It seems that Baidu, one of China’s biggest tech companies and one of the first Chinese companies to launch an AI bot, is not happy with the level of excitement people have with AI and finds it a bit premature.
At a recent industry event in Shenzhen, Baidu CEO Robin Li expressed his belief that significant players in the tech industry should redirect their attention away from developing foundational AI models and instead concentrate on the burgeoning business of creating artificial intelligence-based applications.
Li argued that the continuous redevelopment of large foundational models represents a significant waste of social resources.
“In the AI-native era, we need AI-native applications at a scale of millions,” Li emphasized. Foundational models, characterized by hundreds of millions or even billions of data points, are intricate and expensive, limiting accessibility to only the most sophisticated tech companies.
Li likened a large language model to a fundamental foundation, akin to an operating system, but stressed that developers must rely on a limited number of these large models to build various native applications.
While Baidu did not provide a comment on the matter, many prominent tech companies in China and the United States have already unveiled their foundational models, sparking a race to establish the operating system of the future. Last month, Baidu introduced its latest AI model, Ernie 4.0, positioning it as a competitor to OpenAI’s GPT-4.
Other tech giants, such as Alibaba and Meta, have also released their models. Meta requires a special license for companies with over 700 million monthly active users to use its Llama 2 model.
I argued against attempting to catch up with these dominant tech players, deeming it improbable and ultimately fruitless. Instead, he suggested that the development of AI should follow a path similar to the smartphone revolution. After smartphones became popular and affordable, various applications emerged, some of which became significant tech companies. Li cited examples like Uber and Douyin (the Chinese version of TikTok).
According to Li, there is yet to witness the emergence of the best AI-native applications in China or the United States. The Chinese market has seen a proliferation of competing AI foundation models, leading to what an executive from Tencent called the “war of a hundred models.” However, analysts express concerns about the similarity among these models and the difficulty new entrants may face in competing against entrenched, more prominent players.
Li’s perspective contrasts with OpenAI CEO Sam Altman, who speculated that ever-bigger models might be replaced by smaller ones tailored to specific use cases. The development and use of large AI models have raised questions about transparency in AI development. A study from Stanford University found that 10 of the largest AI developers in the US failed a “transparency test” evaluating the origin of their data and how it is used.
In response to such concerns, China’s government has taken steps to regulate the development and usage of large AI models. In July, China’s Cyberspace Administration released some of the world’s first government regulations governing AI use, potentially narrowing down the field of AI models in China.
