Alibaba Cloud's Zhou Jingren: Alibaba Cloud insists on embracing open source

When the AI giants begin to land applications and enter the battle for industry users, the debate over routes follows.

At the sub-forum of the World Artificial Intelligence Conference on July 5th, Alibaba Cloud CTO Zhou Jingren reiterated Alibaba Cloud's firm stance on embracing open source. "Two years ago, we released the Tongyi Big Model series at WAIC, announcing that the core model of Tongyi would be open source. Today, Tongyi Qianwen has truly achieved full-size, full-modal open source."

In Zhou Jingren's view, the training and iteration costs of large models are extremely high, and the vast majority of AI developers and small and medium-sized enterprises cannot afford to develop them on their own.

In August last year, Tongyi took the lead in joining the open source community, successively launching dozens of models, including language large models, multimodal large models, hybrid expert models, code large models, etc., and began to "dominate the charts."

The appeal of Alibaba Cloud's open source lies in the continuously strengthened capabilities of large models.

Alibaba Cloud stated that over the past year, the performance of the Tongyi model series basic models has continued to improve. In the evaluation of OpenCompass, Tongyi Qianwen Max scored on par with GPT-4 Turbo, becoming the first domestic large model to achieve such results. The newly launched open source model Qwen2-72B, ranks just below OpenAI's latest release, GPT-4o.

Zhou Jingren revealed, "Alibaba Cloud's actively open-sourced performance has reached the level of GPT-4, surpassing many closed-source model versions of large models, truly narrowing the gap between open source and closed source models, and accelerating the application and landing process of large models."

Clearly, Alibaba Cloud aims to compete head-to-head with closed-source players such as Baidu and OpenAI.

Not only open sourcing, but Alibaba Cloud also made a big move by upgrading its own Alibaba Cloud "Bailian" to 2.0 in May this year. Developers can directly call the APIs of Tongyi, Llama, ChatGLM series, and third-party models, like building with LEGO, they can build their own AI applications in just a few tens of minutes.

Developers also voted with their feet. Zhou Jingren revealed, "In the past 2 months, the download volume of Tongyi Qianwen open source models has doubled, exceeding 20 million times, and the number of Alibaba Cloud Bailian service customers has increased from 90,000 to 230,000, a growth of over 150%."

Alibaba Cloud, which is heavily focused on open source and AI ecology, aims to drive traffic to its own public cloud services. As Alibaba Cloud has always wanted to act as an infrastructure provider in the AI era, it is facing rapid attacks from its competitors.

Fortunately, with the penetration of AI, there is still untapped space in the public cloud market in China. Alibaba Cloud internally estimates that in the upcoming public cloud market, there is at least a scale of 10 billion RMB in incremental space each year. Enterprise consumption on the cloud will continue to increase with business development and deepening cloud usage, in a snowball effect Therefore, in order to minimize the threshold for model usage and accelerate the outbreak of AI applications, on May 21st, the Tongyi Qianwen series of models significantly reduced prices, with the flagship GPT-4 model dropping by 97%.

Zhou Jingren introduced that after the price reduction, a large number of customers directly accessed large models on Alibaba Cloud. In the past 2 months, the number of BaiLian service enterprise customers has increased from 90,000 to 230,000, a growth of over 150%.

As the second growth curve of Alibaba Group, Alibaba Cloud is using an open source and open strategy to exchange for new business growth.

Zhou Jingren emphasized that Alibaba Cloud will continue to embrace open source and openness, creating the "most open cloud in the AI era". Alibaba Cloud will work with industry partners to continuously help Chinese enterprises and developers seize the opportunities of the era of large models through open computing power platforms, self-developed open-source models, and high-quality model services