Robin Li: The large volume model is meaningless, and the opportunity for volume application is greater

On September 5, the 2023 Baidu Cloud Intelligence Conference was officially held, less than a week after Baidu officially announced the approval of the filing and the full opening of the Wen Xin one-word model on August 31. At the Cloud Intelligence Conference, Baidu announced a number of large-model product upgrades.

It is worth mentioning that although Baidu founder, chairman and CEO Robin Li did not come to the conference site "in person", he gave a speech in the form of a digital doppelganger.

Li Yanhong said that the model itself does not directly produce value, and the application developed based on the basic large model is the meaning of the model's existence. The remarks reflect Baidu's ambition in the application of large models, which is also an important direction for all large model companies to focus on at this stage.

A staff member of a large model enterprise interpreted the "city boundary": "Everyone has long realized that the main market of the big model in the future is the ToB market, and they are all looking forward to the approval of the record. Next, with the liberalization of policies, large manufacturers will catch up first, and small and medium-sized players who do not have the ability to do general large models in the future will definitely do products and services for the demand points of the market segment, and the enterprise-level market will be very lively. ”

At the meeting, Shen Zhen, executive vice president of Baidu Group and president of Baidu Intelligent Cloud Business Group, announced that Baidu Intelligent Cloud Qianfan model platform has been fully upgraded to 2.0.

"Qianfan" is an important starting point for Baidu Intelligent Cloud to enter the large-model enterprise-level market, combining tool chain, computing power, large model and dataset, and enterprise-level security tools.

Such products and services are regarded as the key to the realization of ToB applications in large models, so they have also become a "must-compete place" for global large model enterprises.

At present, many domestic enterprises have launched corresponding products, such as Tencent Cloud has officially announced the industry large model selection store based on the "TI platform", and SenseTime has released the "Daily New SenseNova" large model system. In foreign countries, there is news in the industry that OpenAI, which has developed ChatGPT, is also considering launching a large model store similar to the "App store" for customers to sell customized AI models to enterprises.

Baidu positions "Qianfan" as the world's first one-stop enterprise-level large-model platform. According to Shen Chu, there are nearly 10,000 monthly active enterprises connected to "Qianfan", covering more than 400 business scenarios in finance, manufacturing, energy, government affairs, transportation and other industries.

Shen said that in this round of upgrade, the number of mainstream large models at home and abroad managed by Qianfan platform has risen to 42, and 41 high-quality and industry-specific datasets have been preset, and 10 selected application paradigms such as knowledge Q&A, customer service dialogue, and code assistant have been preinstalled, aiming to reduce the threshold for enterprises to use, train and reason about large models.

Qianfan's tool chain, computing power service, and security service have also been upgraded accordingly, for example, after the upgrade, it only takes one day to run through a model and see the effect; When training a large model in a Vanka cluster, the effective training time reaches more than 95%.

At the meeting, Baidu Intelligent Cloud also released the digital government solution "Kyushu" based on the Wenxin model, the financial solution "Kaiyuan", and updated the industrial solution "Kaiwu".

In addition, Baidu Intelligent Cloud also released the "AI Native Application Family", which can provide 11 products suitable for general scenarios of enterprises for the common application scenarios and needs of a large number of customers, focusing on service marketing.

(Author|Dong Wenshu Editor|Sun Chunfang)