) . "Historical experience tells us that we always overestimate the short term and underestimate the long term. Our current imaginations about AI may be underestimations," Wu pointed out

In Wu ’ s view, the most significant opportunity for generative AI lies far more than one or two new killer applications for a smartphone, instead, the technology will take over the digital world and transform the physical world. "Last year, large models could only assist programmers in writing simple code; today, they can directly comprehend requirements and complete complex programming tasks. While their mathematical prowess was at a high school level last year, it has now reached a level comparable to Fields medalists. They are nearing proficiency at levels equivalent to Ph.D. degrees in disciplines such as physics, chemistry, and biology,"   Wu said.

Aspara Conference, an event kicked off in Hangzhou Yunqi Town, China, Thursday, saw Alibaba Cloud ’ s release of over 100 of its newly-launched large language models, Qwen 2.5, to the global open-source community.The newly released open-source Qwen 2.5 models, ranging from 0.5 to 72 billion parameters in size, feature enhanced knowledge and stronger capabilities in math and coding and support over 29 languages, catering to a wide array of AI applications both at the edge or in the cloud across various sectors from automobile, gaming to science research.  

Wu noted Qwen ’ s open-source Qwen2 swiftly topped Hugging Face ’ s global open-source model rankings in June this year. On Hugging Face, there are nearly 50,000 native and derivative models of Qwen, ranking second globally in terms of model number. The ModelScope community of Alibaba Cloud hosts over 10,000 models and serves more than 6.9 million developers.

Wu said Alibaba Cloud will continue to further reduce user costs to promote AI adption as the cost of inference is a critical factor for application proliferation. In the past year, the price of Qwen API on Alibaba Cloud ’ s Model Studio has dropped by 97%. The lowest cost for deploying a million Tokens is now just RMB 0.50 ( US$00.007 ) . "The cost of model inference has decreased exponentially, outpacing the cadence of Moore ’ s Law,"   Wu said.

Moreover, the AI development is only the beginning. To achieve true AGI, the next generation of models must possess a larger, more universal, more generalized knowledge system, and more complex and multi-layered logical reasoning capabilities, Wu said. He predicted the threshold for building the world ’ s advanced models and staying competitive will reach billions or even tens of billions of U.S. dollars. "The path for AI to possess creative abilities and help humans solve complex problems is clear, unlocking the potential for AI ’ s broad application across various industries and scenarios,"   said Wu.

The evolution of AI computing is accelerating and becoming the dominant force in computing architecture   as Alibaba has seen that over 50% of new computing power demand stems from AI, and AI computing needs are at the forefront of demand. Despite Alibaba Cloud ’ s substantial investments in AI computing capacity over the past year, it still falls short of meeting our customers ’ voracious demand, according to Au.

Wu stressed Alibaba Cloud is investing, with unprecedented intensity, in the research and development of AI technology and the building of its global infrastructure. "Historically, people often overestimate the short-term impact of technological revolutions while underestimating its long-term implications,"   said Wu. "During the early stages of new technology adoption, with low penetration rates and no personal experiences to draw upon, skepticism is natural. This technology will grow despite your skepticism, and you risk missing out on significant trends if you hesitate."