The deployment of foundation models in various industries signifies a transformative shift, but it is not about introducing a single, powerful solution to replace existing workflows. 

Instead, this era of AI agents involves a nuanced and iterative process of aligning technology with engineering practices and market needs. On Thursday, founders of two Chinese AI unicorns, BaichuanAI and MiniMax, joined a forum with the president of Ant Group to discuss deploying foundation models for industries at the 2024 INCLUSION·Conference on the Bund in Shanghai.

After LLM

Baichuan AI has chosen healthcare as its focus for deploying AI models. Headquartered in Beijing, the company is valued at over RMB 20 billion and has received support from major players in China’s technology sector and investment funds from several local governments.

Founder Wang Xiaochuan noted that Baichuan AI is approaching the health sector via two main groups, pediatricians and general practitioners. The company aims to achieve low-cost and accessible healthcare by next year, and to alleviate hospital congestion.

“Creating super applications is not just a technical challenge; it also requires a collaborative development process with the industry,” Wang said. Public reports reveal that Baichuan AI has enlisted experts above the deputy chief physician level from 100 top hospitals to annotate data for its AI model, with the aim of enhancing the model’s accuracy and reliability.

In a separate keynote speech, renowned computer scientist Harry Shum echoed a similar sentiment: “The advent of the AI agent era will not involve a magical, all-powerful model suddenly replacing all workflows. Instead, it will be about the ongoing refinement and integration of technology, engineering, and market demands, ultimately delivering services that exceed human expectations.”

In a panel discussion on the same day, Peng Yang, Chief Executive Officer of Ant International and Senior Vice President of Ant Group, highlighted that over 70 countries have begun adopting open banking, with increasing reliance on AI to support this transition. In this context, he emphasized that privacy-preserving computing can break data silos, enabling businesses to collaborate without sharing raw data. This fosters the development of new business models and enhances customer service both domestically and internationally, making his view particularly relevant as AI becomes integral to open banking efforts.

Future trends

Since OpenAI released several Sora-generated AI video demos earlier this year, there has been a surge in investment in text-to-video technology. Last week, the Chinese AI startup MiniMax also launched a Sora-like tool video-1, becoming the latest tech company in the country to enter the sector.

While AI model applications are a prominent trend, a key issue becomes increasingly clear: the speed of innovation in applications depends on whether the underlying technology is solid and robust.

At the forum, MiniMax founder and CEO Yan Junjie emphasized his current focus on pushing the boundaries of AI through technological optimization. He stated that if the model’s error rate can be reduced to single digits, it would be considered reliable, allowing AI to evolve from an assistant into a productivity tool. 

“This shift could lead to genuinely mainstream products and drive fundamental changes in the industry,” Yan added.

On the same day, Ant Group’s Alipay launched an independent app called Zhixiaobao, an AI-powered assistant that connects to Alipay’s lifestyle service ecosystem and can quickly book tickets, order food, hail a cab, and find nearby dining and entertainment options.

Ant Group President and CFO Cyril Han summarized the goal of AI model-powered products as being “easy to use, useful, and affordable.”

Emphasizing the second of these three goals, Han added that if users mention that the AI lifestyle assistant is useful, “I think that would be even more meaningful.”