"Within the industry, there is a saying that the era of large models has arrived, and every product is worthy of being remade,” on May 26th, at the opening ceremony of the Zhongguancun Conference, Baidu CEO Li Yanhong (Robin Li) made a passionate remark.
“Baidu wants to be the first company to completely redo all of its products. It's not about integration or connection, but about starting from scratch and reconstructing!"
Li's statement reaffirmed Baidu's commitment to transforming from a search engine company into an AI company. With more than a decade of investment in AI, Baidu has swiftly captured public attention and attracted a significant number of industry experiential clients and developers following the official launch of its ERNIE Bot (a ChatGPT-like service) in March this year.
ERNIE Bot’s first introduction to the public came with hiccups such as the pre-recorded demo and a dip in stock prices. The company is making a recovery from the woes and says ERNIE Bot’s QPS (Queries Per Second) inference response speed has increased by 10 times, while the cost of inference reduced to 1/10th of the original version.
The company is showing an edge in the generative AI wave in China. It has just unveiled the WENXIN Qianfan, an enterprise-level large model production platform, with capabilities in over 400 enterprise scenarios showcased during internal testing.
"Why do we say that large models are changing artificial intelligence?" Li asked rhetorically, “because with increased computational power, large models, and vast amounts of data, intelligence emerges. Skills that were previously untaught, it can now learn."
This shift has also led to the development of artificial intelligence transitioning from discriminative to generative. "Search engines are a typical example of discriminative AI. However, AI can now be used for literary creation, writing reports, designing posters, and more, which are all generative tasks. Generative AI greatly enhances work efficiency. Some research institutions believe that in the next 10 years, the productivity of knowledge workers could increase by fourfold,” says Li.
Li believes in the next decade, half of all jobs worldwide will revolve around “prompt engineering”, which emphasizes the importance of question-asking, more so than problem-solving.
The company will soon unveil its latest foundation model, ERNIE 3.5, which follows a series of releases starting as early as 2019 with ERNIE 1.0, followed by iterations 2.0 in 2019, and 3.0 in 2021.
According to a previous IDC report, Baidu, SenseTime, and Microsoft’s spinoff Xiao Ice have been listed as “leading first-tier companies” in the industry, with Baidu AI Cloud ranking first in terms of “AI capabilities” and “user experiences”, and Xiao Ice ranking first in terms of “market and ecological capabilities” and “commercialization cases”.
The search engine giant's heightened investment in generative AI is poised to unlock greater opportunities for its enterprise-facing business and cloud business. In its latest financial report for Q1 2023, Baidu achieved a 10% year-on-year revenue growth, with the intelligent cloud business recording its first-ever quarterly profit.